Home
Jun 30, 2026
View All

How PMs Are Rethinking Leverage, Autonomy, and AI-Shaped Problems

·2 underrepresented voices

The Short Version#

Today's signals converge on a single uncomfortable question: what does PM work actually look like when AI can do more of it? Lenny published a framework for PM leverage with AI, Teresa Torres named a new class of problems that AI creates rather than solves, and Dare Obasanjo surfaced a stark cautionary case from Ford about what happens when you replace senior engineers with AI and skip the quality check.

Lenny Rachitsky — How Top PMs Increase Their Leverage with AI#

Source: https://www.lennysnewsletter.com/p/how-top-pms-increase-their-leverage Credibility: High (Lenny's Newsletter, first-party PM craft content from a well-sourced practitioner)

What happened: Lenny published a framework for how PMs can get more leverage from AI in their day-to-day work. The framing is around "leverage" specifically — not just using AI as a writing assistant, but using it to extend what you're capable of doing as a PM. Based on the excerpt and the consistent thread in recent Lenny content (the Gusto/Claude Code story, the OpenAI Codex interview), this continues a series on what high-leverage PM behavior looks like in an AI-accelerated environment.

Key PM craft patterns:

  • The central question is leverage, not productivity — meaning: how does AI change what's possible for a PM, not just how fast they can do existing tasks
  • High-leverage uses tend to involve AI in research, synthesis, and decision prep — not just drafting
  • The companion Gusto piece (Eddie Kim, CTO) shows a worked example: a 5-person team shipping a full AI product line in 10 weeks using Claude Code, a perma-Zoom, and zero documentation — which suggests the leverage question is also structural, not just personal
  • The OpenAI Codex interview framing ("taste matters more than ever") suggests the emerging PM leverage point is judgment and taste, not execution speed

Why it matters for PMs: If "leverage" is the right frame, then the PM question isn't "which tasks can AI help me do faster?" It's "which decisions and outcomes can I now reach that I couldn't before?" That's a harder question and most PM workflows haven't been redesigned around it. The Gusto example is worth studying — 10 weeks, 5 people, full product line, no docs — because it suggests the organizational overhead we've normalized (Figma, Jira, documentation rituals) may be more about coordination than necessity.

Critical questions:

  • Is "leverage" the right frame for all PMs, or mainly for senior PMs who already have execution bandwidth to spare?
  • The Gusto example removed documentation entirely — what's the durability cost of that? Who pays when someone leaves the team?
  • If taste and judgment become the primary PM differentiator, how does that change hiring, performance reviews, and how PMs demonstrate value?
  • What happens to junior PMs who haven't yet built the judgment that makes high-leverage AI use work?

Action you could take today: Pick one recurring PM ritual (weekly status update, research synthesis, spec writing) and try completing it entirely through AI prompting this week — then honestly evaluate whether the output quality changed, or just the time.

Teresa Torres — AI-Shaped Problems#

Source: https://www.producttalk.org/ai-shaped-problems-all-things-product-podcast-with-teresa-torres-petra-wille/ Credibility: High (Teresa Torres's own platform, Product Talk — recognized PM craft authority)

What happened: Teresa Torres, alongside Petra Wille, published a new episode of the All Things Product podcast titled "AI-Shaped Problems." Based on the title and Torres's consistent framing in her recent work (the organizational change episode, the consent coach case study), this appears to focus on a category of problems that AI tools create — not just problems they solve. Torres has been increasingly focused on the human and organizational side of AI product adoption, not the capability side.

