Work with me

I’m Dayna Blackwell, and through Blackwell Systems I help teams build AI agents in Go and run them in production. The method is the one this guide uses: measure on your own data, say what the evidence can and can’t show, and recommend only what holds up.

What I can help with

Architecture and production reviews of agent systems. A review of an agent you have built or are about to: the loop and its limits, tool design, context management, retries, safety and approval, and what happens when a run crashes halfway through. You get findings ranked by risk, each traced to code or to a run, and what to change first.

Eval design and measurement. Evals for an agent whose answers vary: task sets with tiers, graders written as code, ranges around every rate, and comparisons that tell a real improvement from chance. Then you can tell whether the next prompt or model change helped.

AI cost optimization. An audit of what your LLM calls cost and why, across prompt caching, model routing, batch processing, context and tool-output size, and the shape of the agent loop, with every saving measured before and after on your workload.

AI opportunity assessment. A fixed-scope, vendor-neutral read on where AI fits your business and where it doesn’t, in plain language, for teams that want a straight answer before they build anything.

Full list of services, including fractional CTO work and technical due diligence.

Background

• The book. Building AI Agents in Go, which builds a crash-triage agent from a 30-line loop to production, with every result measured.
• Bide, a library for durable agents in Go.
• GCF, a data format for LLMs that ships in Google’s Chrome DevTools MCP and more than 30 other projects.
• Upstream work: over 50 pull requests merged into projects from Google, Anthropic, GitHub, and HashiCorp, including the MCP SDKs for Go and Python.
• Before that, five years as the technical owner of Best Western’s loyalty platform, serving tens of millions of members at more than 5,000 hotels in over 100 countries.