AI Agents Are Now a Daily Habit. Making Them Reliable Is the New Problem.

For most software teams, AI agents are no longer something to try out — they are part of everyday work. A new survey of more than 550 engineers, published this week by Temporal, found that 80.8% now use AI agents every single day, up from 47.3% just one year ago. That jump — nearly 71% in twelve months — is not gradual adoption. It is a fundamental change in how software gets built.

But daily use does not mean smooth use. More than four in ten engineers said they hit agent-related problems every day. Teams struggle most with three things: keeping track of what an agent was doing when something went wrong, debugging failures, and keeping costs in check. The shift from adoption to reliability is now the story. The teams pulling ahead are not simply the ones who moved fastest — they are the ones who treated those three problems seriously from the start.

Cost matters more than most teams expect. Nearly 80% of engineers say token costs are already limiting what they can do. Tokens are the tiny chunks of text that AI models read and write — and every model charges for them. A team running agents all day can rack up large bills fast, and right now those bills are cutting real projects short before they reach their potential.

The most striking number in the report: 92.3% of engineers have tried to build software they would previously have just bought. Agents are changing the build-versus-buy decision for small teams. A startup that once purchased a scheduling tool or an analytics product can now have one engineer build a working version in a week, with agents doing most of the heavy lifting.

If your team is not yet using agents every day, you are behind the typical engineering team. But speed alone is not the advantage — the businesses winning right now are those that treated reliability, cost controls, and clear ownership as first-class problems from day one.