July monday.com Product Updates

monday.com's July 2026 updates: agents are now becoming more like your colleagues

The biggest release I've seen from monday in a while, and you’ll see what I’m talking about in the title

I've spent the last few weeks building agents on monday.com for clients, so I came to July's release with a fairly specific set of questions. Spoilers: it answered most of them.

This was a big month of releases. Agents are clearly the headline and not as a single update but as a cluster of them all pushing the same way. Workflows come in at second, with a very useful set of improvements. Thirdly (and the one I think gets missed) is how much monday has expanded its external connections: MCP, bring-your-own-agent, Slack, the cloud AI connectors and the option to buy more API capacity.

Put those three together and it shows that monday is opening its borders. Agents from outside can come in and work as governed members of your account. monday itself can now be driven from outside, whether that's Slack, Gemini, Copilot, or Claude and Cursor over MCP. You now no longer have to be inside monday to work with monday.

That matters, because the "you have to be in the tool for the tool to be useful" ideology is the one of the biggest adoption problems every work platform has. It's the reason good implementations still fail. Seeing monday tackle it head on is very encouraging.

Here's the breakdown, roughly in the order I'd rank it.

Agents: from tool to teammate

monday shipped out a group of agent updates this month. Now they’re going with more so the approach of: an agent isn't meant to be another feature you configure, it's meant to be a member of the team. Having built a few now, I believe that framing holds up.

They work from the moment you create them

Agents are active as soon as they're created. There's no setup step and no manual saving. They can also be shared so anyone in the account can use one. There's also a template library of more than 30 pre-built agents, each opening with a starting prompt that's ready to run.

This template library is a new initial introduction into agents. The hard part of getting an agent working isn't the technology, but instead it’s having a blank page in front of you. Opening something that already does roughly what you want and adjusting it is a very different experience from describing an agent from scratch.

Agents have also gained multi job management, so one agent can handle several workflows with a separate trigger for each. They can produce real files now too, HTML among them, which you can open, edit and build on instead of getting text back in a chat window.

What has impressed me most in practice is how well they navigate boards. That may sound dull, but an agent that understands your board structure, relations and column types well enough to act on them reliably is impressive and very helpful.

Agents now have an identity…

Agents get a profile card. Anyone in the account can see what an agent is for, who owns it, what tools it uses and how to work with it, without exposing sensitive configuration. Admins get cost and activity visibility through the Agent Directory. Creators can preview how their agent looks to everyone else before it goes live.

This is the update I'd point to if someone asked what actually changed this month.

Once an agent has an identity, people start treating it as a worker rather than an automation. You refer to it by a name. You assign it things. You talk to it. The tangible result is that you can add capacity to your business without hiring for it.

Think about the administrative functions in your organisation that already live partly on monday: like a payroll process, an HR request queue, a legal review step, a finance approval. Each one has repetitive work that needs doing, needs to be discussed and needs someone accountable for it. You can now set up an agent for that function, add it to items on a board, have it prepare data the way you need it and deal with it much as you would a colleague.

The profile card is what makes that feel real and enjoyable to use - a small change but with a big effect on how teams relate to it.

Governance arrived with it

Two updates matter here and they're the reason why I'd let an agent near a client process at all.

Human-in-the-loop approval for sending Gmail. You can set an agent to always pause for approval before it sends an email, using a simple ‘always ask’ or ‘don't ask’ toggle. The agent drafts, shows you the draft in the chat, before then waiting for you to approve or reject it.

The biggest worry with any autonomous system (and this isn't specific to monday) is that it might contact someone on your behalf without you knowing. I don't want a bot emailing my clients. I don't want it emailing anyone before I've seen what it plans to say. Being able to gate that one action is what moves email automation from intriguing but not helpful, to something I'd actually deploy. monday has said toggles for other sensitive actions, such as posting in Slack, are coming. A very strong step in the right direction.

