Master Scaling AI Coding Agents: Cut Costs with Best-of-N

​Your coding agent's biggest cost isn't planning. It's the executor.

​Every agent session runs dozens of edits and test-fix cycles, often testing several candidate solutions in parallel — and what happens in that executor layer decides both the speed and the bill.

​Kwasi Ankomah, Lead AI Architect at SambaNova Systems, joins Data Science Dojo for Part 2 of the SambaNova Webinar Series to show how parallel execution, best-of-N selection, and disaggregated serving turn agent costs into something predictable. With 15 years across financial services, consulting, government, and startups, Kwasi specializes in multi-agent orchestration, deep agent architectures, and context engineering.

​What You'll Learn
→ Where coding agents actually spend time and money
→ How parallel executors change the economics of running more candidates
→ What best-of-N selection is and why it works
→ Why fast, affordable inference is essential for test-time compute
→ How disaggregated serving shapes throughput, latency, and utilization
→ How to evaluate your own agent infrastructure's real cost drivers