Hello Community! 👋

We are excited to announce an update to AI Agents: A redesigned AI Agent Builder experience! 🤖

With this release, we’ve rebuilt the agent builder to reduce friction, eliminate dead ends, and speed up your workflow, all while maintaining the full power and capabilities you rely on.

🚀 What’s New?

1. Side-by-Side Builder (Configure & Test)

No more hopping back and forth between separate tabs to build and verify your agents! The chat builder and form builder now sit side-by-side as complementary tools.

• Live Real-Time Sync: See your changes sync instantly across both panels as you work.
• Stage-Based Workflow: Easily switch between the Configure and Test stages at the top of your screen to tune and test without ever leaving your view.

2. Introducing "Scopes" (Formerly Active Locations)

We’ve renamed "Locations" to Scopes to better match how teams naturally think about where an agent should operate.

• Clear Boundaries: "Scope" explicitly defines the total reach or boundary of your agent, making it easy to specify where an agent can work, execute actions, or check for triggers.

3. Appoint Agents Directly During Setup

Choosing where your agent works is now an integrated part of the creation process.

• Context-First Building: Define the scope right at the beginning so you can build with full context in mind.
• Non-Blocking & Flexible: Appointment is completely optional during creation, so it will never block you from saving an agent.
• Centralized Management: View and remove existing appointments directly from one place during setup, or appoint agents straight from the agents table.

4. Sentence-Style Triggers & Actions

Setting up rules and logic now reads in plain, natural language!

• Natural Configuration: Rule setups read like normal sentences (e.g., "When status changes in the sub-items of the agent’s scope… post a comment in the work items in the agent's scope").
• Progressive Fields: Form fields appear dynamically as you need them, automatically incorporating your custom scope and appointment choices into the wording so the setup fits your exact use case.

5. Clearer Action Filters

Upfront, surfaced filters give you complete visibility into agent logic before it runs.

• Targeted Execution: Easily review active conditions at a glance so your agents run exclusively on the work that matters, saving action runs and avoiding unwanted triggers.

📖 Learn More & Share Your Thoughts

Plus, we’ve updated all existing agents in our AI Agents Library to leverage these new features, so you can explore updated examples or add them to your workspace!

To get step-by-step instructions and explore best practices for building your AI agents, check out our dedicated Help Center Article.

We’d love to hear how these builder updates streamline your workflows! Which new capability are you most excited to try out? Drop a reaction (👍 Like / 🙂 Agree / 💡 Insightful) and share your thoughts in the comments below! 👇

Before implementing them, teams often created tasks with inconsistent fields and workflows, making reporting and automation difficult. By introducing standardized request, project, and enhancement item types, we were able to improve data quality, streamline reporting, and simplify user onboarding.

The biggest benefit has been ensuring that every team follows a common structure while still maintaining flexibility for their specific business processes.

I'm curious to hear from other admins: Which Wrike feature has had the biggest impact on your organization's efficiency?

Hi Community! 

What a Collaborate 2026! 🎉

Our annual customer conference brought together exciting product announcements, real-world inspiration, and a look at what's next for Wrike. And, as always, one of the biggest highlights was our Product Keynote, where we shared our vision for the future of work and some powerful new capabilities to help teams get there.

This year, we introduced Wrike's vision as The Trusted Work Delivery Platform for People and AI, powered by the Work Intelligence® Graph.

The idea? Help organizations accelerate every stage of their transformation journey, from Strategy → Delivery → Learning, connecting big-picture goals with everyday execution and continuous improvement.

So, what does that look like in practice? Let's explore some of the biggest announcements! 👇

🌟 From strategy to execution: Introducing Wrike Strategic Portfolio Management

Big plans need more than great ideas. They need visibility, resources, and a clear connection to execution.

The new Wrike Strategic Portfolio Management (SPM), powered by Shibumi®, helps organizations connect strategic initiatives with the work happening across their teams.

With portfolio-level visibility, scenario modeling, capacity planning, and live execution data, leaders can make more informed decisions, evaluate priorities, and keep strategic plans aligned with reality.

🔗 Explore Wrike SPM 

🤖 Turn ideas into actionable work with AI Work Intake

What if you could go from a collection of ideas, notes, or documents to a structured project plan with the help of AI?

With AI Work Intake powered by Wrike MCP, you can use natural-language prompts and content from whiteboards to build structured work in Wrike, leveraging your existing blueprints and workflows.

The result? A Gantt chart populated with dependencies, milestones, and critical paths, helping teams move from planning to execution faster.

👉 Explore AI Work Intake via MCP

🔗 More possibilities with Wrike MCP

We're continuing to expand what AI assistants can do with Wrike!

The latest Wrike MCP server edition enables external AI assistants to interact with Wrike work, bringing information from multiple sources into Wrike Whiteboards and supporting both reading and updating work.

And there's more on the horizon: Intelligent Resource Allocation will help AI assistants identify the right people for the right work.

👉 Learn more about Intelligent Resource Allocation with Wrike MCP

👉 Find full MCP documentation here

🧠 Bring your own AI agents into Wrike as Workmates

Your AI agents can become part of your team in Wrike!

With External Workmates, you can connect your own AI agents to Wrike and bring their capabilities directly into your workflows.

And because trust and control are essential when working with AI, this experience is designed around a comprehensive AI Governance framework, including scoped permissions, access controls, auditability, and human validation.

During the keynote, we saw an example of a custom market-intelligence agent delivering a report through Wrike's proofing and approval process.

👉 Meet External Workmates in Wrike Labs

💡 Looking for inspiration? Visit our Workmates Library to explore more possibilities!

📊 Plan and manage complex work with the new Gantt Chart

The new Gantt Chart brings portfolio-level timelines with cross-project dependencies and milestones into one view, helping teams better understand how work connects across projects and programs.

This builds on a series of improvements we've introduced over the past few months, including Cascading and Required Fields, improved Request Forms, Workflow Governance, Project Baselines, and more.

Together, these capabilities make it easier to plan, coordinate, and manage work at scale.

👉 Discover what's new in Gantt

✨ More updates worth exploring

The Product Keynote wasn't the only place to discover what's new! Here are a couple of additional updates worth checking out:

💬 Threaded Conversations
You asked, we listened! 🙌

Comment Threads are now available in Wrike Labs, making it easier to keep conversations focused and organized, even when there's a lot happening in your task's comment stream.

With threaded replies, you can group related discussions together, reply directly to specific comments, and keep track of conversations without losing context. Plus, threads support rich formatting, attachments, and API and MCP integrations!

This update was inspired by feedback from our very own Wrike Community, so a big thank you to everyone who shared their ideas and helped shape this improvement! 💚

👉 Explore Comment Threads and try them in Wrike Labs

🎨 Make your Dashboards your own with color customization

Add more visual clarity to your reporting with Dashboard Color Customization, now generally available on Pinnacle and Apex plans.

👉 Explore Dashboard Color Customization

🔒 More control and confidence with AI Governance

As AI becomes a bigger part of everyday work, governance matters more than ever. We're continuing to strengthen Wrike's AI Governance capabilities, with new admin controls already available and additional enhancements planned for Q4.

👉 Explore AI Governance and Admin Features

💚 Now, we'd love to hear from you!

That's a lot of exciting innovation, and we're only getting started!

Which of these announcements are you most excited about? 🚀

Is there a new capability you can't wait to try, a feature that could make a real difference for your team, or something you'd love to learn more about?

Let us know in the comments! We'd love to hear your thoughts, questions, and ideas as we continue shaping the future of work together.

We're exploring AI integration (Claude via MCP) to automate ticket creation and template population for our creative workflows.
Questions:1. Does Wrike support MCP (Model Context Protocol) or have an official Claude integration?2. What are the API capabilities for programmatically creating/updating tickets with auto-filled templates?3. Any rate limits or constraints for API-driven automation?4. Examples or docs for third-party AI integrations?
Use case: Auto-generate campaign tickets with pre-filled custom fields to reduce manual entry.
Thanks!

Hey Community! 👋

Welcome to Wrike Agent News!

This monthly digest features updates, learnings, and how-tos from the Wrike R&D team.

This month is about the work agents create. Until now, when an agent created something, it produced a single, bare item - one task, folder, or project, built field by field, with a plain-text description. Useful, but a long way from the structured, ready-to-run work your teams actually operate on. September closes that gap from both ends: agents can now generate a whole standardized structure in one go, and write descriptions that read like a person laid them out.

On top of that, we’ve started reworking the whole agent setup experience. The first piece, appointing an agent right inside the settings form is already available, with a bigger builder redesign on the way.

Here are the primary updates landing this month:

🏗️ New Action: Agents Can Create an Item from a Blueprint

Agents could already create a single task, folder, or project, but only a bare one, field by field. Now they can spin up a whole blueprint, so a trigger produces the entire standardized structure, the parent plus all its subtasks, in one go. A recurring process kicks off and the agent launches the right template automatically: no more building the structure by hand or manually triggering the blueprint at the right moment. You pick the blueprint when you set up the action, not the LLM at runtime.

🎨 New Action Upgrade: Agents Can Create Items with a Formatted Description

With the first iteration of create-item actions, the item’s description came out as plain text. Now, the agent can write a properly formatted task: headings, tables, bullet and checkbox lists, bold, colors, so the created task reads like something a person laid out, not a wall of text. It’s the same formatting support as the change-description action, and the generated description shows up formatted in the logs and the playground too.

Why this matters: Combined with blueprint creation above, an agent can now produce a complete, well-formed item the instant a trigger fires, with no human cleanup step required.

🎯 Setup Gets Simpler: Appoint an Agent Right Inside the Settings Form

People built the trigger, wrote the actions, saved it… and never completed the separate step that tells the agent where to actually run. So the agent existed but never did anything.

We’ve folded that crucial appointment step into the agent form. There’s now an Active locations section right in the settings, where anyone who has AI agents in the space can add or remove the tasks, folders, and projects the agent should run on, and it all applies when you save. Open an existing agent and you’ll see every location it’s currently appointed to (including ones added from elsewhere), ready to edit in the same place. Creating an agent with no location is still allowed, so this never blocks you; it just makes the finish line a lot harder to miss.

