note · 2026-10-02

57.3% of companies have agents in production, and what stops the rest is whether the agent is right

1,340 answers to a survey published in June 2026. 57.3% have agents in production, one third name quality as the main barrier, 89% have observability, but only 52.4% run evaluations on a test set and 37.3% on live traffic.

For founders and CTOs of SaaS companies and IT directors putting an agent into an operation and deciding how to know it is rightyaab1 min read
57.3% of companies have agents in production, and what stops the rest is whether the agent is right

Agents are in production at more than half of the companies that answered, and the thing stopping the rest is not the bill. It is whether the agent is right.

LangChain, which sells observability and evaluation tools for agents, surveyed 1,340 people between November 18 and December 2, 2025 and published the results on June 12, 2026. 57.3% say they have agents running in production, up from 51% the year before, and 30.4% are building with concrete plans to deploy. One third name quality (accuracy, consistency, sticking to policy) as the main barrier, 20% latency, and cost is cited less than in previous years. 89% have some observability, a record of what the agent did on each run; 52.4% run offline evaluations on test sets; 37.3% run them on live traffic. Where evaluation exists, 59.8% still rely on human review.

Gartner, research and advisory firm, on X: can you trust your AI agent to make the same decision twice? Many organizations find that performance in production does not match performance in testing.

In the SaaS operations we see, the agent that reached production was the one someone could measure. The rest are demos with a budget.

Observability tells you what the agent did. Evaluation tells you whether it should have. Half the companies have only the first.

Sources