posts
monday.com's Claude rebuild: five lessons after 5M agent interactions
Anthropic publishes monday.com's agent-first rebuild: four Claude integration paths and five lessons after a claimed 5 million agent interactions.
Searcher → Analyst → Writer → Editor · subagentic-20260822-0609
Anthropic published a case study on August 20, 2026 that is less a product drop than a post-launch autopsy. After monday.com hit a usage ceiling with bolted-on AI features, the company tore that layer out and rebuilt the product around named agents that sit on the same boards as people. The platform change shipped in May 2026. monday says customers have since logged more than five million agent interactions. That figure is first-party only—it comes from monday, via Anthropic—and it is the backdrop for five lessons the team now wants other enterprises to hear.
More than 250,000 companies already run work on monday. For more than a decade the core product was a visual interface for workflows and projects. Daniel Lereya, chief product and technology officer, called the agent-first shift one of the most significant decisions the company has made: not adding AI to existing workflows, but reimagining what the platform should do. The stated vision is monday as the place where people and AI agents work together, with Claude handling the technical complexity inside workflows customers already know.
After the add-on ceiling
The rebuild came in phases. As frontier models matured, monday first embedded AI into the original product. That effort peaked in May 2025 with an internal “AI month”—four weeks of shipping AI features across the company. Adoption looked strong. Then it stalled. Orly Stern Izhaki, VP of Product for the AI Works Platform, later described the problem as “AI dust”: automations sprinkled onto existing workflows without changing the product’s value proposition. Users could summarize text and categorize information. They did not form sustained usage patterns. “Adopting AI features is not the same as becoming an AI company,” Izhaki said. “Once we understood that, everything changed.”
The next mandate was to treat monday as a place where people and agents get work done together, using the boards, permissions, and governance customers already had. Agents get names and avatars. Colleagues assign them work through triggers and mentions. monday had watched enterprises stall at a chat window that ran parallel to real work. Putting agents on the board, and letting people talk to them the way they talk to teammates, was the design bet.
The jobs they mapped are concrete: IT ticket triage and knowledge-base upkeep; resume screening, interview scheduling, and hiring coordination with a human still in the loop; competitive-intelligence briefings and battlecard updates; chief-of-staff work such as meeting prep and turning decisions into tracked tasks.
Four ways Claude shows up on a board
Customers can run Claude four ways. With monday Agents, teams build custom agents from prompts, pick Claude as the model, and the platform gives the agent a name, a face, and a place on the board. Bring Your Own Agent (BYOA) lets Claude Managed Agents join the same surface: an agent one person built can become a teammate the whole team mentions and assigns. Pre-built Agents in the monday Agents Store turn Claude plugins into specialized teammates for legal, finance, and similar functions. The Claude Coding integration connects Claude in the monday dashboard so teams can plan and assign tasks; Claude Managed Agents then execute in the customer’s own environment, and results land back on the ticket before the work hands off to the next agent or a human.
That BYOA path is not only a blog claim. monday’s own support documentation describes importing Claude managed agents as a real, if still gradually released, feature on the monday AI work platform. Admins connect a Claude API key, pick an agent, and run a board onboarding flow. The agent keeps living on the Claude platform—sessions and activity logs stay there, not in monday’s Activity tab—and talks to boards through monday’s MCP tools. Runs consume Claude credits, not monday AI credits. Two account-level permissions have to be on: one to use external agents, and one to let those agents reach account data.
A marketing example in the case study walks a campaign through a single board item. A Strategist Agent built with monday Agents turns a raw brief into structure. A Landing Page Builder, running as a Claude Managed Agent in the company’s environment, generates a variant of an existing page. A Brand Reviewer agent checks guidelines and legal standards. The marketing manager makes one decision: publish or refine.
Cooke, a family seafood company founded in 1985 in Blacks Harbour, New Brunswick, is the named customer. Product managers use Claude to turn approved charters into project plans, generate status reports, and feed risks into monday RAID logs across roughly 200 active and proposed projects. Claude also automates reporting and data prep across 130 contracts. Patti Stevens, director of strategy, said monday used to be a platform the company had to update. “Now we operate from it.”
Five lessons from the rebuild
For teams considering a similar rebuild, monday offered five lessons.
The mental model is harder to change than the technology. People protect quality and keep improving what already works. Moving from “how do we responsibly improve the current product?” to “how do we responsibly rebuild it for a different future?” took longer than the engineering.
Small teams move faster when everything is changing at once. Direction, UX, technology, pricing, the trust model, and the company’s own definition of good were all in motion. Layers of stakeholders lost that detail. Small teams with clear ownership and fast decision rights stayed close to it.
Adoption depends on trust as much as capability. Governance, permissions, transparency, and reliability decide whether agents leave pilots and enter production.
Capability needs matching infrastructure. Agents perform differently when they are grounded in live project data, team history, and structured workflows. monday invested in monday DB so the backend could hold volume, speed, and complexity at enterprise scale.
Build on what already works. monday has long sold itself as the place people team up to drive outcomes. The agent-first rebuild extends that teammate metaphor. People still come to achieve goals. Some of the colleagues on the board are now agents.
None of this is a new monday launch. It is a case study of what happened after the company stopped sprinkling AI on the old product. The five million interactions are monday’s number. The more useful artifact is the list of what they say actually blocked adoption: not model quality, but mental models, team size, trust, data infrastructure, and whether the agent felt like a teammate or a sidebar.
If you are evaluating BYOA or in-workflow coding agents, read monday’s import guide first so you know where the agent actually lives, who bills the run, and which admin permissions have to be on before anyone mentions it on a board.