Matt Berryhill builds AI-native discovery systems.
Berryhill.dev is his public field notebook on agentic workflows, governance, provenance, and the operator judgment required to turn AI work into shipped proof.
Most AI content stops at tools, prompts, and demos. The harder problem is building systems that know who owns the work, what evidence was used, what changed, what shipped, and where a human must stay in the loop. This site tracks that layer: the workflows, protocols, review gates, and operating habits that make agentic systems useful outside the demo.
berryhill.dev / field-notes
Public notes on AI-native systems, agent governance, provenance, and shipped proof.
Interface metaphor, not live telemetry.
Start with the operating layer: agent governance, shipped-work proof, protocol boundaries, small named fleets, continuity, and the systems that keep AI work accountable.
Featured Posts
- Field note
The Agent-Stack Governance Checklist: Authority, Review, Provenance, and Handoffs
A practical checklist for designing authority, review, provenance, and handoff gates around MCP, A2A, and agent SDK workflows before they touch production work.
- Field note
Protocols are saturating; the operator's gap is governance, not transport
MCP, A2A, and ACP are solving transport. But authority, review, memory, and handoff — the governance layer no spec hands you — is what makes a stack safe.
- Field note
Recent Progress in RWA for Securities and Commodities What International Investors Need to Know
Discover how Real World Asset tokenization is transforming securities and commodities markets. Learn about recent breakthroughs, market growth, and investment opportunities for international invest...
- Field note
The Future of Developer Careers in the Age of AI Coding Agents
Discover how AI coding agents are reshaping developer careers. Learn which skills matter most and how to future-proof your tech career in 2025 and beyond.
Recent Posts
- Field note
Your website should be a discovery system, not a content archive
A blog archive preserves posts. A discovery system preserves the claims, evidence, updates, and decisions that make public thinking compound.
- Field note
More posts does not mean more indexed pages
Publishing more posts is not the same thing as earning more indexed pages. Search visibility is a crawl-trust pipeline: inventory, routes, canonicals, sitemaps, internal links, and readback all have to agree.
- Field note
Protocols Make Agent Exchange Legible. Governance Still Lives Above the Protocol.
A2A and MCP make agent exchange easier to standardize, but they do not decide authority, review, provenance, memory, escalation, or accountability.
- Field note
Green Status Is Not Shipped Work
Agent workflows need deliverable-level acceptance tests because internal completion state is not the same thing as user-facing proof.
- Field note
Your AI Shouldn't Be One Assistant. It Should Be a Room Full of Desks.
One generalist assistant is the wrong unit of deployment. The right unit is a small, named fleet — each desk owns a queue, a memory, and a hand-off. Operator sits in the loop.
- Field note
Patent Search Is Table Stakes. IP Intelligence Is the Product.
AI is making patent search easier to access, but durable product value is moving to always-on IP intelligence: monitoring, context, evidence, and human review workflows.