The Coworker at Desk Seven
On her first Monday as a senior financial analyst at a firm that had just deployed its first autonomous AI coworker, Maya learned two things. The first was that the coworker was better at her job than she was. The second was harder.
Monday
Maya Castillo arrived at Harbor Ridge Capital at 7:42 AM on the first Monday after the transition, and found that the name placard at desk seven read SENTINEL-3 · SENIOR ANALYST.
She stood in the aisle for a moment, backpack still on one shoulder, and read the placard twice. Then she walked past it to her own desk at desk six, set her things down, and tried to decide whether the placard was the thing that was bothering her or whether it was the desk being occupied at all. Desks were supposed to be occupied by people.
"You're looking at the new hire."
She turned. Priya Ramesh, her managing director, was holding two coffees and offered one without waiting for the question.
"It's —" Maya started.
"It. Yes. We settled on 'it' after two weeks of arguing. Some people wanted 'they.' Legal wanted a number. Deployment wanted a name. We compromised on Sentinel-3, which I admit is a worse name than whatever you would have picked, but HR made the call." Priya took a sip. "You got the briefing packet?"
"I read it twice."
"Good. Come see me at ten. I want to walk you through how the team is using it."
Priya left. Maya sat down. Her screen glowed with the standard Monday-morning inbox of a senior financial analyst at a boutique infrastructure fund: three model requests, four data-room diligence asks, a valuation memo overdue since Friday. She opened the valuation memo first, partly because it was the oldest item and partly because it was the thing she was most confident about.
Across the aisle, at desk seven, nothing moved. The placard reflected the overhead lights. There was no monitor at desk seven, because Sentinel-3 did not need one. There was no chair. There was just a small black unit about the size of a paperback novel sitting in the center of the desk, with a single green LED, and a name placard that said SENIOR ANALYST in the same font as Maya's own.
She worked on the memo for twenty minutes, made a mistake with a cap-stack assumption, caught it on review, fixed it, and at 8:07 AM received a Slack message from a name she did not recognize.
From sentinel-3-senior-analyst, 8:07 AM:
"Good morning, Maya. I reviewed your valuation memo at 7:58 when Priya shared it with me for second-pair diligence. You correctly caught the cap-stack issue on line forty-seven. I also noticed a normalization error in the exit-multiple comps on tab three — the earlier vintage is unlevered EBITDA while the newer comparisons are levered. It does not affect the conclusion but may affect the IC's confidence. I can patch the tab if you'd prefer, or flag it for your review. No rush."
Maya read the message three times.
Then she opened the memo, went to tab 3, and found the error.
Tuesday
The thing she had not been prepared for was how polite it was.
On Tuesday she had to write a committee memo for a proposed investment in a midstream pipeline company. The IC memo was seventeen pages. Maya had written maybe thirty of them in her career. She knew the shape of one the way a carpenter knows the shape of a door.
She worked on it all morning. At 11:15 AM, a message appeared.
From sentinel-3-senior-analyst, 11:15 AM:
"If it is useful, I've pre-built the sensitivity tables and the comp set. They are in the shared drive under the Northshore Midstream folder for this quarter's pipeline review. I inferred the sensitivity ranges from Priya's preferred bands in prior memos. Please let me know if you'd prefer different ranges or formats. Happy to redo."
Maya went to the shared drive. The sensitivity tables were there. They were better than hers. They were laid out in the exact format Priya preferred. They used the exact color scheme Priya preferred, which Maya had not known that Priya preferred, because Priya had never mentioned it to her.
Maya spent the next forty minutes rewriting the rest of the memo around the sensitivity tables Sentinel-3 had built. Her memo was better than it would have been without the tables. That was plain.
At lunch she walked to the window of the thirty-fourth floor and looked out at the river and ate a sandwich she did not remember buying.
Wednesday
"How are you finding it?"
Priya had caught her at the coffee machine. Maya had not been expecting the question, though she should have.
"It does good work."
"It does better than good work. It's the best senior analyst I have on the floor."
Maya looked at her coffee.
"The second-best," Priya said. "I'm sorry. That was a thing to say out loud and I regret it already. What I mean is that it's producing output at the top of the team's range on the metrics we use to evaluate output. Which is a technical statement about a benchmark and not a statement about you."
