What’s shaping my thinking.

The book I recommend first when someone asks how to work with AI without turning the conversation into a technical lecture.

Grounded experiments and observations from someone who uses the tools daily. The back catalog is unusually useful.

Long-form writing about AI, products, and how the way we work is changing, with enough detail to be useful.

A working model for turning completed projects into instructions, tools, and checks that improve the next run.

A concrete operating model for giving small teams clear ownership while shared systems carry context across the company.

The 11-star exercise pushes a product past the first feasible answer, then works back to the version worth shipping.

A file-based approach to agent workflows where folders define stages, markdown carries context, and each output stays open to human review.

Peter tests AI tools through product work, detailed tutorials, and interviews with people shipping the systems.

Real people walking through real AI workflows. I care much more about the work on screen than another round of predictions.

A useful conversation about code agents, parallel work, AutoResearch, and what changes when experimentation gets much cheaper.

A full coding session that builds a small GPT and exposes each step between tokens, attention, training, and generation.

The visual explanation I send when someone wants intuition for what an LLM is doing without starting from the math.

A visual walk through tokens, embeddings, attention, and the calculations inside a transformer.

A useful overlap of product judgment and hands-on AI tutorials.

He teaches by building, including the awkward middle where the first attempt does not work. That is the useful part.

The clearest technical intuition I have found for models, agents, and the changing shape of software work.

Frequent, concrete experiments about how AI changes teaching, research, and everyday knowledge work.

Working notes from building AI products and changing how a company operates around them.

Operators show their real workflows, including where AI helps, where it breaks, and what still needs judgment.

He tests specific AI tools and workflows through product decisions, tutorials, and interviews.

His ICM work treats folders, markdown, and reviewable files as the control surface for agent workflows.

His build videos keep the failed attempts and fixes that polished demos usually remove.

Visual explanations that make the mechanics of neural networks and transformers easier to inspect.

Concrete lessons from building, monetizing, and distributing consumer apps instead of founder theater.

Useful signal on Claude Code, developer tools, and the constraints behind agentic software work.

Experiments in agents, orchestration, and composable systems that are close to the problems I am building around.

A daily thinking and building tool, especially when a project needs a lot of context held together.

I use coding agents for scoped implementation work, parallel exploration, testing, and the operational details that make software ship.

Where product context, decisions, research, and operating systems stay connected instead of disappearing across documents.

A practical foundation for shipping authenticated products with Postgres, storage, and real-time behavior without unnecessary ceremony.

I use Pages, Workers, D1, and Wrangler to deploy sites and run product infrastructure at the edge.

I use Doppler to keep environment secrets out of repositories and deliver the right configuration to each runtime.

Build, test, metadata, screenshot, and release workflows run through versioned automation beside the code.

I use release-aware errors and traces to connect a production failure to the build and code path that caused it.

I use product events and funnels to inspect how a shipped flow behaves after it reaches users.

I use RevenueCat for subscription state, purchase restoration, entitlements, and store-side release work across mobile apps.

I use the shipped interaction examples to study timing, transitions, and motion that explains a state change.

A reference library for hover states, transitions, feedback, and small interface details.