StarvardYour First Coworker Might Be a Robot: Using AI Without Handing It Your Job
Somewhere in the last few years, millions of workers quietly gained a new coworker. It is fast, it is confident, it never sleeps, and it is wrong more often than it admits. The question for your career is not whether this coworker is coming to your workplace. It is already there. The question is whether you will be its supervisor or its assistant.
The good news: the early research says the robot is best understood as a very fast, very careless intern. Treat it that way and it can make you better at your job. Treat it as an oracle and it will eventually embarrass you in front of people whose opinion matters.
What the studies actually found
The clearest evidence comes from a study by Erik Brynjolfsson, Danielle Li and Lindsey Raymond, who watched what happened when a large software company gave a generative AI assistant to 5,179 customer support agents. As MIT Sloan summarized the work, agents with the tool resolved about 14 percent more issues per hour. The striking part is who gained. New and lower skilled agents improved the most, resolving around 35 percent more chats per hour, while the most experienced agents barely moved. The tool seemed to carry the hard won know how of the best agents, their phrasing, their patience, their sense of what to say next, and hand it to the newest person on the floor.
Two details deserve attention. Agents followed the AI's suggestions only about 38 percent of the time, choosing for themselves the rest. And customers noticed the difference: requests to speak to a manager fell by about 25 percent.
Writing shows the same pattern. In a study published in Science by Shakked Noy and Whitney Zhang, 453 college educated professionals did realistic writing tasks, half of them with access to ChatGPT. The group with the tool finished about 40 percent faster, and independent graders rated their work about 18 percent better. Again, the weakest writers gained the most, and the gap between strong and weak performers shrank.
The intern rule
Read those numbers the way a manager would. The gains were real, and they were largest for beginners. The tool did the most good in the hands of people who still made their own calls, accepting roughly a third of its suggestions and overruling the rest.
That is exactly how you would treat a brilliant, careless intern. The intern drafts. You decide. The intern brainstorms twenty options. You pick the two worth having. The intern summarizes the document. You check the summary against the document before it leaves the building.
Danielle Li put the risk plainly in MIT Sloan's account of the study: if you are just typing the recommendations, you may not necessarily be learning. You are more productive, but the source of your productivity lives in the technology, not in you. Productivity you do not own can be taken away, repriced, or handed to everyone else until it is worth nothing. Skill you own travels with you to every job you will ever have.
The skills that gain value
When drafting gets cheap, four human skills get more expensive.
Judgment. Knowing which of the twenty options is right for this customer, this boss, this moment. The studies show the tool proposing and the human disposing. That disposal is the job.
Taste. The difference between correct and good. AI output tends toward the average, a smoothie of everything it has ever read. Taste is how you notice that something is technically fine and still wrong for the room.
Verification. The intern invents facts with total confidence: names, dates, quotes, numbers. Checking its work is now a core professional skill, and it requires knowing the territory well enough to smell an error.
Knowing the work itself. Every one of these skills rests on the same foundation. You can only supervise work you understand. The agents who gained least from the tool were the experts, and that is not an insult to them. It is the receipt showing what expertise is worth.
Six rules for working with the robot
- Let it draft, never let it decide. First passes are its job. Final calls are yours.
- Verify every name, number, date and quote it gives you, every time.
- Give it the boring first pass and keep the meaningful last pass for yourself.
- Never send anything you could not defend out loud in a meeting.
- Keep private and company information out of tools you do not control.
- Spend the time it saves you on learning the work more deeply, not just on doing more of it.
Why this belongs at Starvard
Every generation of workers has had to learn a new tool, and the learning has always been the unfair part: expensive, slow, or locked behind an employer. A free university exists to remove that lock. The research above says the people who gain most from AI are the ones with the least experience, which is another way of saying the people Starvard was built for.
Start with our free course on artificial intelligence to understand what these tools are and are not, then go further with AI agents when you are ready to put the intern to work properly.