monday.com Counted 75M Agent Requests. The Board Needs One Outcome Number.

monday.com reported 75M MCP requests and 6M agent interactions. Activity isn't the outcome. The board needs one outcome number, and the CPO defines it.

Falk Gottlob · · 8 min read

Leadership-amber Falkster cover: a big car dashboard with every gauge needle spinning blurred, except one large clean dial in the center whose needle points steadily at a single mark.
8/9
SeriesThe CPO Transition

Elizabeth Elliott of Morgan Stanley asked the question on monday.com's earnings call this week that every CPO will be asked next year.

"What are some of the leading indicators that should improve first if this transition is working?"

monday.com is in the middle of a big move. Roy Mann said on the call that the company is "moving from managing work to executing it," and it cut about 20% of its workforce during the quarter to get there. The numbers on the table were real. Eran Zinman said customers had made "over 75 million MCP requests," outside agents doing work inside monday.com, with no monetization plan beyond usage and fees. monday agents "surpassed 6 million agent interactions in less than 2 quarters." AI ARR reached $36 million, led by credits, and CFO Eliran Glazer was candid about it: "it's not incremental yet. It's part of our overall ARR."

Then Mann said the line I'd frame. Customers "care a lot about measuring the output they get out of credit."

That's the bar. Not the requests, not the interactions. The output.

This is part 8 of The CPO Transition, part of The Product OS. Part 6 argued that product, design, product marketing, and the field report to the CPO because one loop needs one owner, and part 7 took the customer end of that loop. This part is about the number that owner answers for.

The short version

  • AI-first companies are reporting activity, and the board is about to ask for the outcome. monday.com counted 75 million outside-agent requests and 6 million interactions with its own agents this quarter, and its co-founder said customers care about the output they get.
  • Activity counts can't tell a board whether the transition is working, because an interaction can end with the job done or abandoned, and because most use now runs through agents the company didn't build.
  • The number that holds is verified jobs done: customer jobs completed to a written definition of done, on every path, with the cost per job beside it.
  • The definition is a product decision, as the three different ways Intercom, Zendesk, and Salesforce count an AI resolution show.
  • So the CPO writes it, reports against it, and each of the four seats owns one input to it and stops reporting the metrics that don't move it.

The outside agents already outnumber yours

Look at the ratio before the totals. 75 million requests from outside agents. 6 million interactions with monday's own agents. More than twelve to one.

75 million

MCP requests from outside agents at monday.com, against 6 million interactions with its own agents, by the company's Q3 2026 call.

Mann said it plainly: "We see increase in MCP usage in discovery through the LLM." The customer's assistant is becoming the front door, and a growing share of the work happens through an interface the company doesn't design and can barely see. That changes what a product org can measure. Page views, sessions, and agent interactions all assume you own the surface. You don't anymore.

What you still own is the job. A customer asked for a project plan, a ticket closed, a report filed. Either it got done or it didn't, whichever agent asked. So the scorecard has to be built on the job, counted the same way on every path: your interface, your agent, someone else's agent through MCP, or a person on your team.

I think this is the biggest measurement change in software since the move to the cloud, and most dashboards haven't noticed yet.

Done needs a definition, and the definition is the product

Here's where it stops being a reporting question.

Rajesh Beri compared how three vendors bill an AI "resolution," and they mean three different things. Intercom's Fin bills confirmed resolutions and "assumed resolutions," where the customer "exits the conversation without requesting further assistance." Zendesk, since May 18, bills only verified resolutions, where an LLM check confirms the issue was solved. Salesforce bills its Help Agent when an issue is resolved "autonomously, start to finish." Same word. Three products.

Each definition tells the team what to optimize. Count silence as success and the product learns to end conversations. Count verified fixes and it learns to solve them. That sentence isn't finance's and it isn't sales's. It's the most important spec in the company, because every agent, eval, and roadmap call bends toward it.

So the CPO writes it. One sentence per core job, what done means, who confirms it, and what doesn't count. Publish it inside, put it in the eval, and put it in the contract, since buyers already ask for it. Then report one number against it every month: verified jobs done, with the cost per job beside it, which is the margin work from part 5 brought to the board.

Four seats, four inputs, one number

This is where the one-function case gets concrete. Under one outcome number, each seat stops reporting its own success and reports one input to the shared one.

Product reports the bar and the decision queue against it, the shift The PM Transition lays out. Design reports the repair rate, how often a person has to fix what the product produced, the number The Design Transition argues design should own. Product marketing reports story lag, the days between a change shipping and the assistants describing it correctly, from The Product Marketing Transition. The field reports request to delivery, Atlassian's clock from part 7, beside the reversal rate The FDE Transition puts on every engagement.

And each stops reporting the metrics that don't move the number. Features shipped. Screens produced. Launches run. Engagements booked. Those were fine proxies when building was the hard part. When building is cheap, they mostly measure motion.

Split the four seats across four executives and you get four dashboards, each green, with no one accountable for whether the job got done. That's the practical reason the seats sit under the CPO: the inputs only add up when one person owns the sum.