Why this matters: It removes the most common reason a carefully built agent does nothing. Build it and point it at its work in one flow.

🌪️ Filters Now Sit in Plain Sight on Every Action Card

Admins kept discovering runaway action consumption only after the fact: an agent attached to thousands of tasks by mistake, or a test run burning through the monthly quota over a weekend. Per-action filters already existed to prevent this, but they were tucked behind a toolbar toggle and easy to miss. Now the Filters section sits at the bottom of each expanded action card, always visible, with a line explaining what it's for: filter out items you don't want the agent to act on (like completed tasks) to narrow its scope and reduce your monthly AI usage. Same filters as before, just impossible to overlook.

🔮 On Our Radar: A Simpler, Side-by-Side Builder (Mid-October)

We’ve heard a lot of feedback that setting up an agent is more cumbersome than it should be: too many steps, too abstract, and easy to lose the thread of what you’re actually building. We hear you, and we’re reworking the whole setup experience to bring something much better 💯

Here’s what’s coming 👇

• Appoint during setup ✅ (Already live): Appoint where the agent will work directly in the setup flow, instead of a separate appointment step afterward. It’s optional, so it never blocks creating an agent, and any locations already appointed (from any path) show up and can be removed. This is the first piece of the redesign and already shipped this month; appointment from the agents table is still on the way.
• “Locations” ➔ “Scopes”: We’re renaming Locations to Scopes so the wording matches how people already think about where an agent applies.
• Side-by-side builder: Chat and form settings sit next to each other instead of on separate tabs: complementary tools, not separate modes. The switch above them is now stage-based (Configure ➔ Test) and the two stay in sync live, so you can tune and test without leaving the screen.
• Sentence-style triggers and actions: Triggers and actions read more like plain language (e.g., “When status changes in the subitems of the agent’s scope… post a comment in the work items in the agent’s scope”). Fields show up progressively, and your own scope and appointment choices feed into the wording, so setup feels specific to your case rather than a generic form.

This is easier to show than to describe, so here’s a quick look at how the experience will change:

👉 Watch the Preview Video

Timing: We’re aiming to release this in mid-October. We’ll follow up with details as it lands.

💬 Webinar Alert

Want to see a risk agent built live and hear about our latest features? Join our community webinar on September 29.

👉 Click here to Enroll.

Questions or Feedback?

Need a different agent type, found a bug, or have feedback? Use the in-product feedback link or let us know below. Need help with AI features? Join an Ask an Expert session.

Found this update helpful? Let us know with a quick reaction (Like 👍, Agree 🙂, Insightful 💡) or drop your thoughts in the comments below!

Accomplishing this took some very simple instructions for the agent.

General Instructions:
You identify vendor names mentioned in task descriptions and map them to the closest matching existing vendor value.

Change Custom Field Action Instructions:
Read the triggered task description, identify the vendor name mentioned there in the Vendor Name section, and set the XXX field options to that value. If no reliable match is found, leave the field unchanged.

I changed the name of the custom field, because it can be whatever makes sense for your space/context.

What other contextual jobs have you given to Wrike AI Agents?

Want to see how Wrike AI Agents can help you stay on top of project health and identify what needs your attention?

On September 29, 2026, we’re hosting a live Wrike Essentials session focused on Wrike AI Agents, where we’ll walk through a practical Project Health Monitoring use case and show how AI can help teams stay ahead of potential issues.
👉️ Save your spot

Here’s what we’ll dive into:

• Explore AI Agents: See how AI Agents can support teams and help simplify everyday work.
• Monitor Project Health: Get a clearer view of project progress and identify areas that may need attention.
• Spot Potential Risks: Surface signals such as overdue work, blocked tasks, or upcoming deadlines that could impact project progress.
• Turn Insights Into Action: See how AI Agents can help identify what needs follow-up and guide next steps.
• Build a Practical Use Case: Follow along as we create a Project Health Monitoring AI Agent and see how it can be applied to real project workflows.
• Connect AI to Your Work: Learn how you can think about AI Agent use cases within your own Wrike workspace.
• Live Q&A: Bring your questions and get answers from our experts.
  

📅 Event details:

Date: September 29, 2026

Time: 10.30AM PST

Where: Register here 🚩

We really hope to see you there! 🙂

Here are some functions that can already be done by AI Agents, but would be better done by automation rules, if implemented:

• Change the status of a task if someone assignes it to himself
• Change the status of a task if someone assignes it to another user
• Remove a specific assignee from a task only if there are also other asignees
• Filter for tasks with a specific duration (e. g. reminder for tasks that have a duration of more than one day)

I deployed two AI agents to send weekly risk reports to our Teams channels and they worked for a while, but recently they've been inconsistent. Each time the log notes "Response status code: 422, message: Unprocessable Entity". Then randomly it will work the next week. Then break again. No changes have been made.

Any thoughts or guidance?

Hi everyone! 👋

I'm Marco, Product Manager for AI Agents at Wrike.

We are looking for a few users who have built an agent on Anthropic's Managed Agents (the hosted agents in the Claude API, not standard chat assistants) to help us test how they perform when connected to Wrike.

🎯 What does participation look like? You bring an agent you have already built, we connect it to Wrike, and you point it at real work. I want to watch the complete end-to-end path:

• Connecting the provider
• Bringing your agent in
• Granting access to the right places
• Watching it perform actual tasks
  

The goal is to hear every point where the workflow slows down or doesn't behave the way you expected.

💡 Built on a different platform? If you have built something similar on another platform, please let us know in the comments anyway. We would still love to get in touch and explore your setup.

📬 How to join: Simply leave a comment on this post, and we will reach out to you!

I tried to set up an Agent with quite simple tasks, e. g. removing our Dummy User as assignee where other users have been assigned additionally.

The result was horrible - the agent assigned about 20 random users to several tasks and even told me it hadn't done any change!

The whole AI integrations seems to be in an experimental stage. So I strongly recommend to use it also in experimental enviroment only!

I would highly appreciate if wrike would focus on more felxible automation capabillities instead.

Feature Updates

• Table & Gantt Views: Track Downstream Dependencies with the Successors Column 📊

You can now add a Successors column in Table view and the Gantt chart table. Acting as the mirror image of the Predecessors column, it instantly shows which downstream tasks depend on the current task.

What’s new?

- Add the Successors column directly from the Fields picker in Table view or the Gantt table.
- Easily view dependent tasks by row number directly within the table grid.
- Edit dependencies directly inside the cell using row numbers and dependency types (e.g., FS, SS, FF, SF) along with lag duration (e.g., 9SS+2d).
- Automatic bi-directional syncing: adding a successor automatically updates the Predecessors column on the target task, and vice versa.

How to set it up

1. Open a project in Table view or Gantt view.
2. Open the Fields picker menu.
3. Search for and enable the Successors column.
4. Type a row number into any Successor cell to establish a dependency.

Note: Scheduling behavior remains forward-driven, dates flow downstream from predecessors to successors. Adding or editing a successor updates visibility and dependency lines without triggering backward scheduling.

• AI Agents: Create Items from Blueprints and Add Formatted Descriptions 🤖

We’ve upgraded the "create-item" action for AI Agents, making automated workflow creation and task formatting more robust.

What’s new?

- Create items from blueprints: AI Agents can now launch complete standardized structures, including parent items and subtasks, in a single step. Instead of building bare tasks field by field, automated workflows spin up full project templates automatically when triggered. You simply pre-select the blueprint when configuring the action setup.

- Rich formatted descriptions: Items created by AI Agents no longer default to plain text. Agents can now generate styled descriptions complete with headings, bullet points, checkbox lists, bold text, and colored highlights. This rich formatting is fully preserved across execution logs and the testing playground.

Learn more about AI Agents in this Help Center article.

• Workspace Appearance: Theme Updates 🌈

We’re tidying up the Appearance gallery in preparation for upcoming theming updates. As part of this work, we’ve removed several themes with duplicate color schemes. If you were using one of these themes, we’ve automatically migrated you to the closest available match. Please let us know if you have any feedback.

Community News

• Wrike Community Challenge: September 2026 🔥

The September Wrike Community Challenge is starting next week! Participate for a chance to win an exciting prize. Whether you are a longtime member or new to the forum, everyone is welcome to join in. Stay tuned: full details and prize information will be announced soon!

• What's New in Wrike - August 2026 Digest 📰

The August 2026 What's New Digest is here! This month's highlights include expanded AI Agent actions, dynamic role-based @mentions , people and duration field support, flexible Board view columns and custom swimlanes, plus enhanced filtering for dashboard chart and calendar widgets.

Head over to the Community post for full details.

• Template Tuesdays: Software License Request Form 💻

This week’s Template Tuesdays spotlight features the Software License Request Form template. Designed to streamline license procurement, it captures software, vendor, contract, and cost details upfront, maps responses to custom fields, and routes requests for quick approval.

Check out this Community post for more details.

• Wrike-Up Wednesday: What’s Your Favorite Thing About the Wrike Community? 💛

This Wrike-Up Wednesday, share your favorite part, person, or memorable moment from the Wrike Community. Whether it’s a helpful member, an unforgettable conversation, a favorite event, or simply the welcoming vibe, head to the Community post and tell us what keeps you coming back!

• [Onboarding Series] How to Use Approvals in Wrike ✅

Need stakeholder sign-off before work moves forward? Our latest onboarding guide explains how to start and customize approvals, add approvers and due dates, review files, submit decisions, automate status changes, and track pending or completed approvals.

Visit the Community post for the full walkthrough!

Hey Community! 👋

Welcome to Wrike Agent News!

This monthly digest features updates, learnings, and how-tos from the Wrike R&D team.

This month is about people! Agents can now figure out the right people to notify on each work item instead of pinging the same fixed list every run, and judge whether a change is even worth a notification before sending one. As well as that, we have a batch of new default agents, a Teams upgrade, agent pausing and the MCP v2 launch.

Here are the six major shifts landing this month:

💬 New Action: Agents Can @Mention Users Dynamically

Previously, picking a user in the @ picker meant pinging the exact same person every single run . Mentioning different project owners meant cloning the entire agent.

Now, you can reference role groups directly in your instructions, such as "notify the author" or "this task was completed by the assignees but approval is still expected from the approver." The agent resolves these roles at runtime based on whoever is actually assigned to that specific item. Every name lands as a clickable, live @mention that triggers a real notification .

