The Inner Game of AI: Lenny on PMs and Unfair Advantages

Lenny Rachitsky — The New Inner Game: Your Unfair Advantage in the Age of AI#

Source: https://www.lennysnewsletter.com/p/the-new-inner-game-your-unfair-advantage Credibility: High (first-party newsletter post from a well-sourced PM practitioner with a large verified readership)

What happened: Lenny published what he's calling a playbook for the "biggest upheaval of work we'll see in our lifetime." The essay is framed around what PMs have that AI doesn't — the things that become more valuable, not less, as AI gets better at execution. The central argument: the inner game (judgment, taste, relationships, context, purpose) is becoming the actual job. The outer game (writing, coding, research, summarizing) is increasingly delegated to agents.

Key patterns:

• The outer game (skills, tactics, execution) is being commoditized fast — AI can do most of it, and will do more
• The inner game (who you are, what you care about, how you think, your network) is the durable advantage — it compounds and can't be copied by a model
• The shift isn't "AI replaces PMs" — it's "AI surfaces which PMs were coasting on execution and which ones have genuine judgment"
• Lenny specifically frames this as an opportunity: if AI handles the busywork, PMs with real taste and strategic clarity will be more valuable than ever, not less

Why it matters for PMs: This is the clearest articulation I've seen of the split that's actually happening inside product teams right now. The PMs who are struggling with AI aren't struggling because they can't use the tools — they're struggling because the tools removed the scaffolding that hid unclear thinking. If your value was primarily in being fast at writing PRDs and doing research synthesis, that's a real problem. If your value was in navigating ambiguity, knowing users, and making calls that require context only you have — that's getting more leveraged. The frame also has practical implications for how you spend time: if you're still the bottleneck on execution tasks, something's wrong.

Critical questions:

• How do you assess which parts of your current job are "inner game" vs. execution that could be delegated? Most PMs haven't done this inventory honestly.
• Does this framing overweight individual PMs and underweight team dynamics? The inner game is partly relational — but whose relationships and institutional memory count when everyone's moving fast?
• Is "taste" learnable or is Lenny describing something more like accumulated experience? If so, what does this mean for early-career PMs?
• The essay assumes judgment improves with reflection — but does the pace of AI-assisted work actually create space for that, or does it just accelerate the treadmill?

Action you could take today: Write down your last five significant product decisions and label each one as "inner game" or "outer game." If most of them were execution-heavy (writing, formatting, research synthesis), that's a real signal about where you're spending leverage.

Mistral AI — Mistral OCR 4#

Source: https://mistral.ai/news/ocr-4/ Credibility: High (first-party product announcement from Mistral AI)

What happened: Mistral shipped Mistral OCR 4, described as a "state of the art document intelligence model." The announcement is light on benchmark specifics in the scraped data, but the positioning is clear: this is a purpose-built document understanding model, not a general-purpose model applied to OCR. Mistral has been one of the more disciplined shippers among foundation model companies, so a new model in a specialized capability area is worth flagging.

Key capabilities:

• Document intelligence focus — goes beyond character recognition to understanding document structure and content
• Positioned as state-of-the-art for the category
• Fits into Mistral's pattern of shipping focused, API-accessible models rather than consumer-facing products
• Available via Mistral's API, consistent with their enterprise-first distribution strategy

Why it matters for PMs: Document processing is one of the most underrated AI use cases in enterprise software — it's everywhere (contracts, invoices, financial statements, medical records, forms) and it's been messy to build reliably. Any fintech PM, or anyone building on top of document-heavy workflows, should be aware that the quality bar for this category just moved again. If you've been relying on a general-purpose vision model to handle documents, it's worth evaluating whether a specialized model like OCR 4 changes the build-vs-buy or which-vendor calculus for your stack. For Rocket Money specifically, anything that improves structured data extraction from financial documents — statements, tax forms, pay stubs — is directly relevant.

