Luis Paredes Franco
← Blog

The PM workflow just flipped

Luis Paredes Franco · October 15, 2026

A working prototype used to come late, after the PRD and the design. With AI it can come first. Notes on Atlassian's State of Product 2027 report, and five things that are working for me.

Original post on LinkedIn ↗Atlassian State of Product 2027 ↗

I just read Atlassian's State of Product 2027 report, which surveyed 1,000 product professionals. One section confirmed something I have been seeing in my own work.

1What the report says

  • 54% of teams now use AI to produce working prototypes or specs engineers can build from.
  • On 19% of teams, PMs submit near-complete specs or even code for Engineering review.
  • 52% say prototyping is no longer limited to design or engineering.
  • 67% agree it is getting hard to define their role as the lines between Product, Engineering and Design blur.

2The flow is inverting

For years, a working prototype came late in the process, after the concept, the PRD and the design were defined. Teams could test mockups along the way, but real feedback on a functional product took a long time to arrive. With AI, that has changed: a working prototype can now be one of the first steps, not one of the last.

We are no longer going only from PRD to mockup. More and more, we go from mockup to PRD. You build something tangible, put it in front of users, and document what you learned.

3Faster building is not faster deciding

The same report found that 86% believe deciding what to build matters more than ever, and 82% of teams deal with low-effort AI work that someone else has to clean up.

4What is working for me

Be obsessive about scope and the MVP

With AI, adding "just one more feature" costs almost nothing, so it is easy to get distracted. Many of those features will not matter to the client or will not generate enough value. Be clear on the problem and delimit the MVP before you start prompting.

Use the prototype to validate the problem, not just the solution

A prototype is the fastest discovery tool we have ever had. Test it with real users before committing. Building the wrong thing faster is still building the wrong thing.

Follow the rules of your platform

Connect AI to your codebase, for example through GitHub, so it understands your architecture, rules and styles. That way your prototype reflects how your product actually works, and you get a much clearer sense of what is realistic to build.

The PRD still matters, just at a different moment

Writing it after the prototype is fine, but it should capture why, not just what: the problem, the success metrics and what is out of scope. Skills and agents help here. For example, one that reads a mockup and breaks it into epics and user stories in Jira keeps documentation up to speed with building.

Embrace iteration

AI prompting is iterative by nature. Your first output is a starting point, not the answer.

5The role is expanding

The PM role is not disappearing. It is expanding. The value is shifting from writing the spec to deciding what deserves to be built, and that is harder, not easier.

How has your workflow changed? Are you still going from PRD to prototype, or have you flipped it too?