Code-writing assistants often behave like a chef who, to boil a single egg, rereads the whole cookbook. When asked to make a tiny fix, they reopen dozens of files and retrace long-familiar connections. This is unnecessary work — energy going to waste.
Researchers proposed the E3 method: first gauge the actual complexity, perform minimal actions, and broaden the search only upon failure. The algorithm resembles an experienced chef who cooks an egg from a tried-and-true recipe, without wasted motion. It's a smart approach instead of panic.
The goal isn't to test a robot, but to build engineering-sound AI that doesn't inflate tasks like a black hole devouring everything around it.
as written by the authors arXiv:2607.13034
Large language model (LLM) agents increasingly automate multi-step engineering and informatics workflows, yet they rarely ask how much effort a task actually requires. They often follow a maximum-context-first strategy--re-reading files and dependencies they have already seen--turning a one-line edit into a small code-base audit. We argue the missing capability is task-aware execution-scope estimation: judging a task's difficulty, the information it truly needs, and the shortest reliable path before committing budget. We formalize minimum-sufficient execution and the Agent Cognitive Redundancy Ratio (ACRR), and propose E3 (Estimate, Execute, Expand): the agent estimates an initial operating point, executes a minimum viable path, and expands scope only when verification fails. On MSE-Bench--a deterministic benchmark of 121 edits in a capability-controlled simulator--E3 matches the strongest baseline's 100% success while cutting cost by 85%, tokens by 91%, and inspected files by 92%, and further beats a strong adaptive retrieval baseline by 16%; the gains survive held-out instruction wording and essentially every cost weighting. A companion real-model harness (LLM-Case) corroborates the effect on a live gpt-4o agent editing a real open-source library, with every candidate patch graded by actually running the project's real pytest suite against a measured oracle: the over-reading is milder but real, and E3 is the leanest and fastest policy at comparable task success--its one shortfall a provider rate-limit, not a wrong edit. We frame this as a controlled probe of execution redundancy, not a measurement of any deployed agent, and position task-aware execution as a step toward engineering-grounded AI (EGAI)--agents whose effort is anchored in the engineering reality of the task. We release the framework and benchmark.
Without the E3 method, the agent examined 12 times more files than necessary for the simplest edit.