Drew Breunig · 2026-10-24 · notable
Drew Breunig — the state of context engineering in 2026
Drew Breunig traces context engineering from the 2024 long-context promise to 2026's long-running agents. He argues the engineer's job is to choose what goes in deterministic code and what is left to the model.
Drew Breunig argues that context engineering is the choice of when to be deterministic and when to be probabilistic.
What is it?
"The State of Context Engineering" is Drew Breunig's write-up of his AGNTCon talk "Context Fails & Fixes". It looks back at 2024, when context windows grew from about 128K tokens to 1M and then 10M, and the hope was that long context could replace retrieval and agent design. Breunig shows why that hope failed and what replaced it.
How does it work?
Long context brought four failure modes in Breunig's earlier work: context poisoning, distraction, confusion and clash. The essay lists six fixes — context augmentation, tool loadout, quarantine, pruning, offloading and summarization — and shows they fit into a DSPy agent with roughly 16 lines of changes. It then names two newer failures in long-running agents: "fighting the weights", where models trained on serial tasks resist parallel tool calls, and "context inertia", where an agent fixes on one hypothesis and keeps chasing it.
Why does it matter?
For teams building agents, the essay frames prompt engineering (2023), context engineering (2025) and harness engineering (2026) as one line of work, with more of the context moved into plain code each year. Breunig predicts that personal agents that keep collecting memories and skills will build up conflicting instructions, so planned forgetting becomes a key open problem.
Who is it for?
developers building agents and agent harnesses