Live building of agents s02e01 "Everything about context"
Details
Topics: context engineering, context optimization, context compaction, context rot, optimizing tools for context efficiency, retrieval strategies, prompt caching
Hey everyone,
in the next episode of our second season we're going deep into the one resource every agent depends on and keeps running out of: the context.
What is it about
Models keep getting smarter and context windows keep getting bigger, yet agents still lose track of their task, drown in tool outputs, forget important details halfway through, or burn through tokens (and money) for no good reason. More often than not, the difference between an agent that works and one that doesn't is not the model, but what ends up in its context window.
- What is context and how it works
- How do we give the agent exactly the information it needs, at the moment it needs it, and nothing more?
- When should we inject context upfront, and when should the agent fetch it on its own?
- MCP or CLI
- How do we keep long-running agents going with context compaction, without them forgetting what matters?
- How to use prompt caching properly to optimize costs
We will look into context optimization techniques, compare different approaches side by side, and see how they affect agent performance, reliability and cost.
For whom it's relevant
The event will be mostly interesting for people building agents, people heavily using agents and people curious about how agents work under the hood.
Hands-on approach
As usual, this meetup is hands-on: participants will be building, running and tweaking agents that need to solve problems in a context-efficient way.
- Bring your laptop (Wi-Fi will be provided)
- Agentic coding: you don’t need to be able to code yourself, but you should have a coding agent (Claude Code, Codex, opencode) or be ready to install it.
- Previous agent building experience helps, but is not required.
Community