👤 Agents Can Read and Act on People (User) Custom Fields

Agents can now read people-type custom fields, use those values in their reasoning, and assign work based on them (no more “Action failed” when a task’s assignee lives in a custom field instead of the instruction). “Assign this task to whoever’s in the Project Lead field” now works, for both single- and multi-user fields. This closes the gap for teams that store responsible or involved people in fields (populated by intake forms, say) rather than hardcoding them per agent.

🔗 Where It Comes Together: The 2 Features Above, Combined

On their own, the two updates above are useful. Together, they unlock the scenario customers have actually been asking for.

The first taught agents to turn a role into a live, clickable @mention at runtime . The second taught them to read the people sitting inside a people-type custom field. Put them side by side and an agent can read the users listed in a field and @mention those exact people in a comment (dynamic @mentions of users stored in custom fields ).

So an agent can pull the stakeholders from a people field (say, the reviewers on an intake form) and post a comment that notifies those specific users directly: a different set of people on every item, all managed by one agent.

Example prompt: You might give the agent something like:

Post a comment “Hi <users listed in “stakeholders to keep informed” custom field>! Task <task name> from <author> which is assigned to <assignees> is ready for your review!”

🤖 3 New Default Templates

We have rolled out three more prebuilt agents you can turn on without configuring from scratch:

• Revision tracker: On a status change, it updates a custom field and posts a comment.
• Project coordinator: On a new item, it assigns unassigned subtasks, sets missing due dates, and emails the CSM.
• Missed deadline agent: When a finish date passes, it posts to Slack and sends an email.

Good starting points on their own, and useful templates to clone and tweak.

🤝 Wrike AI agents in Microsoft Teams: Sending messages to chats

Agents can now post a notification straight to a Microsoft Teams group chat, not only to a channel. The agent composes a message from the work item’s context, following the instructions you give it in the action, and the Wrike Teams app delivers it (not the agent directly, and not under the name of whoever set up the action). Every message carries a link back to the originating Wrike item.

⏳ Agents That Can Pause

You can now tell an agent to wait before it acts.

An agent triggers on a new Wrike item tied to a Salesforce deal. The deal's data (amount, account, close date) arrives on its own clock: Unito copies it from Salesforce into Datahub, which takes anywhere from a few seconds to a couple of minutes. The agent doesn't wait for it. It reads those fields, finds them empty, and acts on the blanks. Write a wait into the agent's instruction, "wait two minutes before doing anything," and it holds off until the data has landed.

Limitation: this is a timer, not a handshake. The agent waits the time you asked for; it doesn't check whether Unito finished. If the sync runs long, the agent still goes early. Set the wait longer than the sync usually takes, and treat it as reducing the odds rather than eliminating them.

Pro tip: time it before you set it. Watch how long the external update normally takes to land, then add margin.

Conversational Builder Improvements

📁 UI chips in the conversational builder

Choose a location or user from a chip in the chat rather than typing its name and hoping the builder finds it. Especially helpful when appointing an agent to the right location!

💾 Fix: When you try to leave a chat, the builder now reminds you to save and appoint your agent

The conversational builder is meant to be the easy way into agents, but it didn't reliably tell you what to do once yours was ready, so it was easy to close the chat and lose everything. Now, if you try to leave with an agent that isn't appointed yet, a popup lets you Add agent (appoint it right away) or Save for later, so your work doesn't vanish.

🔌 MCP Server v2 is live

Don’t forget we shipped MCP v2: it’s now live, turning your conversations into coordinated work in Wrike. Read more here: Wrike MCP v2 is now live for everyone.

📡 On Our Radar

• Agent action: Create an item from a blueprint
• Agent action: Create a new item with a formatted description
• AI agents working with attachments research - let us know if you’re interested, or schedule a call to share your use cases!

Questions?

Want help setting up any of these? Reach out in the comments. Need a different agent type? Contact your customer success manager. Found a bug or have feedback? Use the in-product feedback link or let us know below.

Found this update helpful? Let us know with a quick reaction (Like👍, Agree 🙂, Insightful💡) or drop your thoughts in the comments below!

• You've already tried using AI agents in Wrike and hit a wall with attachments
• You regularly attach documents, spreadsheets, or images to tasks and projects
• You spend time manually reading, summarizing, or extracting information from those files

What's involved?
A 30-minute Zoom call with me to talk about how AI agents could work with attachments. No prep or demo needed. I'm mainly interested in how your work happens today, and where not having an agent that can handle files slows you down.

👉 Click here to book a time directly on my calendar

Thanks!

TL;DR: Thank you to everyone who participated in our Beta! Wrike MCP v2 is officially out of Beta and available to all users! You can now connect your favorite AI tools (Claude, ChatGPT, Copilot, Gemini, Cursor) directly to Wrike to search, create, bulk-edit work, handle inbox @mentions , and manage whiteboards without context-switching. Check out the Wrike MCP Server Overview to get started today!

Hi Wrike Community 👋

We have some exciting news to share! 🙌

Wrike MCP is the bridge between your conversations and the work that brings them to life. Today, we are opening up Wrike MCP v2! Now, the AI assistant you already use can search, read, create, and update work directly in Wrike with no copy-pasting or context-switching required.

It works across the AI tools you already use:

• Claude 
• ChatGPT 
• Microsoft Copilot 
• Gemini Enterprise 
• Cursor
• and any other MCP-compatible client. 
  

Thank you to our Beta testers for all the great input! Wrike MCP v2 is out of Beta and now generally available for all teams. Check out our documentation for more details.

What's new in v2?

• Unified search - across tasks, folders, and projects, and spaces.
• Cleaner context - custom fields show up only when they're filled, your assistant understands custom item types vs. base types, and it uses the same numeric IDs you see in the Wrike UI, so links and references are right the first time.
• Lighter navigation - your assistant skims the key fields first and drills into detail only where it matters. On average this cuts token usage by about 25 percent.
• Bulk read & create - explore complex hierarchies and build multi-item work (a project with several tasks) in a single call instead of many.
• Inbox -  summarize, prioritize, and reply to @mentions without leaving your assistant .
• Whiteboards - create and work with whiteboards directly through your assistant.

How to Use It (Prompts to Try)

1. "Tell me the current status of our launch project in Wrike."

2. "Summarize all milestones due this week across our top projects and highlight any risks."

3. "Turn these meeting notes into tasks under the Q3 Launch project, assigned with due dates."

4. "What's in my Wrike inbox that needs action today?"

5. "Which approvals are still waiting on me, and how long have they been pending?"

How do I get started?

• Head to the Wrike MCP Server Overview
• Pick your AI assistant, and follow the setup guide.

Works with your go-to AI tools:

- Claude / Claude Code: Search, read, create, and update tasks, projects, and folders directly from Claude, keeping the conversation tied to real work.

- ChatGPT: Work with your Wrike tasks, projects, and folders to create summaries, understand priorities, and prepare clear status updates.

- Microsoft Copilot: Manage Wrike work from Copilot so your team can act on projects without leaving Microsoft 365.

- Gemini Enterprise: Take Wrike actions directly in Gemini so you can move projects forward without switching tools.

- Cursor & other MCP clients: Connect any MCP-compatible client to read and act on your Wrike workspace.

💬 We'd love your feedback!

Comment on this post and tell us what you’re building with Wrike MCP v2! Our team is reading every reply to see how you're using it.

Javier Santolaria

Wrike MCP Team

My name is Andrea and with my team we are currently working on making Wrike's AI Agent builder more intuitive and easier to navigate, especially for folks who are just starting to explore it.

We are looking for users to try out a new prototype and give us honest feedback on the design. You do not need to be an AI expert to participate! In fact, getting a "fresh eyes" perspective is exactly what we need to ensure the new layout makes sense.

The session will be a 30-minute video chat, where we will ask you to try completing a quick task using the new design.

Interested?

Please grab a spot on our calendar: https://greatquestion.co/wrike/csuv8rh

Thanks for helping us build a better experience for everyone!

Hello Community! 👋

Today we’re sharing a guide to set up a Blueprint Kickoff Agent that reads a newly created project, reviews its title, description, and custom fields like Client Tier and Urgency, then automatically assigns sub-tasks, sets due dates, and notifies the CSM with a kickoff summary.

Below, you’ll find the full configuration, prompts, setup instructions, and customization ideas so you can plug this into your project delivery workflows ✔️

Agent Goal / Use Case

Teams that launch projects from blueprints often lose time on the same repetitive setup work every single time:

• Sub-tasks are created but left unassigned.
• Due dates need to be manually calculated from the start date.
• Someone has to notify the CSM or account owner that the project has kicked off.

This takes 20-30 minutes per project. It's done inconsistently. New starters skip steps. By the time someone notices, the project is two sprints in and half the sub-tasks are still unassigned.

An automation can set a fixed assignee or copy a static value. It can't read the project's brief, infer "this is a Tier-1 enterprise account doing a rush campaign", and decide that the creative work goes to the senior creative lead on compressed dates. That's judgment, not matching.

How It Works

1. A new project spawns from your blueprint (or any trigger condition you configure), landing in your designated scope (e.g., "Active Projects").
2. The agent reads the project's title, description, and custom fields (client tier, urgency, project type).
3. Action 1 assigns each unassigned sub-task to the right team member, based on the task's title and the project's context.
4. Action 2 sets due dates on each sub-task, calculated from the project's start date and urgency.
5. Action 3 sends an email via Gmail or Outlook to the account CSM with a kickoff summary.

One trigger, full setup, zero manual intervention.

Features Used:

• Send email action: Agent emails the CSM via your connected Gmail or Outlook account.
• Parent item field reading: Each sub-task reasoning includes the parent project's custom fields (client tier, urgency).
• Action filtering: Each action targets only unassigned or undated sub-tasks (now explicitly surfaced in plain text), so repeats on the same project don't overwrite work.
• Action naming: Actions are labeled "Assign Sub-tasks", "Set Due Dates", and "Notify CSM" in the activity log.
• Start/due date action: Agent writes calculated dates to each sub-task.

