Play, Prompts, and the Perils of Incrementalism

In our latest episode, Michelle Lee (IDEO Play Lab) makes the case that play unlocks the next billion-dollar AI market. She reminds us that kids don’t stop at answers—they ask what if and turn shoes into cars or planes. That divergent mindset is exactly what product teams have lost.

“Play is one of the best ways to challenge the norms, to think wide, imagine new possibilities.”

Michelle shares:

* How IDEO discovered billion-dollar opportunities (like PillPack, later acquired by Amazon) by staying curious.

* Why teams should sometimes use older, glitchier versions of AI tools, because the “mistakes” spark better ideas.

* Why incrementalism burns teams out and how designing for attitudinal loyalty beats chasing short-term metrics.

🎧 Listen here → Play unlocks the next billion-dollar AI market

Uncomfortable Truth: Most “AI strategies” today are adult strategies — converging too quickly, chasing predictability, and mistaking incremental progress for innovation. That’s why the breakthroughs are happening elsewhere.

Product Workshop: Find your Disruptive Path

If your roadmap looks like everyone else’s, you’re already behind. Our next AI Product Strategy Workshop (Oct 30) is built for teams who want to:

* Go beyond features and efficiency to discover truly disruptive opportunities.

* Use LLMs as intelligent sparring partners to pressure-test fragile ideas before they waste time and budget.

Spots are limited → Register here

Hard-Cutting Take: If your roadmap reads like your competitors’, it’s not strategy—it’s risk management dressed up as vision.

Incrementalism is the Silent Killer

We’ve all felt it: the slow grind of incremental product decisions that look safe but quietly kill ambition. My new piece argues that incrementalism is the silent killer of AI products—a trap for teams rewarded for predictability instead of progress.

Read it on LinkedIn → Incrementalism is the Silent Killer of AI Products

Uncomfortable Truth: Incrementalism feels safe because it rarely fails spectacularly. But it guarantees mediocrity—and in AI, mediocrity is indistinguishable from irrelevance.

AI Launches to Watch

A wave of new releases will reshape how we design and ship AI products:

* OpenAI: Stripe/Shopify integrations + new pre-designed prompts for professionals.

* Anthropic: Chrome plugin + Claude 4.5 Sonnet, a faster, cheaper model that expands prototyping and evaluation capabilities.

* OpenAI Sora 2: Newly launched today, unlocking endless possibilities for video and creative storytelling, signaling a profound shift in how generative tools will shape the creative industries.

These aren’t just upgrades—they’re reshaping commerce and the browser itself. The integration of Stripe and Shopify signals AI’s deepening role in transactions, while Anthropic’s Chrome plugin points to a future where the browser becomes a true intelligent workspace. It’s likely why Atlassian just acquired The Browser Company (maker of Arc and Dia). These moves aren’t incremental improvements; they’re like a rushing river, pushing the entire industry forward whether teams are ready or not.

The next frontier isn’t who has the biggest model—it’s who controls the browser as the operating system for work. And then when we looking beyond, it will be who controls our real world experiences… (more on that soon with an upcoming guest)

When Projects Go Off the Rails

Even as the models improve, they’re only as good as the prompts and evaluations behind them. We’ve seen how easily “comprehensive business cases” collapse when fabricated ROI, vendor costs, and timelines are passed off as fact.

It’s the Wizard-of-Oz problem: behind the curtain, most AI projects are stitched together with fragile assumptions.

Uncomfortable Truth: Most AI decks aren’t strategy—they’re theater. And like any stage play, the curtain eventually falls.

Hidden Pitfalls of AI Scientist Systems

A new paper, “The More You Automate, the Less You See: Hidden Pitfalls of AI Scientist Systems” (arXiv, Sep 10, 2025), warns about the risks of fully automated science pipelines. By chaining hypothesis generation, experimentation, and reporting end-to-end, teams risk producing results that look authoritative but mask invisible errors and systemic failures. (arxiv.org)

Uncomfortable Truth: Automation without visibility doesn’t accelerate discovery—it accelerates blind spots.

Articles & Ideas We’re Tracking

* Prompts.chat → A growing open library of prompt patterns that shows why better prompt design, not just better models, is becoming the key differentiator for teams.

* AI in the workplace: A report for 2025 (McKinsey) → McKinsey highlights that while adoption is accelerating, most organizations hit cultural and skills barriers long before technical ones.

* The Architecture of AI Transformation (Wolfe, Choe, Kidd, arXiv) → This 2×2 framework shows why most companies get stuck in incremental “legacy loops” rather than unlocking transformational human-AI collaboration.

* TechCrunch: Paid raises $21M seed to pioneer results-based billing with AI agents → A new startup model where AI agents don’t just assist but transact, shifting billing to results instead of hours.

* Harvard/Stanford study on ROI of GenAI → New research explains why so much GenAI spend has failed to generate returns: productivity gains get trapped in organizational silos and misaligned incentives.

* Beware coworkers who produce AI-generated ‘workslop’ → Surfaces a new term—workslop—to describe AI outputs that look polished but lack real substance, shifting the burden downstream to humans.

Hard-Cutting Take: The ROI isn’t missing because the models are weak—it’s missing because organizations are. Incentives, silos, and incremental thinking kill more AI projects than hallucinations ever will.

This is a public episode. If you would like to discuss this with other subscribers or get access to bonus episodes, visit designofai.substack.com