/ About this event
Two things most engineering teams have backwards.
Longer context does not make an agent smarter. Past a point it makes it reason worse, and you pay premium prices for the privilege. And the cheapest model is rarely the cheapest one: it retries, it fails review, and you end up buying the same piece of work three times.
Neither of those is a pricing problem, and that is the whole point: you do not pay for tokens, you pay for tokens that produce no value. Nobody decided how your agents spend them. It arrived with the tools.
Two levers decide it: which model does the work, and what is in its context when it starts. The first is routing - frontier models for planning and review, mid-tier for implementation, the cheapest tier for the mechanical grind, judged on cost times success rate rather than price per million. The second is the bigger one. An agent let loose on a large codebase spends your best model on searching instead of thinking. The fix is architectural: a code intelligence layer in front of your agents, where the expensive model asks the questions and a cheap, locally run system that already knows the codebase answers them.
What we get into:
* Why the cheapest model is usually the most expensive one, and the one metric that shows it * Plan-Build-Grind-Review: which model tier belongs to which phase of the work * Why long contexts degrade reasoning, and what that costs you before it costs you money * The code intelligence layer pattern: retrieval in front of the agent instead of inside it, and how you know it is working
Everyone in the room runs these tools. Almost nobody has decided how. You leave with a framework you can apply on Monday and a way to argue about it as architecture, not budget.
This event is a part of #SFTechWeek - a week of events hosted by VCs and startups to bring together the tech ecosystem. Learn more at www.tech-week.com.