Prerequisites

• Custom fields on the project:
  - Client Tier
  - Urgency
  - Project Type
  - Optional: CSM Email if you want the email recipient read directly from the project instead of using tier-based logic

Agent: Blueprint Kickoff Agent

Getting Started

• Under Space Settings, click on AI Agents, and select “Or build it yourself here” option.

• You will then see a split view, with the conversational builder on the left and a configuration section on the right. In the configuration section, you can manually add instructions and select the location, trigger, and actions. We will use this section to build the agent.

Select Agent Scope

Add this agent to one or more tasks, projects, or folders so it can do its work. The agent can watch for triggers and act only on these items and their subitems.

Prompt (Complete Configuration)

Paste the below in General instructions section

Trigger:

When a new item is created

Action 1: Assign Sub-tasks (Change assignee)

Change field values -> Assignee -> the subitems of the work item in the agent’s scope

Filter: Item Type - All Tasks, Assignee - is empty

Add Instructions:

Action 2: Set Due Dates (Change due date)

Change field values -> Change due date -> the subitems of the work item in the agent’s scope

Filter: Item Type - Task, Due date - empty

Action 3: Notify CSM (Send email)

Send email -> Select Gmail/Outlook -> Select account ->  the work item in the agent’s scope

Setup Steps

Step 1: Create and Scope the Agent

1. Space Settings → AI Agents → + Agent/Get Started→ Select “Or build it yourself here” option.
2. Name it Blueprint Kickoff Agent.
3. Agent scope: Select the tasks, projects, or folders that define this agent's scope. The agent will monitor triggers and take action only within these items and their subitems.

Step 2: Add General Instructions & Connect Email

1. Paste the General Instructions above into the agent instructions.
2. Replace all placeholder user chips with real people by typing @ and selecting users from the picker.
3. Update the CSM email addresses to your actual team addresses.
4. If you plan to use the email action, connect the correct Gmail or Outlook account in the agent settings.

Step 3: Configure Actions

1. Trigger Set to "When a new item is created "
   
   
2. Action 1 (Assign Sub-tasks):
   - "Change assignee for sub-items of the work item..."
   - Filter: Where Item type is Task AND Assignee is empty.
   - Paste the Action 1 prompt.
   
   
3. Action 2 (Set Due Dates):
   - "Change due date for sub-items of the work item in the agent’s scope"
   - Filter: Where Item type is Task AND Due date is empty.
   - Paste the Action 2 prompt.
   
   
4. Action 3 (Notify CSM):
   - "Send email for the work item in the agent’s scope"
   - Paste the Action 3 prompt.

Step 4: Testing Playground

Toggle the stage selector at the top from "Configure" to "Test" to try sample prompts alongside your form configuration without changing screens.

1. Test with a sample project that includes a start date, a Client Tier value, an Urgency value, and several unassigned sub-tasks (ideally created from your real blueprint).
2. Verify sub-tasks are assigned, dates are set, and the email payload generates correctly in the test pane.

Customization

• CSM email on the project itself:
  Instead of hardcoding CSM emails by tier in the general instruction, add a custom field "CSM Email" on the project and have the agent read it directly. Update Action 3's recipient logic to: "Read the 'CSM Email' custom field on the triggered project. Send the email to that address."
  
  
• Adjust team structure:
  Replace the user chips to match your organization (e.g., Creative agency: designer, copywriter, traffic manager; Consulting: engagement manager, consultant, analyst).
  
  
• More milestone granularity:
  If your projects have 5+ milestones (not 3), expand the date rules in the general instruction accordingly.
  
  
• Notify multiple stakeholders:
  Add more sentence-style email actions, such as Notify Account Lead, Notify Client, or Notify Delivery Manager.
  
  
• Skip email entirely:
  If you only want assignment and scheduling, remove Action 3.
  
  
• Add status change:
  Add a fourth sentence-style action after assigning and dating to update each sub-task's status (e.g., from "Draft" to "Ready").

Tips

• Start with one blueprint, not all of them: Appoint the agent to one scope tree where one blueprint spawns projects. Verify assignment and dating logic before rolling it out broadly.
• Check the activity log per sub-task: Each sub-task log shows the agent's reasoning (why this assignee, why this date).
• Filter-on-empty prevents re-runs from overwriting work: The Assignee is empty and Due date is empty filters (explicitly displayed in the sentence builder) ensure that manual changes aren't overwritten.
• Keep prompts tight: The shorter and clearer the general instruction, the easier it is to debug.
• User chips are your control plane: You decide which people the agent can choose from. Keep the list current.
• Email sent from your account: The Send email action uses whichever Gmail or Outlook account you connected in settings.

Found this AI Agent setup guide helpful? Leave a reaction (👍 Like / 🙂 Agree / 💡 Insightful) or share your thoughts in the comments below! 👇

Hi everyone 👋

I'm Tomas, a Product Designer at Wrike. My team and I are currently working on an exciting new capability—AI Work Intake integrated directly into the Wrike MCP—and we'd love to invite a few of you to co-develop and shape it with us over the coming weeks!

💡 The Idea: Instead of hunting through a form library and re-typing everything, an AI agent finds the right request form or blueprint and fills it out with you directly in the chat.

🎯 Who's a good fit?

• You rely on Request Forms and/or Blueprints in your daily intake process.
• You're already using or have experience with the Wrike MCP, or you are willing to try.

🤝 What does participation look like?

• A 60-min kick-off call: You'll try the new tools live on your own account.
• Testing at your own pace: Try it out on your real work, with a direct line to our team.
• Bi-weekly check-ins: A short check-in roughly every two weeks to iterate together. (We know calendars are tight: if you can only join the kick-off call and skip the recurring check-ins, that's completely fine and still super valuable to us!)

To be upfront: This feature is still in its early stages and won't be perfect—you might hit a few rough edges, and that's exactly the feedback we need.

Would testing the work intake side of MCP be something you'd like to work on with us?

📬 How to Join: Simply reply to this post or drop an email to tomas.ledba@team.wrike.com.

Thank you so much for helping us build the next generation of AI workflows in Wrike! 🙌

Here are the Agent test results

And here are the final results

The AI frustrations are too much for a simple task

Hello Community! 👋

Today we are sharing a guide to set up an Intake Triage and Route Agent. If your team manages a shared intake folder or request queue, you know how much manual effort it takes to review every submission, judge its actual priority, and ensure the right team is looped in.

Below, you will find the exact chat prompt to feed the Wrike AI Conversational Builder, the final settings configuration, and a breakdown of how it executes once it is live in your workspace ✔️

Agent Goal / Use Case

Standard automation rules apply the exact same static logic to every new task. They cannot read a messy description, evaluate actual urgency, or intelligently determine if a request belongs to Production, Marketing, or Operations.

By simply describing your routing logic to the conversational builder, you can build a single agent that handles content analysis, judgment, and routing all at once.

This setup gives you:

• Native Urgency Mapping: The agent translates text signals straight into Wrike's native Importance field.
  
• Direct Notifications: Task owners receive an immediate, contextual email overview with a link back to the work item.
  
• Built via Chat in the Side-by-Side Builder: Configure using natural language on one side and watch live updates and testing on the other.

💡 The Conversational Shift: You no longer have to manually piece together complex menus or separate tabs. Open the builder, use natural language, and let the sentence-style rule setup keep your settings synchronized live in the Configure → Test stages.

Built Entirely via Chat: The AI Conversational Builder

Open up the Conversational Builder and paste the text block below to build this agent entirely through chat:

Go to Space Settings → AI Agents → + AI agent/Get Started → paste the below in the chat box

The builder will ask clarifying questions or present options as you go. You simply chat back and forth with the builder to create and refine your agent while viewing live form synchronization next to the chat panel.

Final Agent Configuration

How It Works After It Is Built 🚀

Once you review the chat output, save your configuration, and set the scope to your Intake Projects folder (directly in setup or from the Agents Table), the workflow runs entirely behind the scenes:

1. The Trigger Phase: A request submitter fills out an intake form or creates a new task in the scope. As long as the custom dropdown field Request Type is marked as New, the agent instantly wakes up based on the sentence trigger condition.
   
   
2. Parallel Processing: The agent reads the task name, description, and custom fields, evaluating them against the master instructions provided during the chat build. The two actions process concurrently using the same analytical baseline:
   
   - Priority Assessment: The agent looks for keywords like "ASAP," "broken," "launching tomorrow," or explicit calendar dates. It maps its findings directly to the native Importance field.
   - Identifying the Team: It scans the context to figure out the nature of the work (Production, Marketing, Operations) and includes this routing detail in the email payload.
   
   
3. Execution and Notification: The field updates are logged automatically in the Wrike task activity stream. At the exact same moment, an email is constructed and dispatched to the task owner with a synthesized summary, assigned team details, and a direct link back to the Wrike task.

Tips and Customization

• Connect a Shared Inbox: For Action 2, link a team mailbox (like intake-triage@yourcompany.com) during the email account connection phase so notifications come from a centralized system rather than an individual account.
• Monitor with the Grouped Log: Open the agent log to view runs grouped by session. You can see the exact text parsed and the rationale used for the Importance field and email draft in one timeline view.
• Swap Email for Slack or Teams: You can tell the builder via chat to route the summary to a specific Slack or Microsoft Teams channel instead of email.

Let’s hear how your team handles this today!

• What is the closest thing your team does to sort incoming requests right now?
• How does building via the new Conversational Chat and Side-by-Side View compare to hand-mapping forms and triggers?
• If you could add a third sentence-style action to this agent, what would it be? (e.g., Auto-populating a specialized project blueprint based on the team selected?)

Found this AI Agent setup guide helpful? Leave a reaction (👍 Like / 🙂 Agree / 💡 Insightful) or share your thoughts in the comments below! 👇

Hey Community! 👋

Welcome to Wrike AI Agent News! 

This monthly digest features updates, learnings, and how-tos from the Wrike R&D team.

This month is about two major shifts. First, building an agent no longer means assembling triggers and actions in a form. You describe what you want in chat and the agent comes back built and working. Second, agents got noticeably smarter about doing multi-step work: you set the order of the actions, and each step now builds on what the step before it found. Around those two, we have a wave of governance, integration, and platform updates from across the AI team.

