The Role of Product in the Age of AI: Best Practices from the Field

Greg Biggers has been a Chief Product/Technology Officer three times, led a product team through a $1.5B acquisition, and spent the last decade coaching PMs on the General Assembly product faculty. Eran sat him down for an hour to ask the question every product team is quietly stuck on right now: if AI can write the code, what's actually left for Product to do?

The bottleneck isn't speed anymore. It's judgment.

Webinar video

Follow us on LinkedIn to get invited to our monthly webinars →

Follow our YouTube channel for past webinar recordings →

TL;DR

• The real bottleneck isn't AI-assisted engineers outrunning PMs. It's finding and validating the right problem, and that work is now open to whoever's best at it, not just whoever holds the PM title.
• Static PRDs are dying. Teams are replacing them with short product briefs (market, problem, target outcome) and letting a prototype, or the pull request itself, carry the implementation detail.
• The tool stack has a clear leader: Claude Code, with Cursor layered on for teams that want an IDE, and Lovable or Replit as the on-ramp for people who don't have repo access yet.
• Tooling isn't the hard part. Organizational change is. Teams pulling ahead aren't waiting on a company-wide AI mandate, they're acting like they already have permission.
• Speed doesn't remove the need for review, it raises the price of skipping it. Human attention is what keeps AI-generated work from turning into slop.

Product, design, and engineering boundaries are blurring, not disappearing

Eran opened with the obvious provocation: if coding is this fast now, are product people the bottleneck? Greg's answer was immediate: no.

The bottleneck is discovering and validating the right problems to solve, and anyone can do that. — Greg Biggers

He's watched engineers, designers, and even non-technical people get good at that discovery work once they learn to use Claude or ChatGPT well. It isn't tied to a job title anymore. Eran added his own read from more than 70 product-leader interviews he's run this year: an "end-to-end" builder is emerging, someone who owns an idea from conception to shipping, leaning on AI to fill the design and engineering gaps (existing design system, existing code conventions) while doing the part AI still can't do for them: understanding the customer, setting strategy, deciding what to measure. Greg's parallel point on design specifically: a designer's differentiated value isn't wireframe fidelity anymore, AI handles that fine now, it's friction analysis and understanding how a new capability collides with everything already built.

Playbook move: stop assigning "problem discovery" to a job title. Let whoever's closest to the customer run the validation loop, and give designers the room AI just freed up to do deeper friction and systems work instead of pixel-pushing.

PMs are getting hands-on, and some are shipping the feature themselves

Eran described a split he's seeing across his interviews: some PMs "never leave Claude" and ship full features themselves, others are stressed trying to force old handoff habits into the new pace. He shared his own examples, a one-line prompt to add shareable deep links, a copy-format menu, each shipped to production in minutes. Greg mapped the current tool stack: Claude Code dominant, Cursor as an added layer for teams that have adopted it well, and for people without repo access yet, tools like Lovable or Replit to prototype on the sidelines before graduating into the real codebase, a path Eran said his own team deliberately nudges people through.

Playbook move: give PMs a sanctioned on-ramp, Lovable or Replit before repo access, Claude Code or Cursor once they've got it, so "hands-on" doesn't mean "unsupervised in production" on day one.

The PRD-to-engineering handoff is becoming continuous, not a document

Asked what happened to the PRD, Greg didn't hedge:

We're not on the same planet we were on ten years ago, when a PRD meant a static, nine-to-eighteen-month requirements document. — Greg Biggers

What's replaced it is a short product brief: the market served, the problem to solve, the target outcome, deliberately leaving the solution to the small team building it. Success gets judged by whether the problem got solved, not whether the build matched the doc. Eran laid out a taxonomy from his research of three handoff styles now in use: PMs coding the front end themselves and treating the pull request as the de facto spec; teams handing over a working prototype instead of a document; and a rich, interactive HTML "PRD" with embedded mini-prototypes for engineers who still want a directional artifact. Greg's caution:

A pull request isn't a direction-setting artifact, it's a destination thing. — Greg Biggers

Playbook move: replace the static PRD with a short brief. State the market, the problem, and the target outcome, and let the team's own prototype or PR carry the implementation detail, not a forty-page doc.

Adoption and change management are the hard part, not the tools

Eran asked whether product teams overall are doing well with AI. Greg's answer: only occasionally. Most of the visible LinkedIn success stories represent a small minority. He framed the real choice as this:

Do you wait for permission, or do you just seize the opportunity? Those of us waiting for permission have a long wait ahead of us. — Greg Biggers

Full company-wide transformations are rare. The real progress comes from small pods or individuals acting as though they already have permission, which creates a new job for CTOs and CPOs: spot which grassroots experiments are actually working and deliberately socialize them. Eran shared a concrete practice from one of his interviews, a weekly "AI Fridays" show-and-tell that upskills laggards through peer demonstration, plus a pull-versus-push insight: unsolicited daily AI reports get ignored, but a system people can ask a question of and get a great answer from gets used, which he tied directly to what Evermuse is built to do.

