AI agents have no sense of time and are not aware of it

AI agents have no sense of time and are not aware of it

A new study finds that popular coding assistants can't predict how long a task will take, and they can't reliably tell how long they've already been working. That's a problem for long-running jobs.

When an AI assistant works on a task, it often has no idea how much time is passing. That's the takeaway from a study by two independent AI researchers, done as part of theMATSresearch program. The pair tested two widely used coding assistants, Anthropic's Claude Code and OpenAI's Codex, on their sense of time.

Before each coding task, the agents had to estimate how long they'd need. Then they solved the task and, looking back, reported how much time had passed. The test material came from 200 tasks in a collection called ProgramBench, plus the researchers' own suite of 18 benchmarks.

In the tests, the agents consistently overestimated how much time they'd need. On ProgramBench, both models mostly guessed around 90 minutes, no matter the difficulty. In the second round, Claude was off by three times on average, Codex by six to ten times. The estimates were worst for short tasks, and only in the multi-hour range did some predictions come close to reality.

Coding agents badly misjudge how long tasks take. Points above the dark diagonal mark overestimates: Fable 5 came in about three times over the actual runtime on average, GPT-5.6 Sol about seven times over. The gap is especially wide on short tasks. Source: Ofengenden/Andriushchenko,LessWrong

The same AI behaves completely differently depending on its setup

The results shift based on the software setup the models run in. Claude Code keeps working until it thinks the task is done, a median of about 90 minutes. Codex, on the other hand, stops after roughly half an hour, almost regardless of the task. According to the study, the same language model takes 2.5 times more steps in Claude Code than in Codex on average. So runtime depends on the model and heavily on the surrounding software, known as the harness.

The agents are just as unreliable at judging the quality of their own work. The older models, Opus 4.8 and GPT-5.5, overrated their results by 20 points on average and handed themselves high marks even on failed tasks. In one case, both figured their work was about 70 percent successful. The actual scores were 7 and 14.5 percent.

Claude and Codex rate their own work far too generously. On average, both systems scored themselves about 20 percentage points above the results they actually hit on the tests. The dashed line marks an accurate self-assessment. Source: Ofengenden/Andriushchenko, LessWrong

The researchers say this ability to self-assess matters. For an agent to work reliably on long tasks that run for hours, it has to follow instructions like "iterate on this task for two hours." An agent that constantly misjudges the time is hard to control. Next, the authors want to test whether agents can stick to a set work duration. When the agents got access to a tool that reports elapsed time, they got it right almost every time.

AI News Without the Hype – Curated by Humans

Subscribe to THE DECODER for ad-free reading, a weekly AI newsletter, our exclusive "AI Radar" frontier report six times a year, full archive access, and access to our comment section.

Read on for the full picture.

Subscribe for hype-free coverage.

• Full access to every article on THE DECODER

• Join the comments and community discussions

• A weekly AI news recap via mail

• 6x/year: "AI Radar" — deep dives on the AI topics that matter most

• Daily AI news, always up to date

• Our full ten-year archive

• Covered by a team with 10+ years in AI