💬 Conversational Builder: Describe It, Build It, Appoint It

A CSM sits with a customer who knows exactly what they want: "an agent that emails the requester when a high-priority task sits untouched for a day," but doesn't want to hand-assemble triggers, scopes, and actions. They type that sentence into chat. The agent comes back built, working, and appointed to the work, all in one conversation.

This is the whole capability, not a demo. Build-by-describing now produces agents that actually run. The create flow was hardened so the chat won't hand you a configuration that silently does nothing. You can save an agent by asking ("save it"), and the builder now explains, in plain language, what an existing agent does the moment you open it. This means no more reverse-engineering behavior from a list of actions.

The appointment step closes the loop: a command in chat deploys the agent to its location, so you go from idea to live agent without leaving the conversation. The chat surface itself is steadier. Long responses show a clear timeout notice and retry instead of looking frozen, and your draft survives switching between chat and form view. There's also a thumbs up/down so you can tell us when the builder gets it right.

Pro tip: The conversation is the spec. The more concrete you are about trigger, scope, and action, the better the first build. "When a task in this project is marked Urgent, post a comment tagging the assignee" beats "make it watch for important stuff."

• Why this matters: It democratizes automation. Anyone who can describe their workflow can now build a functioning AI agent in seconds, eliminating the friction of manual form configuration.
  
  
• Current limitations: Appointment ships without location pickers. The agent infers where to act from your command, so name the space, folder, or project explicitly rather than expecting a chooser.

👉 Explore the conversational builder

🔗 Multi-Step Agents: Set the Order, and Each Step Builds on the Last

Two new primitives shipped this month and they're a matched pair:

1. Strict Ordering: Agents run their actions in the order you set. Drag the action blocks in the builder and the agent executes them top to bottom, every run, with the order visible so what you see is what it does.
   
   
2. Context Chaining: Each action now builds on what the action before it did. An agent can triage a task in its first action, and its second action receives that triage decision and acts on it. Before, every action ran blind to the others.

Together, these unlock the read ➔ decide ➔ act pattern that no single action could do previously. A handful of correctness fixes came with it: actions share context even when they run on different items, an action that can't run skips cleanly instead of failing the whole run, and the order in the log now matches the order you set in the builder.

Pro tip: Order your actions deliberately. The later action sees the earlier one's result, so put the "decide" step before the "act on the decision" step, not the other way around.

👁️ The Agent Log Now Reads One Run at a Time

An admin opens the log to answer "what did the agent actually do at 9 a.m.?" and instead of a flat wall of action rows, the log now groups actions by run.

Every action an agent took in a single triggered execution sits together under one session, so you can read its behavior one run at a time. Day-grouping, run times, and spacing were all fixed alongside it, and the log now remembers your view per agent. Filters, grouping, and sort order save individually for each agent and each user, so switching between agents no longer resets your layout.

📧 Email Action: Outlook, Multiple Recipients, and Subitem Support

We have rolled out critical fixes and expansions to the email action based directly on your feedback:

• Outlook Integration: Outlook sending now works end to end. Previously, some scheduled agents logged as failed despite executing internally; this is completely resolved.
• Multiple Recipients: A single email action can now address more than one recipient simultaneously.
• Subitem Scope: The email action now fully accepts subitem scope ("the subitems of that folder, project, or task").
• Working Links: Work-item links composed by the agent now properly resolve to a working URL instead of breaking.

If you previously hit the Outlook wall or the subitem-scope refusal, it's worth re-recommending to your teams.

🤝 New Action: Agents in Microsoft Teams

Agents can now post notifications directly to MS Teams channels! The agent can create a message from the work item's context, following the instructions you give it in the action, and the Wrike Teams app delivers it seamlessly. Every message carries a link back to the originating Wrike item.

Common Use Cases:

• A risk report comes in flagging a high-risk project ➔ Posted to #pmo-risks
• A blueprint kicks off a new project ➔ Announced in #project-launches
• A status field changes on a task ➔ Team channel gets an update with the item link

Pro tip: Name the exact fields you want pulled in. "Include the task title, assignee, and due date" gives you a predictable message every time. "Let the team know" leaves the wording up to the agent's judgment.

Where it stands today: Channel posts only. No group chats yet (that's on the way), and no @-mentioning Teams users.

🛡️ A Steadier Foundation

A handful of stabilization updates keep customer agents responsive under heavy load:

• Faster Responses: Context retrieval was profiled and fixed, trimming the response path so agents return results much sooner.
• Timeout Handling: A genuinely slow response now shows an explicit notice and an automatic resend instead.
• Subitem-Scope Guard Rails: Introduced a hard limit on how many subitems a single run processes, ensuring an agent pointed at a massive folder degrades gracefully instead of dragging down the whole execution.
• Subtask De-duplication: Fixed an issue where agents scoped to "the subitems of..." were running their action twice per subitem.
• Builder Accessibility: The rich-text prompt editor is now fully keyboard-navigable with correct focus through the toolbar.
• Datahub Filter Fix: Selecting more than seven rows in a Datahub field filter no longer hides the text input.

🧩 Expanding Agent Capabilities

📝 Agents Working with Descriptions

A messy intake lands, and the agent rewrites it into a clean summary. A weekly status is due, and the agent synthesizes one right into the description. Simply set up a Change description action and tell the agent to rewrite the whole thing, edit one section, or pull context from other tasks to fill an empty description field.

Current Limitations: Tables render as plain text in logs; nested list indentation is currently imperfect.

⏳ Agents and Duration Custom Fields

Duration-driven workflows like SLA counters, time-in-status, and turnaround tracking can now be automated directly without bridging through a separate number field.

Pro tip: LLMs aren't great at complex math, so keep your duration prompt logic simple!

🔌 A Reimagined MCP Server (v2)

We are launching a brand-new MCP Server experience featuring smarter search, inbox management for summarizing and replying to @mentions , and full approvals support.

If you're already using MCP and want early access, email Javier Santolaria directly.

🏆 Agent of the Month: Intake Triage & Route Agent

Every team with an intake queue has the same job buried in it: a request lands, and a skilled person reads it, decides how urgent it is and which team owns it, and routes it onward. At volume, that's a full-time job done by someone far too senior for it.

Why an automation can't do this: An automation can fire when a request arrives, but it applies the same fixed rule every time. It can't read this request, judge how urgent it actually is, and then route based on the decision it just made. That judgment-then-act is exactly what this month's new chaining makes possible.

The agent: A two-step agent. When a new request lands in your intake project, the first action reads the task and decides its priority and owning team. The second action then acts on that decision, it sets the native Importance field and posts a routed summary to the owning team's Slack channel (or emails the owner), carrying forward exactly what the first action concluded. One agent: read → decide → act, in the order you laid out.

How it works:

1. Trigger: a new task is created in the intake project (scope it with any custom field - e.g. only "Request type = New").
   
2. Action 1 - the agent reads the task and decides priority (High / Normal / Low) and which team should own it.
   
   
3. Action 2 - using that decision, the agent sets the Importance field and posts a short routed summary to the owning team's channel, with a link back to the work item.
   
   
4. The agent log records the run, grouped so you can see the decision and the routing together.

The pattern adapts to support-ticket triage, creative-request routing, and ops intake - anywhere work arrives unsorted and a person currently sorts it.

And yes - you can build this just by describing it. Open the conversational builder and tell it the trigger, the decision, and the routing; it assembles the two chained actions for you.

Build this agent → See this Community post for full details.

📡 On Our Radar

Areas we are actively exploring next:

• More Date Conditions to Trigger On: Moving beyond "has passed" to add more dynamic options like "a week before" or "within the last 7 days."
• Richer Output Chaining: Building on this month's foundation to pass even more complex data payloads forward between actions.

Questions or Feedback?

• Want help setting up any of these features? Reach out to us in the comments!
• Need a different agent type? Contact your Customer Success Manager.
• Found a bug? Use the feedback link in the product or let us know below.

TL;DR: This single-trigger, multi-action Wrike AI agent automatically catches tasks that slip past their deadlines. The exact moment a task's finish date passes, it posts a short at-risk summary to your team's Slack channel and emails the assignee directly. It replaces manual deadline tracking, reduces status meeting surprises, and keeps your project delivery running smoothly.

Hello Community! 👋

Today, we’re sharing a guide to setting up the At-Risk Escalation Agent that automatically detects slipping work, posts a focused message to your project’s Slack channel, and emails the assignee with immediate next steps.

Below, you’ll find the complete configuration, prompts, setup instructions, and customization tips so you can plug this pattern directly into your project delivery and client services workflows ✔️

Agent Goal / Use Case

Every team that runs work to deadlines faces the same quiet problem: a task slips its finish date, nobody notices, and the consequences show up later as a delayed project or a surprised client.

Standard automations can fire when a date passes, but they send the same prewritten note to the same fixed list regardless of how serious the slip is or who needs to hear about it. An automation can notify; it cannot judge. It cannot read the task context, look at the team's specific situation, and choose the right wording for the moment.

The At-Risk Escalation Agent closes that gap by performing the following actions:

• Real-time detection: Scans active tasks and fires the instant a finish date passes.
  
  
• Contextual reading: Reviews the task title, description, assignee, recent comments, and custom fields.
  
  
• Multi-channel routing: Composes a tailored summary for your team's Slack channel and sends an email to the assignee simultaneously.

How It Works

This agent pattern uses a single date trigger to execute two distinct communication actions:

1. The Trigger: A task crosses its finish date without being completed. The agent runs automatically via the "Date reaches" trigger, which is restricted to active tasks so completed or cancelled items are excluded.
   
   
2. Context Gathering: The agent reads the task's title, description, assignee, due date, recent comments, and relevant custom fields.
   
   
3. Action 1 (Post to Slack): The agent writes a short, 3 to 4 line at-risk message naming the task, owner, and context, then posts it to your connected Slack channel with a link back to the work item.
   
   
4. Action 2 (Send Email): The agent writes a fuller version of the summary and emails the assignee via your connected Gmail or Outlook account, providing a clear next step.
   
   
5. The agent log records the run, with the delivery status for both the Slack post and the email visible in one place.

Why this works: One date passing, two channels reached, zero manual follow-up. The team sees the high-level risk in Slack where they already work, and the owner gets an actionable item directly in their inbox.