Agent activity log. Admins and agent creators can now see a permission-controlled log of every agent run, which covers the trigger, the outcome, credits used and step by step details. Admins see usage and cost across all agents with sensitive detail hidden. Creators get full visibility of their own agents, so they can debug a failed run and jump straight to the source.

Anyone who has worked with agents will know why this matters. You need to see what happened over the last week or month: what ran, what worked, what didn't and (more importantly perhaps) what it cost. Without this you're trusting a black box and nobody should scale a black box across their account.

Bring Your Own Agent

You can now connect external agents such as Claude or Cursor into monday workspaces, where they work alongside your team in the context of the work.

The design is well thought through. An invited agent becomes a governed entity with its own identity, separate from the person who brought it in. It can be chatted with, mentioned on items, assigned tasks, triggered and monitored - the same as a monday agent or a person. You give it monday-specific instructions covering how it behaves inside the platform, while its actual brain, meaning its instructions, skills and knowledge, stays on the external platform where you built it.

This last detail is an example of how this is practical: If you've already invested in an agent elsewhere, or you want to try a model monday doesn't offer natively, you don't have to rebuild it. You connect it, govern it through monday, and keep the intelligence where it lives.

If you have IT or compliance stakeholders, the governance side is the part to show them: separate identity, granular permissions, its own action log as well as dedicated cost controls.

Agents in the CRM

monday has introduced a set of purpose-built agents for CRM and Campaigns. Three stand out.

A pipeline monitoring agent. It produces a structured insights report on pipeline health and coverage for sales managers, flagging risks, recommending next best actions, and drafting messages to send to reps.

This is one I had started trying to build myself, so I'm keen to see it working. Pipeline health reports have existed for years and mostly get skimmed. The value is having something that engages with you about it: staying on top of deal flow, noticing what hasn't been updated, checking whether follow-ups happened, and filling in the blanks instead of just pointing at the gap. Reps get a targeted prompt about their specific deal rather than another generic CRM reminder, and that's usually the difference between a nudge that works and one that gets ignored.

A meeting prep agent. It takes the full context of the account, contacts and opportunity history and produces a briefing document. It surfaces company news, generates discovery questions and talking points, and works out whether the call is a discovery conversation or a renewal so the brief suits it.

If it works well with notetaker data, this could take meeting prep from a twenty minute job to something that's already done when you sit down. How context-aware it is in practice is the thing to watch, and I'll be testing that.

A campaign intelligence agent. For monday Campaigns users, this one researches your market, competitors and industry trends, produces campaign insights, and pulls CRM insights out of real sales calls.

CRM also gets a customisable homepage, with widgets for pipeline, upcoming meetings, contacts, accounts, recently visited boards and agents. Campaigns and CRM can now work from one shared contacts board with real time updates, instead of two lists that slowly drift apart. That second one fixes a very common problem: marketing targeting prospects on stale status data while sales can't see campaign engagement.

Webinar: AI in CRM with Agents

We're running a session on this exact topic. How agents change the way a CRM gets used day to day, what's worth automating, where a human needs to stay in the loop as well as what we've learned building agents for clients.

[DATE] · [TIME] · [DURATION]

[REGISTRATION LINK]

If your CRM data is only as good as the last person who remembered to update it, come to this one.

Workflows: The human-in-the-loop block stands out as the premier feature of this release

Agents got the headlines. Workflows got the update I'm most pleased about.

A human-in-the-loop block. You can add an approval step into a workflow. It waits for a user response, currently by email, and you decide how the workflow behaves based on what comes back.

This excites me more than the agent email approval, for a specific reason. Until now, handling human input in a monday workflow meant a well-known workaround: build a workflow that generates something and sets a status, then wait for the user to change that status, which triggers the next part. It works, but every human decision point needs its own status column and some process discipline to go with it. It also makes workflows fragile in a way that's hard to explain to a client.

Now you just say that human input is needed at a certain point. The workflow pauses, the person responds, and then it carries on. That removes a whole layer of extra setup and it means we can automate admin processes that previously weren't worth the complexity. It will change how we design workflows at The SaaS JEDI.

