What 500 technical leaders say about running agents in production

What happened

In December 2025, Anthropic published a survey run with the research firm Material. It covered more than 500 technical leaders across industries and company sizes.

The headline numbers: 57% of organizations use agents for multi-stage workflows, and 16% run processes that span several teams. About 80% said their agent investments were already producing measurable returns. Nearly 90% use AI for development work. Among those using agents for production code, reported time savings were roughly 58% to 59% across planning, writing, documenting and reviewing.

The obstacles are more interesting than the wins. Integration with existing systems topped the list at 46%. Data access and quality followed at 42%, and change management at 39%.

One caveat. Anthropic sells the tools it recommends, and the case studies are company-provided. Treat the numbers as a snapshot of how leaders describe their own results, not an audit.

So what for your business

Look at the three obstacles. None of them is "the AI isn't smart enough". All three are about the systems and people around it.

That fits how these projects tend to go. An agent that can reason about your invoices is no use if it can't reach the system where invoices live, or if half the records are incomplete. And a team that wasn't told how its role changes will quietly route around the new tool.

The early returns point somewhere practical, too. The highest-impact uses reported were data analysis and report generation (60%) and internal process automation (48%). That's back-office work, not flashy customer-facing bots.

What to do this month

1. Start with the integration, not the agent. Map which systems the workflow touches and whether each one has an API you can rely on.
2. Check the data. Pull 100 real records. If a person would struggle to act on them, an agent will too.
3. Name an owner on the business side. Not IT. Someone who lives with the workflow and can say whether the output is right.
4. Tell the team early. Explain what changes in their week and what doesn't.
5. Pick an internal workflow first. Reporting and process automation carry less risk than anything customer-facing, and they're easier to measure.