Proof of Writing
Thinking out loud. Published on Middle Layer (Substack).
What makes an AI product feel cool instead of forgettable
A first-person note on AI product taste, memorable AI product experience, and why the best AI interactions feel vivid, useful, and worth returning to.
Why Anthropic and Claude fit the kind of product work I want to do
A first-person note on why Anthropic-adjacent and Claude-heavy product work fits best: workflow design, operator tooling, eval-aware systems, and turning model capability into durable products.
What teams using Claude Code actually need from product people
A first-person note on what builder-heavy teams using Claude Code actually need: workflow judgment, operator empathy, eval paths, internal tools, and better product decisions around automation.
Why AI labs need product people using Claude Code and Codex
A first-person note on why AI labs increasingly need product-minded operators who can use Claude Code and Codex to prototype workflows, eval paths, tooling, and product surfaces around model capability.
What AI evals actually are
A simple first-person explanation of what AI evals actually are, why they matter, and why the useful ones are tied to real workflows.
What AI benchmarks are really for
A simple first-person explanation of what AI benchmarks are for, where they help, and where they start to become performative.
What agent orchestration actually means
A simple first-person explanation of what agent orchestration actually means, where deterministic logic ends, and where model judgment begins.
Claude Code for Product Managers
A first-person note on why Claude Code matters for product managers, how it fits real product work, and why judgment still matters more than the tool.
My AI research interests: evals, benchmarks, and agent systems
A first-person note on the research-adjacent layer I care about most: evals, benchmark design, agent reliability, orchestration, and tooling.
Why AI product work is becoming orchestration work
Why the most useful AI product work increasingly looks like orchestrating agents, tools, workflows, and judgment rather than managing a static roadmap.
What I've actually built with Claude Code and Codex
Concrete examples of how these tools show up in product work, workflow design, public artifacts, and shipped software.
Where AI workflow design creates leverage
The real gains usually come from workflow design, not from asking a model one better question.
How I use Claude Code and Codex as a product builder
Why I think these tools are most interesting in the hands of product-minded operators, not just coders.
On Compute Economics ↗
Why the cost curve of inference is the most important chart in tech.
Durable vs. Disposable ↗
What separates companies that survive model commoditization.
The Platform-Product Blur ↗
When does a platform become a product? When does a product become a platform?