Agents in workflows. You can add an existing agent, or create a new one, from inside the workflow builder using chat. It reads board context, workspace data and earlier workflow stages to configure the agent for you rather than making you look things up.

This is where agents stop being standalone helpers. An agent you built for one job can now fire as a step in a bigger sequence, as part of a proper multi-stage process instead of something a person has to remember to ask for.

Automation suggestions through Sidekick and automation building through MCP. Sidekick can surface automation suggestions based on your boards and how your team works. Separately, Sidekick and external agents can build and manage monday automations outright using a new MCP API.

I think the suggestions piece will do more than it looks like it will. Automations are badly underused. Most accounts I look at are running a fraction of what they could… and it's rarely because the team doesn't want them. It's because you don't know what's possible until someone shows you. Having the platform watch your patterns and tell you what you could be doing shortcuts a learning curve that normally takes months, or a consultant.

Run on demand. You can trigger a workflow manually whenever you want, either before publishing to test it or afterwards. The pre-publish testing alone justifies this. Being able to validate a workflow with real agent steps before it touches live data takes out a real risk.

Loops. A new block that iterates over a collection, whether board items, subitems or connected records, and runs a set of steps for each one, reading monday data sources as it goes.

Run on demand and loops work well together. Previously, to sweep a board for items missing a connection, we'd add a column, change a status, and let an automation match things up one at a time. Now you loop through everything and have it done, and you can run it whenever you want without the extra status column. Much cleaner for that sort of work.

Full workflow configuration by chat. You can plan, configure, validate and publish complex multi-step AI workflows by describing the goal in the Sidekick chat pane, with MCP support so developers can do the same thing programmatically from external tools. That's where the world is going, and it's very good to see monday there.

Opening the borders: external connections

This is the third theme and it is one I highly encourage you not to skip past.

MCP block in AI workflows. You can add a block that runs CRUD actions against third-party applications using an API key, with any publicly available MCP server.

This is the most interesting item in the release and I want to be careful about it. The obvious question is what it means for tools like Make and Zapier, which we've used for years to connect monday to everything else. My read is that this isn't a straight replacement. It looks like a different use case rather than the same job done natively. It is early and I'd rather watch how it behaves in real builds than call it now. But if it matures, it changes where and when we reach for a middleware layer which is worth watching.

Slack integration. You can mention the monday.com app in any Slack channel, chat or message to trigger the agent. It reads the thread context, retrieves information or takes actions in your account and then posts back into the same thread.

We've seen the same pattern work well with Claude in Slack, so it's good to see monday there too. Being able to pull out insight and push updates to boards from where the conversation is already happening removes a real friction point. Most teams discuss the work in Slack and then someone has to go and reflect it in monday, or nobody does.

Google Gemini and Microsoft Copilot connectors. Gemini Enterprise and Copilot users can manage monday workflows in plain language without leaving their own ecosystem, building tracking boards, summarising updates, redistributing tasks and cross-referencing data across workspaces from the chat interface they already use.

If your organisation has standardised on the Google or Microsoft AI stack, this is significant news. It also shows the borders point nicely: monday is happy to be operated from somewhere else.

API calls add-on. Accounts can buy additional daily API call capacity beyond their plan limit, on a self-service basis.

This one is close to the work we do. The concern with any serious integration build is what happens as you approach the ceiling. You end up engineering around it, building thresholds, retries and sleep intervals into your code purely to stay under a limit. Now there's a choice: pay for more headroom, or write efficient code and stay where you are. Having the option is the point and it means an integration doesn't have to force a plan change.

Access to LLM models through your monday subscription. There's now a single API that lets developers and agents reach leading models from OpenAI and Anthropic through the monday subscription itself, with monday routing the requests and reconciling the usage against your workspace. No separate LLM vendor agreement needed.

Worth flagging if you've been through an AI procurement process. Using monday's existing contract, security approvals and billing instead of negotiating with each model provider takes out a lot of friction, and so IT can set budget caps and rate limits across models in one place.