Features Used

• Post to Slack notification action (NEW this month): The agent posts the at-risk message into the project's Slack channel without anyone needing to log into Wrike.
  
  
• Date-reaches trigger: This trigger runs the moment a date custom field, including the standard Finish date, crosses a threshold. It is restricted to active tasks by the trigger's own settings.
  
  
• Send email notification action: The agent emails the assignee directly via your connected Gmail or Outlook account.
  
  
• Reading task context including custom fields: The agent reads the task title, description, assignee, custom fields, and recent comments to compose the message.

Prerequisites

Before setting up the agent, ensure you have the following pieces in place:

Custom Fields on the Task:

You will want these custom fields available on the task, on its parent project, or on a place the agent can see:

• Assignee Email (Text): The assignee's email address, used as the recipient. If your account already exposes the assignee's email on their user record, you can let the agent read it from the user object.
  
  
• Project Owner Email (Text, optional): Used if you want to CC an internal escalation contact on the email.
  
  

A Slack Channel for the Project 

• Connect your Slack workspace once at the account level, then choose a single destination channel during setup. Pick the channel where the project team already collaborates.

A Gmail or Outlook Account

• Ensure your work mailbox is connected in your agent settings so emails send from your address.

Agent Configuration

Trigger Settings

Getting Started

• Under Space Settings, click on AI Agents, and select “Or build it yourself here” option.

• You will then see a split view, with the conversational builder on the left and a configuration section on the right. In the configuration section, you can manually add instructions and select the location, trigger, and actions. We will use this section to build the agent.

Select Agent Scope

Add this agent to one or more tasks, projects, or folders so it can do its work. The agent can watch for triggers and act only on these items and their subitems.

Prompt (Complete Configuration)

Paste the below in General instructions section

Trigger:

When Field Value Changes -> Date Reaches -> Finished date -> Has passed -> Time zone -> Select scope

Action Configurations

Action 1: Post to Slack

Send Slack message -> Select Slack Account -> the work item where the date reaches

Add Instructions:

Action 2: Send Email

Send email -> Select Gmail/Outlook -> Select Account -> the work item where the date reaches

Add Instructions:

Step-by-Step Setup

Step 1: Prepare the Fields and Channels

• Set up your Assignee Email custom text field if required.
• Ensure your core team Slack channel is ready and active.
• Authenticate your Slack workspace and your email client (Gmail or Outlook) in your Wrike account settings.

Step 2: Create and Scope the Agent

1. Space Settings → AI Agents → + Agent/Get Started→ Select “Or build it yourself here” option.
2. Name the agent At-Risk Escalation Agent.
3. Agent scope: Select the tasks, projects, or folders that define this agent's scope. The agent will monitor triggers and take action only within these items and their subitems.

Step 3: Configure Triggers & Actions

1. Trigger: Set to "When Finish date passes in [Scope]..."
2. Action 1 (Slack): Select "Post Slack notification...", choose your target channel, and paste the Action 1 instruction block.
3. Action 2 (Email): Select "Send email notification...", configure your sender options, and paste the Action 2 instruction block.
4. Paste the General Instructions into the main prompt window.

Step 4: Testing Playground

Toggle the stage selector at the top from "Configure" to "Test" to try sample prompts alongside your form configuration without changing screens.

1. Select or create a test task inside the Test stage view.
2. Fill out the Assignee Email field and set its finish date to a time in the past.
3. Verify the agent triggers immediately, generating the 4-line Slack summary and the structured email draft in the test pane.

Customization Examples

• Project Owner CC on Slack: If your project owner does not always read the team channel, add a second email action that alerts them when the priority field is "High" or "Critical."
• Different Slack channels by region or client tier: Set up multiple instances of this agent (one per channel), with each instance scoped using sentence-style action filters to route Critical EMEA tasks to one channel and Critical AMER tasks to another.
• Add a soft warning version: Configure a sister agent using date conditions that posts a soft warning a few days before the finish date.

Expected Results

• Immediate Impact: Deadline overruns are caught instantly, eliminating multi-day gaps where slipped tasks go unnoticed.
• Team Benefits: Assignees receive direct, non-alarmist emails outlining exactly how to remediate the timeline.
• Organizational Benefits: Client-facing project timelines stay protected, team accountability increases, and risk metrics become highly transparent across your target channels.

Caveats

• The Slack post action commits to one channel per action: If you need messages in multiple channels, set up one action per channel.
• Email send requires a connected Gmail or Outlook account: Emails originate from your mailbox, allowing the assignee to reply directly to a real person.
• The Slack notification does not support threading or @-mentions yet : The agent posts as a top-level message and identifies assignees in the text body.
• The trigger uses the standard Finish date field: Point the trigger at a specific date custom field if your team uses one as its working deadline.

Found this AI Agent setup guide helpful? Leave a reaction (👍 Like / 🙂 Agree / 💡 Insightful) or share your thoughts in the comments below! 👇

Hi Community! 👋

Wrike AI agents are designed to handle repetitive manual tasks, such as updating status fields, copying dates, and routing tasks. However, building them used to mean starting from scratch and needing to know the exact triggers, scopes, and actions before you began.

To simplify this process, we have shifted from setup forms to a conversational approach. Now you can build and configure an agent simply by describing what you want it to do 

How the Conversational Builder Works 👇

Instead of requiring you to fill out a form, Wrike guides you through a step-by-step chat to configure your agent.

💬 It starts a conversation

You can describe your goal in plain terms, such as: "I want my project status to stay accurate." The builder then asks specific questions to establish the mechanics, such as whether the agent should run on a schedule or when a task is completed, and whether it applies to a single folder or a whole project. Each answer determines the next question.

🗺️ It references your actual Wrike workspace

The builder reviews how your specific space is set up, including your custom fields, workflows, and item types. When it suggests an action, it uses your actual field names rather than generic examples so the agent matches how your team works.

🛑 It clarifies what is possible

If you request a capability that agents do not currently support, the builder will let you know immediately. It will then offer the closest working alternative and explain the difference, ensuring that any agent you build will run successfully.

⚡ Build, save, and turn it on in the chat

Once the conversation is complete, the full agent is displayed with all its triggers and actions. You can review the details, make changes, save the agent, and turn it on directly within the chat window. No changes apply to your live workspace until you choose to activate it.

What to Build First 💡

You don't need to be an expert to get started. Here are a few quick, highly useful agent workflows teams are setting up in a single short conversation:

• Status Syncing: Automatically keep a parent project's status field accurate as its underlying individual tasks are completed.
  
  
• Timeline Protection: Copy a due date across dependent tasks so that if one deadline slips, the shift reflects everywhere instantly.
  
  
• Instant Routing: Route a task to the correct team member the exact moment they are mentioned or tagged.

How to Get Started

Creating a New Agent:

1. Navigate to your space's AI Agents settings.
2. Select to start a new agent.
3. Alternatively, open an existing agent and click Edit in chat to update your current workflows visually.

Editing an Existing Agent:

1. Open your relevant space settings and click AI agents.
2. Open the specific agent you want to update.
3. Click the Chat button to describe your changes and update the setup directly in the conversation.
   

For more details, please refer to our Help Center article.

✅ Availability

This feature is live and available for users on Business, Pinnacle, and Apex plans.

We’d love to hear from you! What is the first repetitive task you’re going to allocate to an agent by talking with Wrike AI? Share your use cases or drop your questions for the team below! 👇

TL;DR: Agents can now reach into Slack channels as well as email inboxes. You can use any of your custom fields to tell an agent which slice of work to look at, and agents can now read and set a task's Importance (High, Normal, Low) based on what they find. Additionally, agents now read the full record behind a Link-to-Database field and can see who has approved a task, plus a handful of behind-the-scenes fixes make Wrike steadier on busier projects.

Featured example: the At-Risk Escalation Agent. When a task's finish date passes, the agent posts an at-risk summary into the project's Slack channel and emails the assignee the same moment.

Hey Community! 👋

Welcome to Wrike Agent News!

This monthly digest features updates, learnings, and how-tos from the Wrike R&D team.

This month's theme is earning your trust on real work. April was about giving agents finish power: starting approvals, sending emails, and kicking off blueprint projects. May is about making sure the agent does exactly what you told it to do on the bigger, busier projects your team actually runs.

Five things shift this month:

• Broader reach: Agents reach further by connecting to Slack alongside email.
  
  
• Targeted focus: They see exactly the slice of work you point them at using any custom field and stacked filters.
  
  
• Urgency flags: They can now set the Importance field on a task (High, Normal, Low) based on what they read.
  
  
• Deeper knowledge: They know more about each piece of work they read, including Link-to-Database records and approver decisions.
  
  
• Behind-the-scenes fixes: System updates close real-world gaps in how agents behave on busy projects.

💬 New Action: Agents Can Post to Slack Channels

A risk report flags a slipping project, and the team finds out in their #pmo-risks channel instead of a Wrike comment they need to remember to open. A blueprint kicks off, and the announcement lands in #project-launches. People who live in Slack now hear from the agent in Slack, instead of having to remember to check Wrike.

You connect your Slack workspace once, then pick a single destination channel when you set up the action. That channel is where every message from this action goes. One action maps to one channel. If you need messages to land in different places (risks to #pmo-risks, launches to #project-launches), set up a separate action for each. The agent writes the message based on the work item it is responding to and the instructions you wrote into the action. The message is delivered by the Wrike Slack app and includes a link back to the Wrike work item.

Pro tip: Be specific in your action instruction. "Post the task title, assignee, and due date" works well. "Notify the team" gives the agent too much creative freedom over what it writes.

Why this matters: An agent that flags something but only posts a Wrike comment does not always get heard if the message sits where the wrong people are looking. Slack changes that. Combined with the email action from April, the agent can now reach people in two places: their email inbox or the Slack channel where their team already lives.

Current limitations: No replies into threads and no @-mentioning Slack users yet . Threading and mentions are on our roadmap.

⚡ New Action: Agents Can Set Importance

A rush order arrives with a 48-hour turnaround, a customer comment uses the word "URGENT", or a task is about to miss its SLA. Until this month, agents could spot those signals in the content, but they could not change the Importance field, the native High / Normal / Low marker the rest of the team actually filters on. Teams worked around it with custom priority fields, which split the signal across two places and cluttered dashboards.

