Token Budgets for Coding Agents
Workshop #3 in the TwoSetAI Workshop series
If you're running coding agents in production (Claude Code, Cursor, Windsurf, or in-house tooling), your token bill probably hurts. Not because the model is expensive, but because the system runs more work than it needs to. Every message, every tool call, every retry loop compounds. What starts as a $500/month experiment turns into a $10K/month operating cost.
This session is about the coordination patterns that fix that.
Nicole walks through the routing and coordination logic behind AgensFlow: how the system decides which agent runs, when to skip a node, when to defer work, and why every decision is auditable. She'll show benchmarks she's been running on coding agents like Claude Code, and the patterns that translate to your own agent stack.
The result: lower token spend, more predictable cost, and clearer visibility into why any given run cost what it did.
Walk away with:
Coordination-policy patterns for cutting token spend on coding agents
A framework for deciding when to skip a node vs run it
Live walkthrough of benchmarks Nicole has been running on Claude Code
An approach to token-cost auditability (knowing WHY a run cost what it did)
Who this is for:
Teams shipping coding agents in production and watching token bills climb
Engineers evaluating cost-optimization approaches beyond model swapping
Founders whose AI unit economics don't work yet
Anyone who attended Workshop #2 and wants to see the runtime cost behavior in action
Format: 60 minutes, live on YouTube. ~40 min technical walkthrough, ~15 min live Q&A. Come with hard questions. If you're evaluating AgensFlow for a real project, this is the session to bring your specifics to.
About the series
Workshop #3 in the TwoSetAI Workshop series. Session #1 (July 17) introduced why multi-agent systems fail. Session #2 (July 31) introduced to AgensFlow.
See all upcoming sessions: https://twosetai.com/lab/workshops/
About the instructor
Nicole Königstein is the creator of AgensFlow, an open-source coordination-policy layer for multi-agent systems. PhD researcher on self-improving multi-agent systems. Teaches three live courses on O'Reilly Learning that pull 1,000+ signups each.
About the host
Angelina Yang runs TwoSetAI and hosts TwoSetAI and Decoding AI Builders — long-form interview platforms where she's interviewed 100+ AI founders shaping this era. fast.ai fellow. 2024 Anthropic Hackathon winner. Curates the workshop series and connects the speakers with builders who care.
Hosted by TwoSetAI.
Past events replay
Workshop #1 with Nicole Königstein — Stop Your Agents From Spinning — walked through why multi-agent systems fail in production. Replay:
Workshop #2 - Inside AgensFlow: Design Your Reliability Layer for Your AI Agents
Workshop #3 (Aug 14): Token Budgets for Coding Agents ← this one
Workshop #4 (Aug 28): Self-Improving Agent Loops Without Breaking Your Budget