What AI agents can actually do for a business

Anthropic's 2026 State of AI Agents report surveyed more than 500 technical leaders on where agents are already paying off. Data analysis and report generation topped the list: 60% named it one of the most impactful agentic tasks. Internal process automation came next at 48%, then managing and summarising internal knowledge bases at 41%, then resolving complex customer queries at 35%. None of that is glamorous. It is the backlog work that normally sits on one person's desk until it gets done, which is exactly the kind of task an agent can run start to finish without someone babysitting every step.

The same survey breaks results out by company size, and the gap matters more than the averages. Startups and small businesses expect the biggest near-term impact from customer service (52%) and rate data analysis and reporting just as impactful as everyone else (60%). But they plan to lean on agents for research and reporting at only 44%, against 57 to 58% at mid-market and enterprise firms, and expect software development gains at 41%, well behind mid-market's 60%. The pattern is consistent: smaller companies are betting on the agent that answers a customer and the one that compiles a report, not the one that writes code or models supply chains, because those two jobs are the ones a small team already does by hand every week.

Mostly real, but smaller companies are seeing less of it so far. Across all company sizes, 80% of the surveyed leaders said their AI agent investment had already delivered measurable financial impact. Among startups and small businesses specifically, that figure drops to 72%, the lowest of any segment, against 83% at mid-market firms. Looking ahead, the confidence gap narrows: 79% of startups and SMBs expect measurable financial impact in the future, versus 93% at enterprise firms. Read plainly, that says the technology is not the blocker. Smaller teams are earlier in deploying it, not worse at using it.

Start where the data says small businesses already see the clearest return: a customer service agent that resolves a real share of incoming questions, or a reporting agent that pulls numbers out of the spreadsheets someone currently assembles by hand once a week. Skip the instinct to match what a larger company is doing with coding agents or supply chain optimisation; that is where the size gap is widest, and it is not where your business feels the most pain today. Pick the one recurring task that already eats a person's week, point an agent at it, and measure whether it actually finishes the job before widening its scope.