This month, agents can read the current Importance value, decide what it should be based on the task content, and set it - High, Normal, or Low, directly on the native field. The activity log records the previous value, the new value, and the reason the agent gave for the change.

Pro tip: Be specific about what counts as High for your team. "Mark as High when the task description contains URGENT, ASAP, or escalation language" is concrete. "Be smart about priority" leaves the agent guessing.

Why this matters: This closes the gap between the agent detecting urgency in the content and flagging it where the team actually watches. If your team already filters or sorts by Importance, the agent now contributes to that signal directly on the native Wrike field rather than requiring a custom-field workaround.

Scope: Native Importance field only (High / Normal / Low). Custom fields used as priority proxies are still updated through the existing Update Custom Fields action.

🔎 Sharper Filtering: Any Custom Field, Any Slice of Work

Most teams organize their work with the custom fields they actually use, such as Region, Client Tier, Priority, Effort, a dropdown for stage, or a checkbox for "needs legal review." This month, you can use any of those fields to tell an agent which slice of work to look at. Write something like "tasks where Region is EMEA" or "tasks where Client Tier is Tier 1 and Effort is more than 8 hours" into your agent's instructions, and the agent will work with only the matching tasks.

You can also stack filters. Narrow a set of projects by status, then ask the agent to find the tasks assigned to a particular person within those projects. The agent reads only the matching tasks inside the matching projects.

Pro tip: Use the custom-field vocabulary your team already uses. It does not matter what kind of field you use (text, dropdown, number, percentage, checkbox); they all work the same way now.

Why this matters: When you tell an agent which slice of work to focus on, Wrike narrows the work down before the agent reads anything. The agent looks at exactly the tasks you asked for. This consistency keeps the agent reliable on real projects, even when a project has hundreds of tasks.

📖 Richer Context: Agents See More of Your Work

Where Sharper Filtering controls which work the agent looks at, Richer Context controls how much the agent knows about each item once it is looking. Two changes this month expand what the agent reads.

Link-to-Database fields reveal the full record

Before this month, when an agent reads a task with a Link-to-Database field pointing at a record - say, a "Customer" field pointing at a Customer database, the agent saw only the name of the linked record, like "Acme Corp." Now the agent sees every column on that record: the CSM, the contract tier, the regional contact email, the renewal date, and whatever fields you have configured on the linked database.

This removes the need to copy columns from a linked database into separate custom fields on the work item just so the agent can read them.

Pro tip: If you have built agents that rely on duplicated fields because the agent could not see the linked record, you can stop duplicating now. Point the prompt at the linked field directly.

Agents see approver decisions

When an agent reads a task that has an approval flow on it, the agent can now see each approver's decision by name, decision, and timestamp (who approved, who rejected, and who has not yet responded).

The top use case for this is reminding only the approvers who have not responded after a couple of days. Before this month, the agent had no way to tell who was still pending. Now it does.

Pro tip: Refer to approvers by their role in the prompt: "Remind any approver whose decision is still pending" and let the agent resolve the names from the approval chain. You do not need to list approver names yourself.

Why this matters: An agent that knows more about each piece of work makes better decisions. The same prompt that used to fall back on just the customer's name now has the customer's CSM email, contract tier, and region to work with. The change is invisible in the UI, but highly visible in the quality of the agent's actions.

🛡️ A Steadier Foundation

A handful of changes improve the experience for teams running agents on active projects:

• Blueprint projects: Agents on blueprint projects now run on every task in the new project. Previously, the agent fired only on the first item in the new project. Now every task created from the blueprint triggers the agent as expected.
  
  
• Sub-item scoping: The "all subitems" scope reliably runs on every sub-item when a template folder is copied, ensuring all children run as expected.
  
  
• Log filters: The agent log now has filters. Narrow your view by agent, by action, or by time, instead of scrolling through a long list of every run.
  
  
• Importance filtering: Agents that filter by Importance no longer fail when there are multiple custom fields sharing a name.

🤖 Agent of the Month: At-Risk Escalation Agent

Every team that runs work to deadlines faces the same problem: a task slips its finish date, nobody notices, and the consequences show up later as a delayed project or a surprised client.

An automation can run when a date passes, but it sends the same pre-written message to the same list every time. It cannot read a specific task, judge how serious a particular slip is, and pick which channel and which person to escalate into. That is the difference between a generic notification and a smart escalation.

The Solution

A three-step agent built on the Date-reaches trigger and this month's new Slack action. When a task's finish date passes, the agent reads the task, posts a focused summary into the project's Slack channel, and emails the assignee directly at the same moment.

How it works

1. Trigger: Date reaches → Finish date → Has passed. The agent runs automatically the instant an active task in the project crosses its finish date without being completed.
   
   
2. Context: The agent reads the task details, including its title, the assignee, relevant custom fields (like the project owner's contact details), recent comments, and the original due date.
   
   
3. Action 1 (Post to Slack): The agent writes a short at-risk message and posts it into the project's Slack channel with a link back to the work item.
   
   
4. Action 2 (Send email): The agent emails the assignee directly with a longer version of the summary, detailing what the task is, what slipped, and the expected next step.

The agent log records the run, providing delivery status for both the Slack post and the email. This pattern adapts easily to client-delivery escalation, internal milestone tracking, and contract-deadline reminders.

Click the link below for the complete configuration, prompts, and customization guide.

👉 Solution and configuration

🧩 Combining This Month's Features

Picture a customer success team running an agent on their renewals pipeline. Each task in the pipeline has a Customer field that links to a Customer database containing the columns the CSM team uses to organize their work: CSM owner, contract tier, region, and renewal date. The team wants a daily summary posted into their CSM Slack channel of Tier-1 renewals coming up in the next 30 days.

The Automated Workflow

• Trigger: Scheduled (weekday mornings at 8 AM)
  
  
• Scope: Tasks in the renewals folder where the linked Customer's Contract Tier is "Tier 1" and Renewal Date is within the next 30 days.
  
  
• The Agent Action: The agent reads each matching task, reads the full Customer record behind the link (CSM's name/email, region, renewal date), and posts a short summary into #cs-renewals. Each renewal is named alongside the responsible CSM, the renewal date, and a link back to the work item.

This utilizes three of this month's new features at once: sharper filtering picks the right tasks, richer context reads the CSM details from the linked customer record without manual data duplication, and the new Slack action puts the result directly in front of the team.

Questions or Feedback?

• Want help setting up any of these features? Reach out to us in the comments.
  
  
• Need a different agent type? Contact your Customer Success Manager.
  
  
• Found a bug or have feedback? Use the feedback link in the product or let us know in the comments.

On Our Radar

Areas we are actively exploring next:

• Agent log grouped by run: Making it easier to follow one run of an agent from start to finish instead of scrolling through a long chronological list.
  
  
• More date conditions to trigger on: Moving beyond "has passed" to add more ways to trigger or filter on date fields, including options like "a week before" or "within the last 7 days."

Your feedback shapes our roadmap. Keep testing, keep sharing what works (and what doesn't).

Hello Community! 👋

Today we’re sharing a guide to set up a Project Health Monitor - a pair of chained AI agents that scans every project in your portfolio on a schedule, classifies each one as Green / Yellow / Red based on the state of its sub-tasks, and helps surface risk before it turns into a reporting fire drill.

Below, you’ll find the full configuration, prompts, setup instructions, and customization ideas so you can plug this into your delivery workflow ✔️

Agent Goal / Use Case

PMOs typically learn that a project is in trouble after the slip has already happened, usually because a manager spotted a red row in a status report, or a client raised an issue. By the time the conversation happens, the recovery options are smaller than they would have been a week earlier.

A standard automation rule can flag one late task or one budget overrun, but it cannot reason across the entire project hierarchy to decide whether the project as a whole is still healthy.

This setup gives you:

• A single Project Health field on every project - refreshed on a schedule.
  
• A comment + email the moment a project turns Red.
  
• A clean, testable, two-step pipeline: detection and notification are separate agents you can tune independently.

How It Works 

The pattern uses two chained agents, not one multi-action agent. Here's why that matters:

Note: Actions inside a single Wrike AI agent fire in non-deterministic order. They run in parallel. There is no Action 1 → Action 2 → Action 3 sequence. 

If you need "detect health, then notify based on the result," a single multi-action agent will not work, the notification action might run before the detection action.

The Health Monitor uses two agents:

Agent 1 (Health Detector) 

• scheduled weekly. Walk the full project hierarchy. Sets the Project Health custom field to Green / Yellow / Red on each project.

Agent 2 (Health Notifier)  

• Triggers when Project Health changes to Red. Posts a roll-up comment on the project AND emails the project owner.

Why two actions in Agent 2?

The comment action and the email action don't depend on each other. They each independently walk the project's sub-task tree and compose the roll-up message from the same source data. They produce consistent output because they're reading the same facts, not because one runs before the other.

Features Used

• Recursive hierarchy reasoning - Agent 1 walks every level of sub-projects and sub-tasks, applying overlapping filters across the full tree (overdue + high-priority + over-budget).
  
  
• Subitems of the appointed item scope - Agent 1 acts on every project under the portfolio folder it's appointed to.
  
  
• Send email action - Agent 2 emails the project owner via your connected Gmail or Outlook account.
  
  
• Scheduled trigger - Agent 1 runs on a cadence (weekly, daily, hourly), not on a user event.
  
  
• Custom-field trigger for chaining - Agent 2 triggers on Agent 1's field write. This is the canonical Wrike agent-chaining pattern.
  
  
• Action filtering - Agent 2's actions only fire when the new value is "Red", so Yellow and Green updates don't generate noise.
  
  
• Action naming - Agent 1's action is "Set Health"; Agent 2's actions are "Post Roll-up" and "Notify Owner". Easy to find in the activity log.

Prerequisites

1.Custom field: "Project Health"

• Type: Single-select dropdown.
• Values: Green, Yellow, Red (and a blank default).
• Applied to: Projects in the portfolio folder.

2.Project owner

Each project needs an assignee (or a custom field like "Project Owner" with an email). Agent 2 uses this to address the email. If you don't track project owners today, add a simple "Owner Email" custom field to your projects before deploying.

