AI’s builders spent the week arguing about the end of the world. Your techs still have to beat Wednesday.

A terrifying week, in their own words

This week the people building AI argued in public about how fast it might kill us all. That’s kind of terrifying.

Jacob Coxon went first. He resigned from Anthropic, walked away from unvested equity, and warned that Anthropic and OpenAI were racing to self-improving superintelligence and “gambling with our lives.” On CBS, he said that if you have a super advanced intelligence, “it will be smart enough to kill us.”

On Saturday, Sept. 12, Anthropic CEO Dario Amodei published an essay titled “We Must Pace the Frontier.” Its core line: “We must slow the pace at which we improve the capabilities of AI models.”

Within a day, Sam Altman posted, “I agree with Dario that we need to pace the frontier.” Elon Musk replied in three words: “Dario is right.” Amodei then told CBS the industry “lied to people” about whether the technology had risks.

We’ve heard this before. Bill Joy told Wired readers in April 2000 that we were entering the new century “with no plan, no control, no brakes,” and put human-level computing power about 30 years out. His window is still open.

Full disclosure: Ironworks builds on Anthropic’s Claude models.

The other view, from Richmond

On Episode 4 of Richmond Next, presented by the Greater Richmond Partnership, VCU’s Michael Rao sat down with Accenture’s David Metnick to talk about AI. Rao went after what he called a myth, the idea that AI will entirely replace humans and human leadership. “It is not,” he said. “It is actually going to bring out the very best of humanity.”

Less terrifying. Probably where we land at the end of the day.

His reasoning is the part worth hearing. Rao said AI “is going to give us an opportunity to be humans who have capacity like no other element that exists in society”: the ability “to be compassionate, to be empathetic,” the ability “to make judgements,” and “a conscience.” The full Richmond Next conversation runs about nine minutes on richmond.com, and it’s worth the watch.

We agree with him. None of that shows up on a benchmark, and all of it shows up in how a service business runs. Think about the best technician you’ve ever had. Customers asked for that tech by name because that tech could stand in a hot attic in August, look at a 17-year-old system, and tell an elderly homeowner the truth about whether to replace it or nurse it through one more summer. That’s judgment, empathy and conscience in one conversation, and AI is a long way from doing that part of the job. Take the hold music and the paperwork off that tech’s plate, and you get more of those conversations.

Doom or hope, neither forecast changes whether your technicians finish more jobs on Thursday than they did on Wednesday.

Meanwhile, in a 12-truck shop

Picture a 12-technician HVAC shop somewhere in the mid-Atlantic. The dispatcher builds the board at 7 a.m. By 9:30 one job has run long, a customer has canceled, and she’s rebuilding the day by phone.

Out on a call, a tech finds a failed blower motor and calls the supply house to check stock. Hold music. The first supply house doesn’t have one, so it’s a second call and more hold music. The tech finishes the repair, then thumbs job notes into a phone in the cab. The invoice waits until someone in the office can decode those notes, or until the owner works through tickets at the kitchen table on Sunday night. A Thursday repair invoiced the following Monday means the owner floats the parts for four days.

Run that across 12 trucks and 4 jobs apiece, and the owner is paying for a lot of hold music.

Those minutes got no airtime this week. Every one of them is either on payroll or off the board.

The Thursday math

This is a hypothetical, so here are the inputs. Change any of them and the answer moves.

• 12 technicians, each finishing 4 jobs a day, for 48 jobs on Wednesday.
• An 8-hour day, so each job slot runs about 2 hours, drive time included.
• A $350 average ticket.
• 15 minutes per job the tech loses to the parts call and the job notes, averaged across every job, and won back once the fixes are in.

Fifteen minutes across 4 jobs is an hour per tech per day, and a free hour won’t hold a 2-hour job. The savings won’t land evenly, though. The techs on parts-heavy calls get back more. Say one tech in four gets back enough for a fifth job. That’s 3 more jobs a day: 51 on Thursday instead of 48. At $350 each, that’s $1,050 a day in billed work. Over 250 working days it comes to $262,500 a year, from the same 12 techs and the same trucks.

Too rosy? Cut the lost time to 10 minutes a job. That’s 40 minutes per tech. Say one tech in six picks up a fifth job: 2 more jobs, $700 a day, $175,000 a year.

The fixes tend to be unglamorous. The tech gets a stock answer without sitting on hold. The tech dictates notes on the walk back to the van. Software drafts the invoice from those notes before the truck leaves the driveway, so the owner stops floating parts over the weekend.

So what do you do with all this?

So, back to the doomers for a minute. We’ve run this contrast before, in June with a finance team’s four-day close and in April with the AI model too powerful to release, and the news cycle keeps handing it to us. The noise is the noise. There will be more resignations, more essays, more CEOs agreeing with each other on X, and plenty of people will keep getting worked up about it. Some of them have good reason to.

It’s also fair to point out that Anthropic and OpenAI, the two companies at the center of this debate, are both headed for the stock market. Axios reported this week that Anthropic is still on track to go public in 2026 and that OpenAI is leaning toward 2027. A lot of money rides on how those IPOs go. We build on Anthropic’s models, as we said up top, so weigh our take accordingly. The money and the warnings can both be sincere, and Coxon walked away from his unvested equity when he left. All of it is a good reason to spend more of your attention on your own business than on theirs.

What we like to do is bring all of it down to practical terms. What can your business do about any of this?

Start using AI, in some small way, today. You don’t need a strategy deck or a steering committee for that. Pick one chore that eats time every week, like turning the techs’ job notes into clean write-ups or drafting the Sunday-night invoices, and try it there. Use it like a tool, the same way you use the computer sitting on your desk.

That computer was a terrifying piece of technology at one point, too. People worried about what it would do to their jobs, and plenty of them were sure they’d break it. Now it’s where you run payroll, and most people don’t give it a second thought. AI is likely headed the same way, one boring job at a time.

Rao is probably right about where all of this ends up. In the meantime, your dispatcher still builds Thursday’s board at 7 a.m., and getting a few more jobs onto it than Wednesday’s seems like a pretty good place to start.