Verint made a related move this week. As Mara Sugue reports at IT Brief, it named Abhi Sharma chief product and technology officer, and CEO Dave Rhodes said "Abhi will lead product and technology as one organisation," framing it around customers "getting real business outcomes now from AI." Putting spec and build together is a good step. The outcome number asks for one more: the story and the adoption have to answer to it too.

What I expect

These are my predictions, not facts.

My prediction: by the end of 2027, some AI-first public software companies report a completed-outcome count beside AI ARR, because analysts keep asking Elliott's question until someone answers it with a job count.

My prediction: within 18 months, the definition of a billable outcome becomes a standard contract exhibit that buyers negotiate line by line, the way uptime became one.

My prediction: by 2028, CPO pay at AI-native companies is tied to one outcome number with a cost-per-job floor, not to roadmap delivery.

My prediction: outside-agent usage passing in-product agent usage, as it already has at monday.com, becomes the norm at work-software companies within two years.

Part 6 predicted that board CPO searches will ask for the outcome number. This is what the number is. Next in the series: the story is part of the product, and what the CPO decides about positioning that agents read.

One thing to try this week

Pick the job your customers hire your product for most often. Write the definition of done in one sentence: what counts, who confirms it, and what doesn't count. Then count last month's completions on every path, your interface, your agents, outside agents through your API or MCP, and your own people. If you can't count the outside-agent path at all, that's your first fix, and it's the first slide of next quarter's board deck.

Sources: monday.com, Q3 2026 earnings call, Roy Mann, Eran Zinman, Eliran Glazer, and Elizabeth Elliott of Morgan Stanley, transcript via The Motley Fool, November 9, 2026, quotes as transcribed. Rajesh Beri, "Your Fin Bill Can Include Customers Who Leave Without Replying," beri.net, November 7, 2026, vendor definitions as he reports them. Mara Sugue, "Verint names Abhi Sharma as chief product & tech chief," IT Brief, November 11, 2026, with Dave Rhodes.

Part of the running argument on Product Leadership: the product org chart was calibrated to a build cost that collapsed, and in the AI era the whole product outcome, from customer signal to adoption, belongs to one function under the CPO.

Related answer: What should a CPO report instead of agent interactions?

Share this post

Frequently asked

What did monday.com report about agents on its Q3 2026 call?+

On the November 9, 2026 call, Eran Zinman said customers had made over 75 million MCP requests and that monday agents passed 6 million interactions in under two quarters. AI ARR reached $36 million, led by AI credits, and CFO Eliran Glazer said it's not incremental yet and is part of overall ARR. Roy Mann said customers care a lot about measuring the output they get out of credit. Quotes are from the Motley Fool transcript.

Why aren't agent interactions a good board metric?+

Because they count activity, not results. An interaction can end with the job done, half done, or abandoned, and the count looks the same. And once outside agents like ChatGPT or Claude work inside your product, as they did 75 million times at monday.com, more of the use runs through interfaces you don't design. The number that holds across every path is whether the customer's job got done.

What outcome number should the board hold the CPO to?+

Falk's answer is verified jobs done: the count of customer jobs completed to a written definition of done, on every path, your interface, your agent, someone else's agent, or a person, reported with the cost per job beside it. One number, one definition, owned by the CPO.

Why does the CPO own the definition of done?+

Because the definition is a product decision. Rajesh Beri's comparison shows three vendors defining an AI resolution three ways: Intercom counts customers who leave without asking for more help, Zendesk counts only resolutions an LLM check verifies, and Salesforce counts issues resolved autonomously, start to finish. Whoever writes that sentence decides what the product is trying to do.

What do product, design, product marketing, and FDE report under one outcome number?+

Each seat reports one input that moves the number. Product reports the bar and its decision queue. Design reports the repair rate. Product marketing reports story lag. The field reports request to delivery. Each stops reporting the metrics that don't move the number, such as features shipped, screens produced, launches run, or engagements booked.

What does Falk predict about outcome metrics?+

His predictions, labeled as such: by the end of 2027, some AI-first public software companies report a completed-outcome count beside AI ARR; within 18 months, the definition of a billable outcome becomes a standard contract exhibit buyers negotiate; by 2028, CPO pay at AI-native companies is tied to one outcome number; and outside-agent usage passing in-product agent usage becomes the norm, not the exception.

LeadershipCPOchief product officerproduct leadershipCPO TransitionProduct OSboard metricsoutcome metrics
Helpful?

Keep reading

The short answer

The Claim FilePositioning is one versioned file in the repo that every agent, page, and assistant profile loads. Sort every change by the claim, and track story lag.Read the chapter

Part of

Product Leadership

About the author

Falk Gottlob

Falk Gottlob

Product Executive · Founder, Falkster.AI

Thirty years shipping product, from Microsoft Research and Adobe to Salesforce, where he grew Quip into what became Slack Canvas. Four startup exits, including a $6.5B healthcare merger and a company Microsoft bought. Four-time Chief Product Officer. Now founder of Falkster.AI, building Heidi, the company brain for AI-native businesses. This notebook is written from inside the build, not above it.

Sign in with LinkedIn to leave a comment.

Sign in with LinkedIn

Also on Medium

Full archive →