Proactive agents introduce new context-engineering challenges
In a LangChain Blog commentary, the author argues that agent architecture is converging on an LLM that runs in a loop, calls tools and operates in a sandbox, citing Claude Code and Deep Agents as examples. Based on conversations with agent builders, the author says proactive, always-on, goal-oriented agents introduce open context-engineering questions involving notification pacing, memory, asynchronous execution, goal tracking, alignment and browser use. The author and the broader industry may not yet have general solutions to these problems.
The commentary contrasts the relatively standard context-engineering practices used with coding agents, such as summarizing long threads, offloading large tool results and using subagents, with a new set of challenges arising from proactive agents.
It feels like everyone is converging on the same agent architecture: an LLM, running in a loop, calling tools, running in a sandbox. Claude Code - perhaps the first high profile manifestation of this - came out over a year and a half ago. Deep Agents, our agent harness, launched over a year ago. This pattern has been building for a while.
There’s nothing bad with that convergence per se, but it has led to a sentiment among some AI engineers that I’ve talked to recently that “all the hard problems are solved” or “there’s no new AI engineering problems to solve”.
When people express this sentiment, I think they are largely talking about the context engineering practices we do to make agents work. And to some extent they are right - there are a lot of standard-ish context engineering solutions we’ve all arrived at:
- Summarizing long threads when it gets to be too much for the context window
- Offloading large tool call results
- Skills for progressive disclosure
- Tool search to deal with a lot of tools
- Subagents for context isolation
- Bash as a primary tool