In September 2026, Bolt.new made its first acquisition: Dokai, a San Francisco startup whose AI agents did something rare. They finished work. Multi-step, unattended, inside live enterprise systems: prospect research in, updated CRM out, nobody nudging the process along. That bar, finished work instead of a good demo, is the one that matters, and it's the same bar a small business should hold every agent to.
What is an autonomous AI agent?
An autonomous agent is software that takes a goal in natural language, plans the steps, and completes them across your systems without supervision. The test is simple: did the work get done while you were doing something else, and can you check it afterward?
Here is how an agent differs from the tools it gets confused with:
| Tool | What it does | What you get back |
|---|---|---|
| Chatbot | Answers questions | An answer |
| Workflow automation | Follows a fixed script | The same steps, every time |
| Autonomous agent | Plans and completes multi-step work | Finished work you can check |
Dokai's agents cleared that bar in one of the hardest rooms there is: enterprise sales operations. Given a plain-language instruction, they ran prospect research, scored and tiered accounts, enriched the records, and updated the CRM, end to end. That standard is why the Dokai team now builds agents at Bolt.
What can AI agents do for a small business?
For a small business, agents fit the work you repeat. Every operating business carries a layer of it: chasing overdue invoices, turning inbound emails into quotes, checking stock against reorder points, assembling the Monday numbers from three different tools. None of it is hard. All of it eats the week. That layer is agent work, because it has a clear goal, a known process, and a result you can verify at a glance.
Large companies are already moving. In McKinsey's 2026 State of AI survey, 40% of large organizations report scaling AI agents, up from 27% a year earlier, while smaller organizations held flat at 22% (McKinsey, 2026). The gap is headroom. What kept agents an enterprise project, integration teams, orchestration platforms, six-figure pilots, is the part that's collapsing. Among Bolt's small-business builders, 93% aren't professional developers (Bolt first-party data, 2026), and they ship working software every day. Agents that run that software are the next step.
The practical move is small: pick one repeating job, give an agent the goal and the checkpoints, and keep the review step for yourself. An agent that drafts every quote and queues it for your approval saves most of the hour and none of the control.
How can an agency use AI agents?
Agencies feel the repetition hardest because they run the same work across every client. In the US, 84% of advertising and PR businesses have no employees at all (U.S. Census Bureau, 2023). For a one-person business, the monthly reporting cycle, campaign pacing checks, and status emails compete directly with billable hours. An agent that assembles the client report, flags the pacing problem, and drafts the status note is the staff you never hired.
The economics compound in a way solo work never has. Build the reporting agent once, and it runs for every account. Add a client, and the agent scales; your evenings don't have to.
Why do so many AI agent pilots never ship?
Most stalled agent projects fail the same way: they perform in controlled conditions and fall apart in production. That pattern is how "agent" earned its demo reputation, and the enterprise record shows both halves of it. Gartner expects 40% of enterprise applications to carry task-specific AI agents by the end of 2026, up from under 5% in 2025 (Gartner, 2025), and the McKinsey scaling numbers above show where the money is going.
The agents that ship are built against live systems, verified on finished work, and reviewed like any other employee's output. Bolt's enterprise customers tend to skip the ceremony: roughly half start building before their contract is signed, and the average customer deploys its first app within 34 days (Bolt first-party data, 2026).
Build agents, don't just buy them
Buying an agent means finding a vendor whose product happens to match your workflow. Building one means describing your workflow and getting software that matches it. In Bolt.new, an AI app builder, the same natural-language conversation that creates an app extends to the agents around it.
As of September 2026, Dokai's orchestration technology is being folded into the platform: the machinery that let agents plan and complete multi-step work unattended. Gerry Fernando Patia and the Dokai team now run it from inside Bolt's AI and machine learning organization, and the first capabilities built on Dokai technology arrive in late fall 2026.
Agents can do more than you may think at first pass. Building them clears tedious, repeatable work from the to-do lists of your entire staff, freeing them up to do higher-value work.
Describe the work. Bolt builds the software around it. Start building at Bolt.new.
