Lessons From 25 Trillion Tokens — Scaling AI-Assisted Development at Kilo

Why It Matters

Kilo’s experience proves that AI‑orchestrated development can slash cycle times and costs, giving early adopters a decisive competitive edge in a rapidly evolving software market.

Key Takeaways

• AI agents let developers become orchestrators, not coders.
• Reducing collaboration boosts velocity; single‑engineer feature ownership improves outcomes.
• Trust hinges on latency, context accuracy, and model selection.
• Progressive AI adoption ladder: autocomplete → chat → single agent → orchestration.
• Matching the right model to task cuts cost and speeds delivery.

Summary

Kilo’s co‑founder and CEO Scott outlined how the company processed more than 25 trillion tokens since its May launch and used that data to reshape software engineering. By treating 2027‑level AI tools as core collaborators, Kilo shifted developers from manual coders to orchestrators of multiple AI agents, enabling the team to ship one to two features weekly with only fifteen engineers. The firm deliberately minimized traditional collaboration, assigning end‑to‑end ownership of each feature to a single engineer and eliminating meetings that slow progress. Trust emerged as the critical metric: latency spikes above 200 ms, inaccurate file paths, or irrelevant suggestions instantly eroded adoption. Kilo addressed this by tightening context awareness, selecting cost‑effective models for specific tasks, and continuously improving latency. Scott highlighted concrete examples: Pedro, a one‑person data team, built a full DBT model in weeks by giving the AI full repository context, achieving ten‑fold speed gains. He also noted that Kilo’s release notes show a steady cadence of feature deliveries, underscoring the tangible impact of AI‑driven orchestration. The broader implication is clear: organizations that re‑engineer development processes around trustworthy, context‑rich AI agents can dramatically accelerate delivery while curbing costs. Success hinges on managing latency, ensuring accurate context, and matching the right model to each workload, offering a blueprint for the next wave of AI‑augmented software teams.

Original Description

Scott Breitenother (Kilo Code) Keynote at the Coding Agents Conference at the Computer History Museum, March 3rd, 2026.

Abstract //

Developers aren’t really coding anymore—they’re managing AI, and Scott Breitenother is already running a team where 15 engineers, powered by 25 trillion tokens, ship faster than ever as one-person teams owning entire products, which raises a bigger question: are engineers becoming 10x… or just turning into middle managers for agents?

Bio //

Scott Breitenother leads Kilo, the company behind the fastest-growing open-source coding agent. Before Kilo, Scott founded Brooklyn Data, a data and analytics consultancy he grew from his Brooklyn apartment to a 100+ person consultancy serving Fortune 500 clients before a successful exit.