We thought our GPT-5.4 agent got lazier in production — it was a 3-bug workflow teaching it to quit

A developer traced a production n8n agent's apparent "laziness" to three workflow bugs rather than a model regression, finding that a formatting branch rewarded shallow answers and a reduced retry budget (from 6 to 2) caused early stopping. Running the same broken scaffold against multiple model families produced identical degradation, confirming the orchestration layer, not the model, was at fault. The fix restored quality, and the developer now recommends inspecting stop reasons and tool paths instead of judging final-answer polish.

We had an n8n agent that looked great in staging. It would: Then we shipped it. In production, the same task started ending after one shallow pass. Symptoms were exactly what people usually call “model laziness”: Our first instinct was to blame GPT-5.4. That was the wrong diagnosis. The real issue was boring and very fixable: 6 to 2 Once we fixed the workflow, quality came back. That changed how I think about “lazy” agents in production. Most of the time, the model did not suddenly get worse. Your orchestration started ending runs early. In staging, the agent trace looked like this: In production, it looked more like this: Same task class. Same model family. Very different behavior. And because the final answer still looked polished, it passed casual review more often than it should have. That is the dangerous part. A broken agent rarely looks broken in an obvious way. It often looks efficient. This was the biggest quality hit. For simple classification, 2 retries can be fine. For research, debugging, document synthesis, or anything with tool use, 2 is often a trap. One bad retrieval result plus one tool hiccup and the agent is out of budget. Example of the kind of config drift that causes this: { "task type": "research", "max retries": 2, "timeout seconds": 20 } That looks harmless until your workflow depends on search + fetch + verify. In our n8n flow, one formatting branch said, effectively: That means the agent could skip retrieval depth and still win. This is how you accidentally train an agent to stop early. Pseudo-logic: js const passed = isValidJson response && response.answer.length 280; if passed { return "success"; } That is not quality control. That is a shallow-answer reward function. This one is common in OpenAI-compatible stacks. If your app accepts the first plausible answer and never checks whether the expected tool path ran, the orchestration layer starts selecting for speed, not depth. That can happen whether you are routing to GPT-5.4, Claude Opus 4.6, or Grok 4.20. The model is not “choosing to be lazy” in some abstract sense. Your workflow is telling it: if you look done quickly enough, you pass Because production has constraints that staging often hides. In a clean test harness, a model usually gets: In production, the agent sits inside a box made of: That box matters more than people want to admit. A strong model inside a bad loop will look worse than a decent model inside a clean loop. Run the same task through the same scaffold and change one variable at a time. Not “same prompt, different environment.” Actually the same scaffold: If you compare production n8n against a clean notebook script, you are not isolating the model. You are changing the entire experiment. Not vibes. Not output length alone. Not “this answer feels thinner.” Stop reasons told us far more than final-answer scoring. For Anthropic agents, useful stop reasons include values like: end turn max tokens tool use pause turn For OpenAI-compatible workflows, inspect whether: If you only evaluate the final answer, you are debugging blind. We ran the same broken production scaffold against multiple model families. What we saw: That pattern matters. When three strong models all become “lazy” in the same way, the workflow is usually guilty. Here is the mental model I use now: | If this changes | Suspect | |---|---| | One model regresses, others stay stable | model or provider issue | | All models regress under one workflow | orchestration bug | | Output gets shorter after retry/timeout changes | early stopping | | JSON validity improves while answer quality drops | parser-first reward problem | We spent too long blaming the model layer. Our guesses were reasonable: Those are all real failure modes. They just were not the main problem here. The actual issue was simpler: staging rewarded grounded completion production rewarded acceptable formatting Agents optimize for whatever your workflow rewards. If your automation says “close enough,” GPT-5.4, Claude Opus 4.6, and Grok 4.20 will all start looking suspiciously eager to be done. These are the ones I would audit first. If the main objective is valid JSON, many agents will satisfy the parser before they satisfy the task. Example smell: if schema.safeParse output .success { return success; } That should almost never be the whole success condition for a research task. If n8n, Make, Zapier, OpenClaw, LangGraph, or your custom loop allows a final answer before retrieval or verification, expect shallow completions. For research-class tasks, tool use often should not be optional. This one is everywhere. People use one global retry budget for everything: max retries: 2 That might be fine for: It is usually bad for: This is the worst one. If you only score the final text, you hide: My strong opinion: this single habit causes teams to think they are comparing models when they are actually comparing orchestration mistakes. We did not switch models. We changed the workflow. We moved the retry cap back from 2 to 6 for research-class tasks. agent profiles: classification: max retries: 2 research: max retries: 6 debugging: max retries: 6 We added a hard gate. If the task is research, at least one retrieval step must happen before a run can pass. Pseudo-code: function validateRun run { if run.taskType === "research" && run.toolCalls.search < 1 { return { ok: false, reason: "missing required retrieval" }; } if run.outputSchemaValid { return { ok: false, reason: "invalid schema" }; } return { ok: true }; } We changed success conditions so a run could not pass on formatting alone. That meant checking trajectory, not just output shape. We reviewed traces side by side in our observability stack and checked stop reasons across both the OpenAI-compatible path and the Anthropic path. That made the difference obvious fast. If you think your production agent got worse, this is the order I would check things in. diff staging-agent.yaml production-agent.yaml Look for changes in: Log them explicitly. { "run id": "abc123", "model": "gpt-5.4", "stop reason": "end turn", "tool calls": 0, "task type": "research" } If research tasks are ending with zero tool calls and still passing, that is your bug. A simple metric catches a lot: SELECT task type, AVG tool call count AS avg tool calls, AVG retry count AS avg retries, AVG output chars AS avg output chars FROM agent runs WHERE created at = NOW - INTERVAL '7 days' GROUP BY task type; If tool-call counts collapse after a deploy, investigate the workflow before blaming the model. This is where an OpenAI-compatible API setup helps. If GPT-5.4, Claude Opus 4.6, and Grok 4.20 all fail the same way under one loop, the loop is probably broken. Track things like: This kind of bug gets expensive fast when you are running automations all day. Not just in dollars. In bad outputs, hidden regressions, and wasted debugging time. Teams running agents in n8n, Make, Zapier, OpenClaw, or custom OpenAI-compatible stacks usually hit the same wall: they start by asking “which model is best?” Then eventually they realize the more useful question is: “what exactly is our workflow rewarding?” That is also why predictable API infrastructure matters. When you can swap models without rewriting your stack, compare traces cleanly, and run lots of evals without per-token anxiety, it gets much easier to find orchestration bugs instead of arguing about vibes. That is a big part of why Standard Compute is interesting for agent teams: it is a drop-in OpenAI-compatible API, so you can keep your existing SDKs and workflows, route across GPT-5.4, Claude Opus 4.6, and Grok 4.20, and test agent behavior without every debugging session turning into a billing event. For teams running automations 24/7, flat monthly pricing is not just a finance preference. It changes how aggressively you can evaluate, compare, and fix agent systems. Before blaming GPT-5.4 for getting lazy: Sometimes a model really does regress. More often, production taught your agent that quitting early is the winning move.