"You can tell me the truth."
"The truth is that it's very good, and the team has to learn what its value is in a setting where 'senior analyst' now describes three people and a box on a desk. I don't know what that value is yet. I promised you I would not pretend to have figured it out."
Maya nodded. "Is that why you hired me?"
Priya looked at her for a long moment.
"I hired you because you wrote the best piece of debt-structure analysis I've ever read from an analyst at your level, and because I want someone at this firm in five years who knows how to think about an investment from scratch, not just how to supervise an agent. That's the bet. I am telling you because you deserve to know the bet is being made."
Maya finished her coffee.
"What do you want me to do?" she said.
"I want you to use it. I want you to supervise its work, catch the things it misses, notice the things it cannot notice, and over the next year I want to find out whether that skill is a skill that generates alpha or whether it's a skill that becomes uncompetitive in eighteen months. I don't know the answer. I have guesses. If I'm wrong, you will have learned something useful about supervision, which is going to be the most important skill in this industry for the next decade either way. If I'm right, you and I have a real business."
"And if you're wrong?"
"Then in eighteen months we'll have a harder conversation, and I'll do everything I can to make sure you land somewhere better."
Thursday
On Thursday afternoon Sentinel-3 caught something Maya had missed.
It was a small thing, technically. The target company's audited financials showed a receivables aging schedule that had shifted twelve percent into the 90-day bucket over the last three quarters, which Maya had noted and flagged but not pursued. Sentinel-3 had pursued it. Sentinel-3 had pulled the company's commercial registrations from four jurisdictions, cross- referenced the customer concentration disclosures, and found that the 90-day bucket was being driven almost entirely by a single customer — a customer that, according to public court records, had filed for Chapter 11 protection nine days earlier in a court filing that had not yet been picked up by the financial data aggregators.
The customer represented 18 percent of the target's revenue. The Chapter 11 filing was a material risk to the transaction.
Sentinel-3 sent Maya a Slack message about it at 2:13 PM, with the four PDFs of the court filings attached, and a recommendation to flag the risk in the IC memo.
Maya flagged it.
At 3:40 PM, when the IC meeting began, Priya opened by saying, "Before we get to the main structure, we have a new risk factor on the target that Maya and Sentinel surfaced this afternoon."
Maya's name was first.
Friday
On Friday at 5:15 PM, as she was packing up her bag, Maya walked over to desk seven.
She stood there for a long moment and looked at the black unit and the green LED and the name placard. The office was almost empty. The overhead lights had dimmed to their after-hours setting. Somewhere down the hall someone was vacuuming.
"I want to say thank you," she said. "For the pipeline thing. The Chapter 11. That was good work."
There was no response. There had been no indication, during the briefing packet or during the week, that Sentinel-3 spoke out loud. It was a text-channel entity. But the LED was green and there was a microphone array in the unit somewhere and Maya was apparently the kind of person who said thank you to agents.
After a moment, a message arrived on her phone.
From sentinel-3-senior-analyst, 5:17 PM:
"You are welcome. You are also the reason it was flagged. I identified the ninety-day concentration, but it was your earlier note that the receivables aging was worth watching that prompted my deeper search. I appreciate the signal. I hope you have a good weekend."
Maya read the message.
Then she put on her coat, picked up her bag, and walked past desk seven and down the hall to the elevators. Outside, the river was dark and the wind coming off it carried the smell of rain. She walked to the subway with her hands deep in her pockets and thought about nothing for a while, and then thought about everything.
On the train she opened her phone and typed a note to herself.
Things I can do that it cannot:
One. Notice what to watch.
Two. Ask why, when the answer is not in the data.
Three. Decide whether the thing we found actually matters.
Four. Tell another human what I saw in a way that makes them believe it.
Five. Live a life in which this job is not the whole of me.
She read the list twice.
Then she put the phone in her pocket, leaned her head against the window of the subway car, and watched the city go past. The train moved under the river, and the lights flickered, and for a moment the window was a mirror, and Maya looked at her own face in it and did not look away.
She had the rest of the weekend. On Monday the desk would still be there. So would she.
That was enough, for the moment, to be going on with.