A new Moore's Law for AI agents

The length of coding tasks frontier systems can complete is growing exponentially – doubling every 7 months.

This trend was discovered by researchers at METR. They took the most capable agents from 2019 to 2025, and tested them on about 200 tasks: mostly coding tasks, with some on general reasoning.

Then, they compared the agent's success rate to the length of each task – how long it takes human professionals to complete, ranging from under 30 seconds to over 8 hours.

Across all models tested, two clear patterns emerged:

1. Task length is highly correlated with agent success rate (R² = 0.83)
2. The length of tasks that agents succeed at 50% of the time – the time horizon – is growing exponentially

What comes next?

This exponential trend seems robust, and there's no evidence of plateauing.

Exponentials grow fast. Extrapolating out, this trend predicts:

• 2026: 2-hour tasks

• 2027: 1 work day (8 hours)

• 2028: 1 work week (40 hours)

• 2029: 1 work month (167 hours)

Recently, the trend has accelerated.

In 2024-2025, time horizons doubled every 4 months, down from every 7 months over 2019-2025.

If the faster trend continues, agents might reach month-long tasks in 2027.

However, looking at just one year's data gives a less robust estimate. The rate of progress might slow down.

It might also speed up. Given that the trend has already sped up, it could be on a growth trajectory that's faster than exponential. This fits intuitively: there might be a bigger gap in required skills between 1 and 2 week tasks, than 1 and 2 year tasks.

Additionally, as AIs improve they'll be increasingly useful for developing yet more capable AIs. This could also lead to superexponential growth in AIs' time horizons.

Increasingly capable AI systems could trigger a flywheel of acceleration – agents speeding up the creation of more capable agents, which speed up the creation of more capable agents.

From here, agent capabilities might skyrocket beyond any human's abilities in AI research – and across many or all other domains. The effects would be transformative.

If automating AI research leads to progress this fast, the rapidly increasing time horizon of AI systems might end up being one of the most important trends in human history.