Run self-improving coding agents in production: The full loop at scale

Closing the improvement loop in production

In the series finale, Kwasi Ankomah puts the whole improvement loop on a schedule and runs it at production scale on SambaNova — then judges the capstone coding agents built across the series.

About this webinar

Improving an agent once is useful. Improving it continuously, safely, and at scale is what makes it production-ready.

In Part 5, we used traces to diagnose failures and automatically improve the agent. In this final session, Kwasi Ankomah turns that process into a scheduled production loop: trace, evaluate, diagnose, improve, and redeploy. He adds the safeguards production systems need — versioning, regression gates on the test suite, and human approval of changes — and runs the executor swarm at production scale on SambaNova. The session closes with the capstone, where the coding agents improved across the series are scored on held-out tasks.

What you will learn

• How to automate the full loop — trace, evaluate, diagnose, improve, redeploy — on a schedule
• How to version agent changes so every update is traceable and reversible
• How to use the test suite as a regression gate so no change ships if it makes things worse
• How to keep humans in control by requiring approval for agent changes
• How to run an executor swarm at production scale on SambaNova
• How the capstone agents perform when scored on held-out tasks

Why closing the loop matters

Agents in production face new tasks, changing codebases, and evolving requirements. An agent that was tuned once will slowly fall behind. A closed loop keeps it improving — while regression gates, versioning, and human approval make sure every change is safe. This is how teams move from a promising prototype to a coding agent they can trust at scale.

Who should attend

AI and ML engineers deploying agents to production, platform and MLOps engineers responsible for reliability and release processes, and technical leads planning how to operate agents at scale. Anyone who followed the series will see their work come together in the capstone, but newcomers will still get a complete picture of a production agent loop.

Kwasi Ankomah is the Lead AI Architect at SambaNova Systems, where he builds production agentic AI systems on purpose-built RDU inference infrastructure. With 15 years of experience spanning financial services, consulting, government, and tech startups, Kwasi bridges deep technical expertise with a hands-on understanding of real business challenges. He specializes in deep agent architectures, multi-agent orchestration, and context engineering. Connect with Kwasi on LinkedIn →