How Monday.com Builds AI Products That Teams Actually Use

Monday.com is one of the world’s largest productivity suites, used by teams at orgs like McDonald’s, Coca-Cola, Canva, and 60% of the Fortune 500. In the last year, they’ve gone all-in on AI, shipping three major new products, including Sidekick, their new assistant that helps teams go beyond managing work to actually getting work done. 

We spoke to Or Fridman, AI Group PM at Monday.com, to learn how his team builds with AI.

Here are the highlights!

Co-create with users (while avoiding scope creep)

When Or’s team set out to add AI features to Monday.com, they ran iterative experiments that combined two approaches:

• Nail one job. Build a narrow feature that solves a concrete problem really well (like updating items on a board).
• Leave room for discovery. Give users a blank canvas to explore new use cases, then fold the most valuable behaviors back into the product.

The result was a product that has evolved with user needs, based on real usage - instead of adding AI for the sake of AI.  

Building AI products is different

Traditional features are deterministic: inputs and outputs are defined in advance, and every user action maps to a predictable result. In contrast, AI features are non-deterministic. The same prompt can return different outputs, which means PMs have to design for variability.

At Monday.com, that forced the team to rethink the PRD. Instead of a user story that says “when the user clicks X, show Y,” they had to spell out questions like:

• What does “good” look like? If a user asks about sprint performance, should the assistant return a paragraph summary, a bulleted list, or a chart of metrics? Each choice implies different UX affordances and evaluation criteria.
• What controls keep users in the loop? Do they get a regenerate button, the ability to edit outputs inline, or an explicit undo path if the system makes a change?
• How to establish trust? Should results be paired with citations back to the source data? When should the model hedge or ask for clarification instead of answering confidently?

The basics of product still apply but the scope of the spec is larger. AI PRDs look less like feature tickets and more like data-product specifications: defining the model’s inputs, the expected shape of outputs, the evaluation metrics, and the safety mechanisms around them.

PMs are no longer just shipping features, they’re shaping probabilistic systems.

Compressing feedback loops

AI gives product teams a way to significantly speed up product feedback loops. At Monday.com, PMs use AI-assisted tools to brainstorm, prototype, and even “vibe-code” mockups. Instead of the old linear flow - PM writes a PRD, designer mocks it up in Figma, engineer codes - those steps all blur together. A PM can draft a rough flow in minutes, vibe code a working prototype, test it with users, and refine it alongside design and engineering.

In a world where model capabilities evolve every few weeks, that speed is essential. 

What skills do they look for in PMs? 

The fundamentals of a good PM haven’t changed: be user-focused, strategic, collaborative. But Or adds one more requirement: adaptability. AI evolves weekly, not yearly. The best PMs are the ones who can keep pace, experiment, and fold new capabilities into the product without losing sight of user value.

We asked him if companies should hire PMs who specialize in AI, and Or’s answer was clear: better to hire strong product thinkers who can learn AI than hire for a narrow set of tools that may change in six months.

Looking five years out

Or expects Monday.com to evolve from being the place where teams manage work to the place where they also execute it. He also sees interaction modalities shifting: chat, questions, voice, and images will replace clicks and menus as the primary interface for work.

For a marketing manager, that could mean not just planning campaigns inside Monday, but generating assets and running them end-to-end without leaving the platform.

Actionable takeaways for builders

1. Co-create with users (without losing focus). Involve users early to discover how they want to use AI, but anchor the product around a single, well-defined workflow.
2. Account for non-determinism in PRDs. Define inputs, expected output shapes, trust mechanisms, and evaluation criteria upfront.
3. Compress feedback loops. Use AI internally to turn ideas into prototypes quickly so users can guide what’s worth building.
4. Hire for adaptability. Prioritize PMs who can learn fast over PMs who happen to know the current thing. 
5. Design for new interfaces. Natural language, chat, voice, and images are becoming the default ways users interact with software.

We had such a great time jamming with Or! We’ve been having similar convos with product and engineering leaders at top orgs (like Intercom, Honeycomb, and Vercel) to unpack how they actually build with AI. You can find previous episodes on YouTube.