3.A portfolio folder

A folder that contains all the projects you want monitored. Agent 1 will be appointed here.

Agent Configurations

Agent 1 - Health Detector

Getting Started

• Under Space Settings, click on AI Agents, and select “Or build it yourself here” option.

• You will then see a split view, with the conversational builder on the left and a configuration section on the right. In the configuration section, you can manually add instructions and select the location, trigger, and actions. We will use this section to build the agent.

Select Agent Scope

Add this agent to one or more tasks, projects, or folders so it can do its work. The agent can watch for triggers and act only on these items and their subitems.

Note: Select the same Agent Scope as well while building the next two agents.

Prompt (Complete Configuration)

• Paste the below in General instructions section

Trigger:

At a Scheduled Time -> Select Days, Time and Zone

Action:

Change Custom Field -> Project Health (Example Custom Field) -> in the subitems of the work item in the agent’s scope

Agent 2 - Health Notifier

Agent Scope

• Same scope as Agent 1

Prompt (Complete Configuration)

• Paste the below in General instructions section

IMPORTANT: This agent has two actions. They run in parallel - there is no Action 1 then Action 2. Each action below independently reads the project and its sub-tasks and composes its own roll-up using the format above. Do NOT have one action reference the other.

Trigger:

When custom field -> Project Health changes -> in selected scope

Action 1 - Post Roll-up (Post a comment):

Post a comment -> in the work item where the trigger happens

Add Instructions:

Action 2 - Notify Owner (Send email):

Send email via Gmail/outlook -> Select email account -> the work item where the trigger happens

Add Instructions:

Setup Steps

Step 1:  Create the custom field

1. Go to Space Settings → Custom Fields.
2. Create a single-select dropdown called Project Health with values Green, Yellow, Red.
3. Make sure it applies to Projects in your portfolio folder.

Step 2: Deploy Agent 1 (Health Detector)

1. Space Settings → AI Agents → + Agent/Get Started  → Select   “Or build it yourself here” option.
   
   
2. -Name: Project Health Detector
   - Agent scope: Select the tasks, projects, or folders that define this agent's scope. The agent will monitor triggers and take action only within these items and their subitems.
   - Trigger: "Every week on Monday at 9:00 AM "
   - Action: "Change custom field [Project Health] for sub-items of the agent's scope..." (Set filter: Where Item type is Project). Name it "Set Health".
   
   
3. Paste the General Instructions for Agent 1.
   
   
4. Testing Playground:
   Toggle the stage selector at the top from "Configure" to "Test" to try sample prompts alongside your form configuration without changing screens.

Step 3: Connect email (one-time)

• In Agent 2's settings, connect your Gmail or Outlook account that should be the sender for project-health emails.
• Use a shared team mailbox if available - emails originate from whichever connected account you specify.

Step 4: Deploy Agent 2 (Health Notifier)

1. Click "+ Custom AI agent" again.
   
   
2. - Name: Project Health Notifier
   - Agent Scope: Set it to the SAME portfolio scope as Agent 1.
   - Trigger: "When custom field [Project Health] changes in [Scope]..."
   - Action 1: "Post comment in the work item where the trigger happened..." (Name: "Post Roll-up"). Paste Action 1 prompt.
   - Action 2: "Send email for the work item where the trigger happened..." (Name: "Notify Owner"). Paste Action 2 prompt.
   
   
3. Paste the General Instructions for Agent 2.
   
   
4. Test in the Playground by simulating a project's Health field flipping to Red.

Step 5: Verify the chain

• Find a project with overdue sub-tasks (or create one).
• Trigger Agent 1 manually from the Playground (or wait for the scheduled run).
• Confirm the project's Health field flips to Red.
• Confirm Agent 2 fires automatically - comment posted on the project, email landed in the owner's inbox.
• Check the activity log: each action should show its name ("Set Health", "Post Roll-up", "Notify Owner") for easy debugging.

Customization

Different health thresholds

The classification rules in Agent 1 are tuned for general PMO use. For tighter delivery teams, swap the thresholds, for example:

• Red if 1 sub-task is overdue AND priority is High (instead of 2+).
• Yellow if any sub-task is over-budget (instead of 2+).

Different cadences

• Daily for fast-moving teams (creative agencies, support).
• Weekly Monday 9 AM for standard PMO.
• Hourly during a critical sprint or launch window.

The trigger setting controls cadence with no prompt change.

Email recipient logic

Hardcoded recipient lookup gets messy past 10 projects. Add an Owner Email custom field on the project itself and update the Agent 2 general instruction to: "The project owner is the value of the 'Owner Email' custom field." This scales without prompt edits as projects come and go.

Add a Yellow-tier notifier

Clone Agent 2, change the trigger filter to New value = Yellow, and downgrade the messaging - for example, comment-only (no email). This gives you early warnings without the alarm fatigue.

Add an SLA escalation

Add a third agent that triggers on Project Health = Red AND the project has been Red for 3 consecutive days (use a "Days at Red" counter custom field updated by Agent 1). Action: post a comment that @mentions the program manager . Same chaining pattern, one more link.

Send to Slack or Teams

Agent 2 cannot post directly to Slack/Teams today. The current pattern is: agent emails the project owner, then your existing email-to-Slack integration (or Slack's built-in email channel address) routes it to a channel. Long-term, the Wrike MCP server can read the agent's output and forward it via your LLM of choice.

Tips

• Start with one portfolio folder. Don't deploy across all spaces on day one. Pick a folder with 5–15 projects, run the chain for two weeks, and tune the thresholds before scaling.
  
  
• Use the activity log to validate Agent 1's reasoning. Each project's run shows why the agent picked Red, Yellow, or Green. If you're seeing too many Reds, the rules are tighter than your team's tolerance, relax them.
  
  
• Don't let the actions in Agent 2 reference each other. The comment and the email are independent. Each composes its own roll-up from the same source data. If one of them ends up shorter or differently worded, that's fine, the comment is for the project record, the email is for the owner's inbox.
  
  
• Filter Agent 2 to Red only. Without the filter, every Green and Yellow update will fire the comment and email actions too. The filter on the trigger is the single most important guard against notification noise.
  
  
• Re-runs are safe. If the project goes Red → Green → Red again, Agent 2 fires again. That's correct behavior, each transition into Red is a new signal. If you want to suppress repeats, add an "Already Notified" checkbox the agent reads in its filter.

Let’s hear how your team handles this today

• What's the closest thing your team does manually today? (Friday roll-up email? Monday status meeting? A cron job + Slack webhook?)
  
  
• Would Yellow be more useful than Red for your workflow, i.e., is "early warning" or "explicit alarm" the bigger gap?
  
  
• If you could customize just one thing first, what would it be?  Thresholds, cadence, who gets notified, or the roll-up format? 

Found this AI Agent setup guide helpful? Leave a reaction (👍 Like / 🙂 Agree / 💡 Insightful) or share your thoughts in the comments below! 👇

Hello Community! 👋

Here’s a simple best practice for anyone building AI Agents and looking for a way to keep their actions focused and relevant 🎯

As agents become more capable, it can be helpful to let them monitor broader areas and make decisions automatically. At the same time, when their scope is too wide, they may activate more often than needed or interact with items outside the intended workflow ⚙️

This can lead to things like 👇

• Agents responding to items outside the target area  
• More updates or actions than expected  
• Extra activity across larger workspaces  
• Increased processing across items  
• Less clarity around how the agent is behaving  

The good news is that this is often not a model issue, it’s a scoping issue.

✅ The best practice

Set clear boundaries for:

• Where the agent should activate
• Where the agent is allowed to take action
• Which items it should process

This helps agents stay targeted, consistent, and easier to manage.

👀Keep trigger scope focused

Trigger scope defines where the agent starts paying attention.

A good starting point is to make this as specific as possible.

• Use top-level items only if the agent is meant to support higher-level workflows
• Include sub-items only if they are truly part of the process

This helps the agent activate only where useful decisions are actually happening.

🔒Keep action scope tighter

Action scope defines where the agent can make changes.

In many cases, an agent may need to observe a wider area, while only taking action in a smaller, specific part of that space.

🎛️ Add filters for more precision

Filters add another layer of control by helping the agent focus only on relevant items.

Such as:

• Item type: such as tasks
• Dates: such as items due soon or recently created
• Naming conventions:  such as prefixes, tags, or project codes

These filters help keep processing aligned with the workflow you want to support.

🧪 Test before going live

Before enabling an agent in a live workspace, test it in a sandbox or demo environment.

This helps you confirm:

• When the agent activates
• Which items it selects
• What actions it takes

A quick test run can make setup much smoother later.

📜 Review logs and reasoning

Logs and reasoning are very useful when refining agent behavior.

They can help you understand:

• Why the agent activated
• Why it chose a certain item
• Which condition allowed the action
  

This makes it easier to fine-tune the setup and gives teams more confidence in how the agent works.

We hope this helps you build AI Agents that are more focused, effective, and easier to trust 🤝

If you’ve found other helpful ways to guide agent behavior, we’d love to hear them in the Comments 😊

Thank you to everyone who joined our recent live session on Wrike AI Agents! ✨

We had a fantastic discussion around how AI Agents can help teams automate smarter, move faster, and simplify repetitive work.

If you couldn’t attend live, or want to revisit the session, you can watch the recording here:
👉 Recording link

During the session, we explored:

• How Wrike AI Agents work
• Real use cases and workflows
• A live AI Agent build
• Latest updates and what’s coming next

We also wanted to share a few additional resources to help you continue exploring Wrike AI:

🧠 Wrike AI Forum
Follow the forum to stay updated on Wrike AI news, feature updates, discussions, and our monthly AI digest.

🤖 AI Agent Library & Inspiration
Explore ready-to-copy AI Agent prompts and use cases shared by our team and Community members.

💡 To receive updates directly in your inbox, click Follow on the forum and enable email notifications for new posts.

🙋Have additional questions about AI Agents? Ideas, feedback, or use cases you’d like to share? Drop them in the comments below - we’d love to hear from you.

We’re continuing to work closely with our product team to help you get the most out of Wrike AI, and your feedback plays a huge role in shaping that journey 👇