Temporal survey: 80.8% of engineers now use AI agents daily, but reliability lags capability

A new Temporal report of 554 engineers finds daily AI agent use jumped 70.8% year over year — yet tracking state, debugging, and managing token costs remain the top blockers, exposing a reliability gap that hits small operators hardest.

Temporal’s second annual State of Development Report: AI Agents, released Tuesday, finds that 80.8% of 554 surveyed engineers in the US and UK now use AI agents daily or more, up from 47.3% a year earlier, a 70.8% relative jump. The survey ran April 29 through May 25, 2026, with the largest cohort (29.2%) at companies of 251 to 1,000 employees. ChatGPT, Copilot, Gemini, and Claude lead adoption.

The capability story is unambiguous. 91.1% of respondents say agents have “improved” or “revolutionized” productivity, and 85.5% trust agent outputs at least somewhat. The reliability story is the opposite. 41.1% hit agent issues daily; 9.0% describe issues as “continuous.” Asked what limits productivity, 35.7% rank tracking state first, followed by debugging and managing token and compute costs. 79.8% call cost itself a limiting factor. Median deployment: five agents per engineer.

“Today’s engineers have adopted AI agents faster than most teams have built the infrastructure to run them reliably. The teams pulling ahead are those who trust their systems more, because they’ve solved for state, cost, and reliability,” said Samar Abbas, CEO of Temporal.

At VentureBeat’s AI Impact Series event on May 29, Temporal SVP Preeti Somal described customers “building version 2.0 of the same agent” after first-generation deployments crashed on state loss and runaway token bills, comparing the moment to enterprise lift-and-shift cloud migrations that skipped architectural redesign.

The Temporal data sits alongside Salesforce’s Agentic Enterprise Index, summarized in MarketingProfs’ August 21 update, which shows agents per organization nearly tripling from five to 13 between early 2025 and April 2026, with governance and measurement lagging deployment. Signal vendors like 6sense have started wiring buying intent directly into agent stacks, but consuming that plumbing still requires engineers.

Which is the awkward part for anyone below enterprise scale. If professional engineers running a median of five agents apiece can’t reliably keep them upright, a founder with no infrastructure team isn’t going to close that gap alone. That’s where done-for-you services like LemonLime position themselves: the operator picks priorities, and finished sales and marketing work arrives without owner-built infrastructure. Whether that trade holds up depends on the same reliability question Temporal is asking of everyone else.

Temporal survey: 80.8% of engineers now use AI agents daily, but reliability lags capability

Sources