A builder walked out of the AI race. Should your company still walk in?

The intelligence in artificial intelligence is still today very jagged. — Toby Walsh, University of New South Wales, to Al Jazeera

Three stories from last week belong on your desk. A researcher who helped build AI at OpenAI and Anthropic quit and said out loud what many in the labs believe. OpenAI shipped a model built to take a goal and work the steps on its own. And your team notices when AI writes the things only a leader should write. Underneath all three sits one question: what do you hand off, and what do you keep in your own hands?

The story

If AI could end us, should your company still deploy it?

It is hard not to feel a little undone by last week's news. On September 8, Jacob Coxon, 27, resigned from Anthropic after three years of research there and at OpenAI. His post on X drew more than 90 million views in a day, TIME reported. The people building AI, he wrote, "earnestly believe that it could kill us all by the end of the decade."

He is not alone. Elon Musk has put the odds of an AI catastrophe at 10 to 20 percent. Anthropic's own CEO, Dario Amodei, has said there is a 25 percent chance things go very badly, alongside a 75 percent chance they go very well. Five days before Coxon quit, Senator Bernie Sanders and Representative Greg Casar introduced a bill to ban superintelligence and pause advanced AI development until federal safety rules exist.

Picture a neighbor working in his garage on something that could power the whole street. He admits a one in ten chance it burns the neighborhood down. Now make the neighborhood the whole planet, and hand the garage to a few people.

Who sits at the table?

The empty seat

Who asks the soul questions?

Sam Altman

OpenAI

Demis Hassabis

Google DeepMind

Mark Zuckerberg

More this week

The week AI learned to take an assignment

On September 3, OpenAI unveiled GPT-6 Astra and opened it to paying users the next day. Most model launches blur together. This one is different, because OpenAI built it for long, multistep work: research, documents, spreadsheets, and the computer tasks that eat an afternoon. Toby Walsh, who studies AI at the University of New South Wales, cautioned that the technology still performs unevenly, and that the labs show little sign of slowing down to address cyber risk.

Skip the benchmarks and notice the shift. For three years most of us used AI like a fast reference librarian: ask, receive, repeat. A tool like Astra behaves more like a new hire. You hand over a goal, and it works the steps. Microsoft's 2026 Work Trend Index describes the upside plainly: as agents take on the execution, people move to directing the work, making the calls, and owning the outcomes.

That puts an old leadership skill back at the center of the table. Delegation. The muscles match: a clear outcome, the right context, a checkpoint, and the humility to let go of how it gets done.

This week's work

Delete

What no longer serves the mission

Delegate

The messy middle, to AI or a teammate

Do it yourself

What only you can do

Three doors for every task, before anything reaches an agent.

Your team notices when AI speaks for you

A survey released last week should make every manager pause. Omni Calculator asked 913 working Americans how they feel when their boss uses AI to do the boss's own job. Forty-two percent said it makes the manager seem less competent, and 36 percent said they trust that manager's judgment less. Workers reported managers using AI to prepare performance reviews (27 percent), to handle conflict (13 percent), and to give career advice (12 percent). Among Gen Z workers, 76 percent said messages their manager wrote with AI feel less personal.

Now set that beside a second study. In a Founder Reports survey of 2,000 U.S. workers, more people said they could be completely honest with an AI chatbot about work stress (20.9 percent) than with their own manager (17.2 percent). Read the two findings side by side and the problem is plain. The machine hears the confession. The manager sends a review the machine drafted. We have it backwards.

Microsoft's research points to the fix. When managers openly model AI use, their teams report a 30-point rise in trust toward AI agents and a 22-point rise in critical thinking about how they use it. Use AI in the open, on the work, and your people follow your lead. Let it quietly stand in for you in the relationship, and they stop bringing you what matters.

What the August job numbers tell a leader

For five straight months, companies named AI as the top reason for U.S. job cuts. August broke the streak. Challenger, Gray & Christmas reported that AI slipped to the fourth most-cited reason, tied to 3,462 cuts, its lowest monthly total since December. Before anyone exhales, look at the full year.

116,175

U.S. job cuts tied to AI so far in 2026, about 22 percent of the total

4th

AI's rank among layoff reasons in August, after five months at number one

+37%

Announced hiring plans through August versus last year

Employers have announced plans to hire 119,825 workers this year. Andy Challenger flagged the catch: how long those roles take to fill, and whether employers find people with the skills the new roles demand. So AI keeps reshuffling work far more than it erases it. Some seats go away. New ones open. The people who have already learned to work alongside these tools get first pick of the new chairs.

Also on the radar

Trust holds the key to agents. Harvard Business Review argued last week that trust, more than technology, holds back AI agents at work. People will not give an agent real autonomy while they doubt its reliability.
The church at the table. The Presbyterian Church (U.S.A.) shared reflections from its delegation to the 2026 AI & Church Summit, where leaders from several denominations wrestled with how the church might use these tools without handing its mission to a machine.

The practice

Thirty minutes before Friday

Block 30 minutes before the week ends. Sort your recurring work into the three doors above, the sort Microsoft's Work Trend Index says every leader now faces. Then write your Handwritten List: the three to five moments you will always keep human. For most leaders it includes performance reviews, apologies, encouragement, conflict, and hard news. Let AI help you prepare for those moments. Then write the words and say them yourself.

Weak, because it asks the tool to decide for you:

What should I delegate to AI this week?

Strong, because you keep the third door:

Here is my recurring work for the week: [paste your calendar or list].
Sort each item into Delete, Delegate, or Do it yourself. For anything you
put in Delegate, say what "done" looks like in one sentence and where a
human should check it. Leave the Do-it-yourself column to me.

Share both lists with one person on your team and ask them to hold you to it. Then tell your team where you draw the line and where you happily use AI. Saying it out loud takes the suspicion out of the room. Any current tool, free plan, about 30 minutes.

Higher ground

Scripture does not hand us a forecast for superintelligence. It hands us a posture. When Israel needed leaders who could read a confusing moment and act, the chronicler singled out one tribe.

Men of Issachar, who understood the times and knew what Israel should do.

1 Chronicles 12:32

That is the empty seat at the table above. The room is full of people who can build. It is short of people who understand the times, who ask why we are here, and who know what to do next. If this issue has a single ask, it is that you refuse both panic and indifference and take that seat where you lead.

The next step

Are you leaning in, holding back, or still deciding? Hit reply and tell us. We read every one.

If you are not sure where you stand, find your AI Edge. It is free, it takes about five minutes, and it names the strength you bring to a room like this.