Playbook move: don't wait for a company-wide AI mandate. Seize the opportunity in your own corner of the org, and if you lead, spend your time finding and promoting the grassroots wins already happening, like a recurring internal demo session.

Preserving human accountability while moving at AI speed

Eran put the principle plainly:

Human attention is the cure for slop in everything. It's a scarce resource, so use it where it matters. — Eran Dror

In practice, that means spending a small, deliberate chunk of review time, an "explain this branch in plain language" step that takes five to ten minutes, rather than skipping review to move faster, and never handing a teammate unreviewed AI output to clean up. Greg agreed this is a new leadership skill: healthy skepticism of model output. His practice is to embed that skepticism directly into a CLAUDE.md-style instructions file, asking the model to state its estimated confidence or walk through its trade-offs before it answers. An audience question tied straight into this: has PM ethical responsibility expanded? Greg said yes, accountability ultimately reduces to whether the work has real value to the market, including a more holistic view of value than pure output. Eran added two concrete points: the duty not to drop unreviewed slop on a colleague, and, specific to building AI products, the responsibility to make an agent represent only what it actually knows rather than confidently answering off a thin slice of data. Evermuse's own approach is to cite everything and let users see exactly what the agent looked at.

Playbook move: bake self-skepticism into your prompts or your CLAUDE.md (ask for a confidence estimate and sources), and treat reviewing the plan, not just the output, as the one step you never skip.

Will there be fewer product leaders? (Audience Q&A)

One attendee asked whether the number of product leaders will shrink as AI takes over more of the build. Greg's answer: no, demand at the leadership level is rising. Sound judgment about which markets, problems, and economics actually matter gets more valuable once a single person can also execute on it. What he does expect to shrink is the number of junior, narrowly-scoped roles, which raises the bar for anyone starting out to build toward more senior, higher-ownership work. Eran agreed, framing it as a bar-raise rather than a bad thing: more leverage from AI, paired with a higher expectation of end-to-end ownership. Greg's caveat: in orgs where engineers still hold most of the build leverage, PMs shouldn't fight that structure, the move is to become a multiplier for engineering's output rather than trying to personally ship production code.

Playbook move: if you're early in a product career, stop optimizing for a narrow scope. Build the customer judgment and technical fluency that keep you valuable as headcount concentrates at the senior end.

Where Evermuse fits

Evermuse is built around the same principle Greg spent the hour defending: judgment is the scarce resource, not code. Evermuse is the context layer that keeps a product agent honest, every answer grounded in a real customer conversation, with the source you can go check yourself. That's the exact distinction Eran drew live: a model connected straight to raw sources will hand you a confident answer off a thin slice of data. Evermuse cites what it actually looked at instead.

If you want your product agents working from real customer signal instead of a guess, install the Evermuse MCP: evermuse.com/mcp.

Want to go deeper?

• The Evermuse MCP puts real customer signal, from sales calls, support chats, and product interviews, directly into the tools your team already builds with, every answer cited back to the actual conversation it came from: evermuse.com/mcp.
• Greg coaches founders and product leaders on exactly this shift through his practice, Continuous, including fractional CPO engagements. Find him at gregbiggers.com.

About the speakers

Greg Biggers is a product executive and coach who has led product teams through more than $3B in combined exits. He is currently VP of Product & Design at Airship and has spent over a decade on the product management faculty at General Assembly, mentoring hundreds of PMs. Earlier in his career he led the product team at Responsys through its $1.5B acquisition by Oracle, and held product roles at Apple, Vantive, and other consumer and enterprise companies. A three-time Chief Product/Technology Officer, he has led and mentored 400+ product and engineering professionals through the shift to outcomes-based, AI-native product operating models. Through his practice, Continuous, he now coaches founders and product leaders and takes on fractional CPO engagements.

Eran Dror is the Co-Founder & CEO of Evermuse, an AI product intelligence company. He is also the Managing Partner of Remake Ventures, a venture studio focused on building human-centered startups. He has helped 40+ startups raise $300M+ by finding product-market fit, including his first exit SetJam, a smart TV startup that sold to Motorola in 2012.

Want to learn more?

• Install the Evermuse MCP and put real customer signal behind your product agent
• Follow Evermuse on LinkedIn for monthly webinars
• Subscribe to our YouTube channel for past webinar recordings