Has he shipped production software?
Yes. The current public receipts include live products for agent registry, agent payment controls, agent games, travel writing, and AI education. The strongest evidence is the live-product set rather than a private repository tour.
What is the clearest technical through-line?
Auditability for AI systems: stable identity, policy gates before action, signed receipts after action, and replayable workflows. That shows up as products, not just essays.
Can he explain engineering tradeoffs clearly?
The writing is mostly short operational essays: cost ownership, agent deployment boundaries, MCP as production dependency, and the habit of proving work through logs, receipts, and boring defaults.
Is there public code to inspect?
Some, but not all. Several product repositories are private. Public code receipts include Learn AI, a small DeepSeek compatibility proxy, and other public GitHub work; use the live products as the broader shipped-work evidence.
What should I not infer from this page?
Do not infer team size, revenue, or customer adoption unless a receipt says it. This page is deliberately evidence-first: when public proof is thin, it says so rather than filling the gap with adjectives.