THE CONTEXT
GRAPH
A working lab focused on production AI agent reliability.
Production is where agents break. We study silent failure modes in real-world AI systems: state drift, tool misalignment, decision gaps, trace collapse.
For teams actively running AI agents in production.
Prefer async? Drop your email for the newsletter.
Same ideas as Compare notes, in your inbox.
Production failures are predictable.
- Agents modify state through tool calls.
- Approvals disappear between systems.
- Context degrades across sessions.
- Logs don’t explain behavior.
- Failures stay invisible until they’re expensive.
Most teams optimize prompts.
Few teams model decision state.
This is for teams shipping real agents.
- AI-native SaaS
- Vertical copilots
- Automation platforms
- LLM infra teams
- Founders scaling agent workflows
Not for
- Prompt experiments
- Academic prototypes
- Hobby projects
What we're exploring
Decision Architecture
How to structure agent decisions so they can be validated, replayed, audited, and simulated.
State & Drift
Why agents degrade even when prompts stay constant.
Beyond Retrieval
Why embeddings don't model execution state.
Every Multi-Agent Framework Ignores the Same Problem
Read →The Vocabulary Problem in Agent Infrastructure
Read →MCP Solved the Pipes. Who Solves the Water Quality?
Read →Context Engineering in 2026: From Karpathy's Tweet to Production Infrastructure
Read →How Context Graphs Prevent the 7 Silent Agent Failures
Read →Production AI systems need structural reliability.
We're comparing notes with teams scaling real deployments. No sales pitch. Just architecture.