A coding agent has been working through a long refactor for hours. Its transcript -- every file it has read, every edit it has made, every command's output -- keeps growing with each step, and it is now approaching the model's context window limit even though the refactor is not finished. Rather than letting the request fail once the limit is hit, the harness periodically has the model condense the older parts of that transcript into a shorter summary of what has been done and what still remains, keeping the most recent, most relevant steps in full and folding the rest down. What problem does this address?
- That the model is physically unable to read more than a few hundred tokens of any tool output, regardless of how the transcript is managed
- That every tool call an agent makes must be executed twice, once for the model and once for a human reviewer, unless the transcript is condensed first
- That an agent's own working transcript keeps growing with every step it takes, and left unmanaged will eventually exceed even a large context window; periodically condensing older, lower-relevance steps down to a summary keeps the most useful information within the window instead of letting the request simply fail once the limit is reached
- That retrieval-augmented generation is required before any agent can be given more than one tool, and condensing the transcript is how that retrieval step is implemented
Why C? And why not the others?
Correct answer: C. That an agent's own working transcript keeps growing with every step it takes, and left unmanaged will eventually exceed even a large context window; periodically condensing older, lower-relevance steps down to a summary keeps the most useful information within the window instead of letting the request simply fail once the limit is reached
As an agent loop runs for longer, the transcript of everything it has read, done, and observed keeps accumulating within the same context window, and unlike a fixed document that starts and stays a known size, this working transcript has no natural ceiling short of the task simply ending -- so a harness that wants to keep going past the point a raw transcript would overflow the window needs to actively curate what stays in full context versus what gets condensed into a shorter summary, treating the window as a limited resource to be filled with the highest-signal information rather than an ever-growing raw log. The option claiming the model can only read a few hundred tokens of any output is wrong and describes a much smaller, fixed limit than modern context windows actually have; the real problem is the never-ending growth of an ongoing session, not a tiny per-call cap. The option claiming every tool call must run twice is wrong and unrelated to transcript size. The option tying this to retrieval-augmented generation is wrong: condensing a transcript that already exists in-session is a different mechanism from retrieving external documents, and nothing about giving an agent multiple tools requires RAG.
Source: Anthropic, 'Effective context engineering for AI agents' (anthropic.com/engineering/effective-context-engineering-for-ai-agents) -- describes treating the context window as a finite resource, agents summarizing completed work and curating the highest-signal tokens as sessions grow long