RSI Is The Buzzword Of The Week

In a week when the bosses of the big AI labs said they think the pace of AI development needs to slow down, they also seem to be doing the exact opposite.

Recursive Self-Improvement (RSI) is the idea that an AI can build the next version of itself. We can already use AI models to dramatically improve the performance of our systems. The thinking is that, eventually, the model will be able to build a better model, kicking off an exponential improvement curve.

Anthropic says Claude now does 26% of the research and development work on its next models. In March, that figure was 1%. Google DeepMind has also hinted at similar improvements.

So what does this mean for product teams?

AI is already an amazing tool for product teams. Right now, it’s an amazing tool but you still need people with enough context to prompt the AI well, and enough taste to judge the output. But taste is improving fast.

Context is still the harder problem. That depends on memory breakthroughs that don’t really exist yet, as well as people being disciplined about keeping the context available to AI systems up to date.

I think this is a net positive for people (assuming we can sort out alignment). But I am net positive on AI in general. What are your thoughts?

This Week's Poll

Last week’s poll
Last week we asked: As AI increases your team’s capacity, where is the hardest decision moving? Choosing which work to stop or kill led with 44.4%. Deciding what evidence still needs humans and which standards and guardrails to enforce each received 22.2%, while changing roles and teams received 11.1%.

This Week’s Updates

Enabling the Team

Want more confidence in AI? Give it more context. by Halina Mader
Give AI company data, institutional knowledge and work context to improve trust and cut rework; adoption metrics alone miss the point.

The executive loop by Jade Rubick
Effective leaders establish what needs to happen, work through the constraints and trade-offs, then build the conditions that make it happen.

Product

The most important product decision is what you don’t build by Liam Nugent
Make the lifetime cost of features visible, solve each customer need directly and reward removing complexity as well as shipping new work.

Will AI Eliminate PM Jobs? by Itamar Gilad
AI broadens what PMs can do, but judgement, coordination, strategy and deciding what to build remain essential even as tasks become automated.

Research

Do Participants Say Less to AI Than to Human Moderators? by Lucas Plabst, Jeff Sauro, Eva Sundberg and Jim Lewis
In one controlled study, participants spoke 45% less to AI moderators; shorter interviews raise questions about depth, not proof of poorer insight.

AI, Margins, and the New Division of Labour: Three Reversals Reshaping How Research Operates by Kate Towsey
Proactive insight systems need research craft and operational expertise together; automating output cannot replace strategic research leadership.

Design

Five eras of UX design and the lessons to keep in the AI era by Patrick Neeman
Each UX era repeats old mistakes; retain human factors, context and feedback as AI changes the tools and expands what designers can make.

AI in Design 2026: The inflection point is here by Designer Fund
Research with 900 designers shows a shift from static deliverables to building products and systems, expanding design’s scope and responsibilities.

Dev

Maybe We Shouldn't Be Reviewing All This Code by Rachel Laycock
Move judgement, learning and architectural alignment before code is written, then use automated checks for the constraints machines can verify.

We are all Product Engineers now by Laurie Voss
As coding costs fall, engineering value moves towards understanding customers, shaping solutions and owning product outcomes across the lifecycle.

Seen an interesting article online? Share it with us, and we might feature it in our next issue!
Click here to share an article

Who Controls The Customer Journey In Agentic AI?

Expedia, Target and Chase on designing AI that can act for users without losing trust, transparency or control.

AI can increasingly recommend, personalise, purchase and complete tasks on a customer’s behalf. The harder question is not what AI can automate, but what it should automate, and how teams preserve trust as more decisions move into the background.

At UXDX San Francisco, Rachel Been from Expedia, Purvi Shah from Target and newly announced Chris Stang from Chase will explore how leading consumer organisations are approaching that balance. Moderated by Alex Burke, CEO of Optimal, the discussion will cover where automation should stop, when users need explicit approval, and how products can make AI-driven decisions more transparent without adding unnecessary friction.

Explore the San Francisco agenda for 5-6 November, and use NEWSLETTERSF26 for 10% off tickets.

FREE COMMUNITY EVENTS

Video Of The Week

If AI Can Build Products, What Are The Humans For?

AI has made it dramatically easier to turn an idea into working software. The harder question is how teams avoid moving faster in the wrong direction when execution is no longer the main constraint.

In this talk that I gave earlier this year, I look across the product delivery lifecycle to show where AI can already take on design, code, testing and deployment, and where human judgement still matters most. Roles may merge, but understanding customers, choosing direction and coordinating decisions do not disappear.

Watch the full session to see why faster building changes the product team, and what people need to become better at when shipping is no longer the hard part:

Want to go into how careers and leadership are shifting as AI compresses the ladder? My ebook Managing Your Career In The Age Of AI explores how to build judgement, relationships and influence in a world that keeps trying to automate the surface of the work.