Scott Bell is a bankruptcy lawyer, and until recently he treated AI the way most owners do: a useful assistant that drafts things while you stay in charge of everything. Then he tried agents — AI that executes whole workflows on its own — and his week stopped looking like a lawyer's week.
His story ran in The New York Times and Entrepreneur this week. The agents now handle his client intake, answer routine inquiries, and organize financial documentation. Over time he let them take on more: triaging which emails are urgent, flagging legal issues, drafting preliminary case summaries. In some cases the agents message clients directly, asking follow-up questions and collecting the details a case needs — without Bell touching it.
His explanation for why it works is refreshingly unglamorous: “With bankruptcy, a lot of the work is forms-based,” he told the Times. That's the honest version of most businesses. A big share of what fills your week — intake, follow-ups, sorting documents, routine replies — is forms-based work wearing different clothes. Which is exactly the work agents are good at.
The shift: you assign goals, not tasks
The platform Bell uses lets him hand over an outcome — onboard this client, prep this file — and the system breaks it into tasks and distributes them across specialized agents. That's the structural change worth paying attention to. With a tool, you give instructions and check the output. With agents, you give a goal, and the system plans its own steps.
Bell described the experience as both efficient and unsettling. Hours of administrative work finished in minutes, sometimes faster than he could follow. And that speed forced a change in his role he didn't plan for: instead of doing tasks, he was supervising outcomes — reviewing decisions made by a system working on its own plan.
The owners who win with agents won't be the best prompters. They'll be the best managers — clear job definitions, real reviews, early catches.
Why this matters even if you're nowhere near agents
Here's the gap this story exposes. Goldman Sachs surveyed thousands of small businesses this year: 76% now use AI, but only 14% have integrated it into core operations. Almost everyone has the tools. Almost no one has made AI responsible for a real workflow. The missing piece isn't a better subscription — it's the management layer: knowing how to define a job, hand it off, and check the work.
And agents raise the stakes on that skill. When AI only drafts things, a bad output costs you one edit. When AI executes end-to-end, an early mistake propagates — a wrong assumption in the intake shows up in the summary, the reply, and the file. The oversight question changes from “is this paragraph right?” to “is this system doing the job I actually gave it?”
How to practice managing before you hire the agents
Pick one forms-based workflow. Something repeatable with clear inputs and outputs: new-client intake, quote follow-up, weekly reporting. Not your judgment calls — your paperwork.
Write it as a job description, not a prompt. What comes in, what should go out, what good looks like, and what the AI should never do without you. If you can't write the job down, an agent can't do it — and honestly, neither can a human hire.
Review it like a new employee's work. Bell didn't hand over case summaries on day one. He started with intake and paperwork, checked everything, and expanded the job as the work held up. Two weeks of supervised runs will tell you what to delegate next — and what to keep.
The technology is arriving either way. The Times piece isn't about a Silicon Valley startup — it's about a bankruptcy practice. The owners who'll be ready aren't the ones waiting for agents to get better. They're the ones already managing their AI like it works for them. Because soon, it will.