A new survey from Temporal Technologies paints a clear picture of where AI agents stand in real engineering organisations today: adoption has gone vertical, but infrastructure is not keeping pace.

The 2026 State of Development Report surveyed 554 engineers and engineering leaders across the US and UK between April and May 2026. The headline number is striking — 80.8% now use AI agents daily or more. A year ago that figure was 47.3%. That is a 70.8% relative leap in frequent use in twelve months.

For business leaders still treating AI agents as a speculative investment, these numbers are not from a vendor white paper projecting future adoption. They are from practitioners actively running agents today.

How Many Agents Are We Talking About?

The scale of deployment is also growing faster than most organisations anticipated.

The median respondent runs five AI agents. The average is 10.7. A meaningful portion of the survey reports running well over 100 agents across their stack. This is not experimental use — these are agents embedded in daily workflows, handling real tasks with real business stakes.

That growth curve creates a practical problem. Most organisations built their operational infrastructure before agents arrived. Deploying ten or a hundred agents through infrastructure designed for human-controlled software processes was never going to be seamless.

The Reliability Gap Is Real

Adoption enthusiasm and operational confidence are not the same thing.

91.1% of respondents say AI agents have improved or revolutionised their productivity. 85.5% trust agent outputs at least somewhat. Those are strong endorsements.

At the same time, 41.1% encounter agent-related issues daily or more. 9.0% report problems continuously.

The pattern this creates is one that Temporal’s CEO Samar Abbas described directly: “Engineers have adopted AI agents faster than most teams have built the infrastructure to run them reliably.”

This is not a fringe problem. It is the majority experience. Agents are delivering value and causing friction at the same time, because the tooling to observe, manage, and recover from agent failures is still catching up to the pace of deployment.

Failures in agent-based workflows are different from traditional software failures. When an agent fails mid-task, the state of the work is often unclear. Did it complete? Did it partially complete? Does the next step in the workflow need to compensate, restart, or wait? Without durable execution infrastructure, these questions often get answered the hard way — by humans investigating what happened after the fact.

What This Means for Business

The 80% daily use figure will likely be used to justify accelerating AI agent investment. That is a reasonable response. The reliability data should inform how that investment is structured.

Buying agent tooling without investing in the infrastructure to run agents reliably is the pattern that produces the 41% who hit problems daily. It is the pattern that makes productivity gains feel inconsistent — impressive when things work, frustrating when they do not.

A few things stand out from this research for business decision-makers:

The adoption gap is widening. Teams that have figured out agentic workflows are compounding their advantage. Teams still on the sidelines are not just falling behind on capability — they are falling behind on the operational learning that comes from running agents in production.

Infrastructure is the real bottleneck. The productivity gains are real, as 91% of practitioners report. The question is whether your organisation’s systems can support agents running at scale without constant manual intervention to recover from failures.

Skills are a constraint too. The report notes genuine concern among practitioners about agents eroding human skills. Organisations deploying agents without upskilling their teams are creating a dependency they may not fully understand yet. The people who know how to work effectively with agents, not just alongside them, are becoming a scarce resource.

For organisations ready to move agents from experiments into production, the combination of the right tooling and the right team capability is what separates deployments that compound in value from ones that plateau at limited use cases.

If you are assessing where AI agents could have the fastest impact in your business, or want to understand how to build the infrastructure to run them reliably, our AI advisory team works through exactly these questions with business leaders across industries.

For teams looking to build the internal skills to work effectively with agents, the Enterprise DNA learning platform has structured training on AI tools, data literacy, and the practical skills engineering and analytics teams need to get the most from their investments.