Portfolio-level resource utilisation

(Enterprise plan only)

Portfolio and resource managers can now see planned versus actual allocation per person across every project in a portfolio, in one report. It shows total availability, planned effort and actual effort, aggregated with a per-project breakdown, with colour coded flags for over and under allocation, plus search, filter and direct navigation into any project's resource planner.

I'm pleased to see this. Portfolio is one of monday's more underrated capabilities and watching resource utilisation and capacity management get layered into it steadily is one of the more interesting things happening on the platform.

Capacity management is hard. It's one of the areas where organisations most often give up and go back to a spreadsheet, because the real picture spans projects, involves people who are only partly allocated and changes every week. A platform-native answer that's actively improving, rather than a static report someone has to maintain, is worth paying attention to. Planned versus actual, in one view, across a portfolio, is what most delivery businesses have been building by hand for years.

Vibe, briefly

Vibe keeps moving at a pace that's hard to keep up with, but don’t worry we’ve got it all covered here. This month you can duplicate any Vibe app in one click, choosing whether to keep the existing board connections or generate independent boards. Agents can create, modify and remove Vibe apps and manage publication. You can build board view and dashboard widgets with Vibe. You can click on a visual element, such as a nav bar, a card or a button, to reference it precisely in a prompt, which saves a lot of going back and forth and a fair number of AI credits. You can disconnect boards from an existing app without rebuilding it. Vibe can read historical board data to show trends over time. And admins on Enterprise can restrict which roles are allowed to create or modify apps, separately from who can publish them.

My favourite is the Vibe community, a browsable gallery of ready made apps built by other people, which you can customise and launch instead of starting from nothing. It's the same idea as the agent template library and it works for the same reason. The blank page is the bottleneck. Builders who contribute get visibility across the ecosystem, which should keep the gallery worth looking at.

Also shipped with this release

A few more in brief. AI Portal Builder in monday Service lets you describe a portal in plain language and get a structured, branded portal back, then refine it through chat with the changes showing live. Historical trends in Sidekick means you can ask how your data has changed over time rather than only seeing today's position. You can submit a monday form through a conversation with Sidekick or an external agent, which fills in what it already knows and prompts you for the rest. Multiple forms can now edit the same item, with each person getting their own form view showing only their relevant fields, so one item can be built up by different people at different stages without duplicate submissions. And there's a personal AI credits widget in the avatar menu showing your own usage, what's left and a breakdown by capability.

What I'd do about this month

Three things.

Build one agent properly. Pick one administrative function that already lives on monday, start from a template, and give it an identity and an owner. Two weeks of actually using one will teach you more than a quarter of planning it.

Switch on the activity log and read it before you scale. The governance features shipped alongside the agent features for a reason. Look at what your agents did and what they cost before you add more.

Revisit the workflows you decided weren't worth automating. If you skipped automating a process because it needed a human decision in the middle and the status-column workaround was too fragile, that objection has gone.

The general advice hasn't changed either. Don't try to adopt thirty-odd updates at once. Pick the one or two that match a problem you already have.

How The SaaS JEDI helps

The SaaS JEDI is a monday.com implementation partner working with clients across the UK and South Africa. We've been building agents on the platform for clients over the past few months, so on this release we can help you:

• Design, build and govern monday agents, covering identity, ownership, approval gating and activity monitoring

• Connect external agents into monday under proper governance, with the right permissions and cost controls

• Redesign workflows to use the human-in-the-loop block, loops and on-demand runs instead of status-column workarounds

• Set up CRM agents, and get your pipeline data into a state where they're actually useful

• Build integrations using the MCP block, and advise on where it does and doesn't replace existing middleware

• Configure portfolio-level resource and capacity reporting

• Advise on AI credit governance, budgets and the new model access route

• Train your teams so any of this gets used rather than just announced

If you'd like to work out which of July's updates are worth prioritising for your account, get in touch, or come along to the webinar.

Stephen MacLennan Founder, The SaaS JEDI