monday.com AI agents monitor your boards, decide what needs doing based on rules you set, and then carry out the work: creating and updating items, assigning owners, changing statuses, drafting messages, researching and summarising, and following up across connected apps like Gmail, Slack, Salesforce and HubSpot. They differ from automations because they interpret context and choose an action rather than following a fixed if-this-then-that rule. In May 2026 monday.com rebuilt itself around them, repositioning from a work management platform to an AI Work Platform. The catch is that an agent is only as good as the boards, permissions and process underneath it.
An agent is an AI worker that lives inside your monday.com account and operates in the context of your actual work. It watches for a trigger, reads the boards and documents you have given it access to, applies the instructions and priorities you have written, and then takes action inside the platform. Crucially, it runs whether or not you are logged in, and it works within your existing permissions rather than around them.
There are two routes in. You can use a pre-built expert agent for a common job, such as the CRM lead agent or the CRM sales agent. Or you can open the agent builder, describe what you want in plain English, and let it assemble the instructions, triggers and tools with you. Neither route requires a developer, though you do need admin permissions on the account.
The most common point of confusion is how agents relate to the automations you already have. They are not the same thing, and one does not replace the other.
Evaluate the situation, weigh it against your priorities, and pick the next best action inside the boundaries you have defined.
Follow fixed if-this-then-that logic. Utterly predictable, instant, and still the right answer for most routine board mechanics.
The honest rule of thumb: if you can write the rule down completely, use an automation. If the decision needs someone to read a request and judge what it means, that is agent territory.
An agent can read a request as it lands, decide how urgent it is, route it to the right owner and set the status, without a human doing the sorting first. This is the single most common starting point we see, because most teams already have a queue that someone is manually distributing every morning.
With web search enabled, an agent can go and find information rather than only working with what is already on the board. Combined with access to your documents and, optionally, the meeting records and transcripts held by the AI notetaker, it can produce briefings that pull from several sources at once.
Agents connect out to email providers, messaging platforms, calendars, CRMs, developer tools and file storage. The list of native integrations runs past thirty and includes Gmail, Outlook, Slack, Teams, Salesforce, HubSpot, GitHub, Stripe and Google Drive. If the connector you need does not exist, tools on demand can generate one from a plain-English description, though that feature is still in beta and only supports API key authentication.
Rather than one sprawling instruction set, an agent's work is broken into jobs. Each job has its own trigger and its own scoped instructions, and jobs can run in parallel or on a daily, weekly or monthly schedule. Every agent gets two jobs by default: one for when it is assigned to an item, one for when it is mentioned.
Agents retain memory across runs. They can hold on to your preferences, your working style, and behavioural rules such as always checking before an email goes out. You can inspect that memory, add to it, ask the agent to forget something, or switch memory off entirely if you would rather each run started clean.
Agents can be reached from outside the platform through WhatsApp and Telegram, and can sit in on live meetings alongside the notetaker, responding in the meeting chat with the same tools and data they use elsewhere. On the board itself, live agent mode shows the agent's avatar moving between cells as it works.
monday.com now ships three overlapping ways to get AI help, and teams routinely reach for the wrong one. Sidekick is the conversational assistant you talk to. Automations are the rule engine. Agents are the thing that executes on its own.
| Capability | AI Agents | Automations |
|---|---|---|
| Runs without you present | ✓ | ✓ |
| Interprets ambiguous input | ✓ | ✗ |
| Generates written content | ✓ | ✗ |
| Chooses between several actions | ✓ | ✗ |
| Fully predictable output | Bounded | ✓ |
| Consumes AI credits | ✓ | ✗ |
| Connects to external apps | ✓ | Via integrations |
| Full run log with reasoning | ✓ | Basic |
| Requires admin permissions | ✓ | ✗ |
Find the AI agents tab in the left navigation, then choose new agent. You can describe the agent you want in plain text, pick one of the suggested starting prompts, or start from blank.
Left navigation → AI Agents → + New Agents
Worth checking the agents made for you section first. Those templates are generated against your role and the data already sitting on your boards, so they are often closer to what you need than a blank page.
The brain tab holds the instructions the builder has drafted for you. Read them and tighten them. An agent cannot act outside what it has been explicitly permitted to do, so vague instructions produce a vague agent that does very little, and over-broad instructions produce one that does too much.
This is the step teams skip, and it is the step that decides whether the agent is useful.
Under knowledge and access, connect only the boards the agent genuinely needs, and decide per board whether it can read only or read and write. You can also upload supporting context as PDF, DOC, TXT, Excel or CSV, which is useful for FAQs, service standards, or anything governing how the agent should respond.
The agent will proactively suggest boards it thinks are relevant. Treat those as suggestions, not instructions.
Describe the tool you need in the builder chat and it will identify the integration, ask any clarifying questions, and surface the connection step. Read tools go live immediately. Write tools arrive inactive on purpose, so you review the setup before anything is sent, created or deleted on your behalf.
Guardrails are worth the five minutes. You can restrict an agent to one Slack channel, or to emailing only certain contacts, and the setup applies your answer across every relevant tool at once.
The activity tab logs every run: what the agent did, when, which apps it used, and how many AI credits it consumed. When something fails you get a notification and a panel explaining what went wrong and what to change.
Pause the agent at any point if the behaviour is not right. In a paused state you can keep editing but it will not run. Note that changes made mid-run only apply from the next run onward.
Agents inherit the quality of the account they run in. An agent asked to triage requests on a board where half the statuses are unused and ownership is inconsistent will make confident decisions on bad information, which is worse than making none. Before you build anything, check which column you are in.
We've had a great experience with Greg and Hayley at warbble·digital, who have been endlessly helpful and patient with all our team, thank you!
- Parker, H. · Jul 2025 · Customer Success Training
Start your trial through warbble·digital and you won't spend it guessing. As official monday.com partners, we'll point you to the right product and the right plan, so your trial proves value before you spend a cent.