Critical questions:

• How does OCR 4 perform on messy, real-world documents vs. clean PDFs? The gap between benchmark and production performance in OCR has historically been wide.
• What's the pricing and how does it compare to using a general-purpose multimodal model for the same task?
• Does it handle handwritten content, non-standard layouts, or multilingual documents? Those are the hard cases that matter most in production.
• Is this differentiated enough from what AWS Textract, Google Document AI, and Azure Form Recognizer already do well, or is this primarily relevant because of Mistral's European data residency story?

Action you could take today: If you have a document processing workflow in your product or on your roadmap, pull up one of your hardest test cases (a messy statement, a non-standard form) and run it through Mistral OCR 4 via their API. Compare against your current solution and check if the delta justifies a switch.

Vercel — Deploy from Claude Design to Vercel#

Source: https://vercel.com/changelog/claude-design-and-vercel Credibility: High (first-party changelog entry)

What happened: Vercel shipped a direct integration that lets you deploy from Claude Design straight to Vercel. This is a one-step closing of the loop between AI-generated design output and production deployment. Previously, you'd use Claude Design (Anthropic's design tool) to generate UI, then manually move it to a repo and deploy. Now that handoff is automated.

Key capabilities:

• Direct deploy path: Claude Design output → Vercel hosting, no intermediate steps
• Fits into Vercel's broader push to become the deployment layer for everything AI-generated (v0, Claude Design, and their Eve agent framework are all converging on this)
• Combined with their recent WebSocket support and redesigned trace viewer for Workflows, Vercel is building the scaffolding for AI-assisted development end-to-end

Why it matters for PMs: This is another data point in a pattern that's been building all month: the design-to-production pipeline is compressing. What used to be a multi-day handoff (design → spec → engineering → review → deploy) is becoming a same-session workflow for prototypes and early-stage product work. For PMs, this changes the conversation about what counts as a prototype — a Claude Design mockup you can deploy and share with users in minutes is meaningfully different from a Figma file that requires engineering time to make real. The practical implication: your bar for "test this with users before investing in it" just got lower.

Critical questions:

• What's the quality ceiling here? Can you actually ship production features this way, or is this still prototype-quality output that needs engineering cleanup?
• How does this fit into existing team workflows where design handoff is a formal process with review gates? Does this accelerate things or create chaos?
• What's the security and compliance story for deployments that originate from an AI design tool rather than a reviewed codebase?
• Does this create tension between design tools (Figma) and Anthropic's design products? Figma has been building the same pipeline from a different direction.

Action you could take today: If you've been using Claude for design exploration, try the full path: design something in Claude Design, deploy it to Vercel, and share the URL with a teammate. Time the whole thing. That timing is the new baseline for "how fast can I test an idea with a real URL."

Teresa Torres — Creating Experiences (All Things Product Podcast)#

Source: https://www.producttalk.org/creating-experiences-all-things-product-podcast-with-teresa-torres-petra-wille/ Credibility: High (Teresa Torres's own blog, first-party content)

What happened: Teresa Torres joined Petra Wille on the All Things Product podcast to talk about creating experiences — and this is a different episode from the organizational change one that was already covered. The title and framing suggest it's about the PM craft of experience design, not just process or discovery methodology.

Key patterns: Based on the episode framing and Teresa's consistent body of work, the conversation likely covers how PMs think about the full user experience, not just features — the distinction between shipping functionality and creating something people actually find valuable and memorable. Teresa has been consistent that continuous discovery and outcome-thinking are prerequisites for building experiences rather than just outputs.

Why it matters for PMs: Teresa is one of the clearest thinkers on the gap between "we shipped it" and "it worked." In an era where AI makes shipping faster, the question of what makes something worth shipping — what experience it creates, not just what it does — becomes more important. If shipping velocity increases but product judgment doesn't, you get faster-produced mediocrity. Teresa's framing cuts directly against that.

Critical questions:

• Is "creating experiences" a useful frame when AI features often feel ambient or invisible to users? Does the experience design framing translate to AI-assisted products?
• How do you measure whether you've created an experience vs. just shipped a feature? Teresa's frameworks usually have a concrete answer — worth checking the episode for it.

Action you could take today: Listen to the episode (link above). Specifically listen for how Teresa distinguishes between building features and creating experiences — and whether her framework changes how you'd describe the last thing you shipped.