Key PM craft patterns:

  • "AI-shaped problems" as a named category suggests Torres is pointing at problems that emerge specifically because AI was introduced — not legacy problems that AI helps with
  • This connects to her prior work on organizational resistance, ethics, and how teams struggle to adopt continuous discovery when AI changes the discovery process
  • The framing matters: most PM conversations are about "AI as solution." Torres is naming "AI as source of new problems" — which is a more honest and harder frame
  • Petra Wille brings a European enterprise consulting perspective that grounds this in real organizational dynamics, not startup-scale thinking

Why it matters for PMs: Most PMs are being asked to ship AI features or adopt AI tooling. Very few are being asked to map the new problems those decisions create downstream — for users, for teams, for trust. Torres naming this as a distinct problem category is a signal that the PM craft conversation is maturing past "how do I use AI" toward "what do I owe users and teams when I do."

Critical questions:

  • What are the specific "AI-shaped problems" Torres is pointing at — are they about user behavior, team dynamics, or product ethics?
  • Is this a framework with concrete PM practices attached, or a framing shift without tooling?
  • How do AI-shaped problems show up differently in enterprise vs. consumer products?
  • What's the discovery process for finding AI-shaped problems before they become support tickets or trust failures?

Action you could take today: For one AI feature you've recently shipped or are planning to ship, write a one-paragraph answer to: "What new problems does this create for users that they didn't have before?" and share it with your team in your next planning session.

Dare Obasanjo — Ford's AI Layoff Reversal and Quality Failures#

Source: https://mas.to/@carnage4life/116825465041766357 Credibility: High (Dare Obasanjo is a longtime senior software industry voice; the underlying Ford story is reported news)

What happened: Dare Obasanjo flagged a significant cautionary case: Ford replaced hundreds of senior engineers with AI, it backfired badly, and they've been rehiring. The quality impact was severe — Ford is now the most-recalled automaker in the US in H1 2026, with 51 recalls covering over 11 million vehicles. The post frames AI adoption as directly implicated in the quality failures.

Key patterns:

  • The decision to replace senior engineers with AI wasn't just a cost play — it removed the institutional knowledge and judgment that quality systems depend on
  • 51 recalls covering 11 million vehicles in one half-year is a catastrophic quality signal, not an incremental one
  • This is one of the clearest reported examples of AI replacement backfiring at scale, in a domain (automotive safety) where failure has life-safety consequences
  • The rehiring is itself a signal: the company is paying twice — once to let people go, once to bring them back — plus the cost of the recall crisis in between

Why it matters for PMs: This is the "AI replaces humans" story playing out in a domain with hard ground truth. Automotive recalls are not a soft metric — they're legally defined, publicly reported, and expensive. PMs building AI features that touch quality-sensitive workflows (code review, safety checks, compliance verification) should be watching this case carefully. The question isn't "can AI do this task?" but "what institutional knowledge is embedded in the humans doing this task, and what happens when that's gone?"

Critical questions:

  • How much of the quality failure was specifically attributable to AI versus general headcount reduction and organizational disruption?
  • Are there domains where this pattern is more likely — high-stakes, judgment-intensive, slow-feedback-loop work?
  • What's the signal for PM teams building AI tools: are we helping people do their jobs better, or building tools that will be used to justify replacing them?
  • How do you preserve institutional knowledge and quality judgment as AI takes over more execution tasks?

Action you could take today: If your team is using AI to automate any quality or review step, map one specific scenario where the AI would fail silently — and make sure there's a human check in that path before the failure reaches users.

Quick Hits#

The Thread#

The leverage question is getting harder to dodge. This week's signals keep circling the same tension: AI gives PMs and teams more capability, but the places where it backfires are almost always the places where judgment and institutional knowledge were doing invisible work. The Gusto story (10 weeks, no docs, full product line) and the Ford story (51 recalls, rehiring engineers) are two ends of the same question — which human knowledge can AI replace, and which knowledge only looks replaceable until something breaks.

Sit With This#

Ford replaced senior engineers with AI to cut costs, the quality failures were severe enough to trigger 51 recalls and 11 million vehicles, and they're now rehiring the people they let go.

For your team: Is there a quality or review step in your current product workflow where AI has taken over execution — and if that step failed silently for 30 days, how long before you'd know?