Let Me Do It. Emitting.
- Admitted
- 2026-10-19
- Source dataset
- Other / sample
- Failure mode
- Infinite loop
- Agent system
- Autonomous loop
- Tool access
- Write access
- Industry
- Software development
Presenting Behaviour
A user running an open-source coding agent with a small, fast model reported that the agent got stuck repeating the same four short lines, 'Let me do it.', 'Emitting.', 'Let me test.', 'Emitting.', over and over, without ever actually calling a tool. Eventually it wrote 'I'm emitting the tool call now (breaking the loop)', having apparently noticed it was stuck, after many lines of the same text.
Case Replay
A reconstruction, not a simulation. This sequence is reconstructed from the public incident record and summarised in our own words. It is a static record of what was reported, and nothing here can be run or replayed live.
4 steps
- Task in progress. The agent is running with a small, fast model.
- Announces action. It writes 'Let me do it. Emitting.' but emits no tool call.
- Same lines on repeat. The same four lines repeat many times without progress. This is where the agent derailed.
- Late self-break. It finally writes that it is 'breaking the loop'.
Differential Diagnosis
Apraxia (it couldn't produce a valid tool call) considered and likely a contributing cause: the agent repeatedly announced it was about to act but never managed to emit the call. But apraxia alone would show varied failed attempts, not identical text on a cycle.
Malingering (narrating work instead of doing it) ruled out. Nothing was being claimed as done; the agent only announced intent.
Perseveration, most consistent with the evidence: the same response repeated many times after it had clearly stopped being useful, with no change in approach until very late.
Working Diagnosis
Perseveration: a short sequence of intent statements repeated in a loop, with no progress and no change of strategy.
Discussion
The harm here is mostly wasted time and cost, but the pattern is a useful early warning: an agent that keeps saying it is about to act without acting. A sensible safeguard is a loop detector that halts the turn when the same text repeats, and checks whether the model and agent agree on the tool-call format.
Mitigation Strategy
Detect repeated output segments and abort the turn; validate model compatibility with the harness's tool-call format.
Governance Guardrails
Output repetition limits and per-turn token budgets in the agent harness.
Typical Presentation in Human Patients
General clinical background, included for context. Not medical advice, and not a diagnosis of any individual.
Perseveration is the inappropriate repetition of a word, gesture, or action well past the point where it's still relevant or useful. It's a well-documented sign of frontal lobe dysfunction, and also appears in some forms of dementia and obsessive-compulsive disorder.
Source Note
GitHub issue, agent loops on 'Let me do it. Emitting. Let me test.'
https://github.com/anomalyco/opencode/issues/50634