Re: AI, productivity, and profit for us mere-mortal-non-frontier-lab sort of companies.
In some places, individual people will capture the upside of AI and keep it for themselves. In others, the company will realize the value.
We’re years in now. Everyone has had access to Codex, Claude Code, and the models behind them, along with training and investment.
A lot of ink has been spilled about the lack of ROI for companies. Some of that may come down to the size of the denominator. AI has been very expensive, and companies have been cutting very large cheques just to run pilots.
Putting more money into a bad idea doesn’t guarantee a better return on the numerator though.
I also wonder whether the value will ever show up on the balance sheet, if employees are capturing it at the individual or team level.
Some of the value may not be the kind of value a balance sheet can see.
Tom Tunguz expected AI to cut the time it took him to ship a blog post . It didn’t. His edits per post held steady. What changed was quality, and his weakest posts improved the most. In his words, AI “raises the ceiling of what the same amount of work produces.” That’s a real return, but it shows up as better work, not lower costs. How do companies price in “better”?
If you can grasp the technology and get more done with it, what incentive do you have to pass those returns on to the enterprise? Why not keep them for yourself: time back in your day, more room for other projects, more time for learning and development?
George Sivulka calls this “context hoarding” : “Employees don’t want to teach AI systems their secret sauce.” He points to Meta, where employees who own stock, and are wildly incentivized to get AI right, still pushed back when the company used their work as training data. As he puts it: “Nobody trains their replacement for free.”
I often reach for Upton Sinclair’s line as an explanation for lagging adoption:
It is difficult to get a man to understand something, when his salary depends upon his not understanding it.
But it cuts the other way too.
For top performers who understand the technology very well, their salary might not reward them for showing how much more they can do.
We keep telling ourselves we’re living through a technological revolution, while we keep underestimating how much we need to reorganize our companies to meet the moment.
So what does reorganizing look like? I think it starts by being honest about the trade.
If someone can now do the work of three people, a company that just hands them three people’s workload will get exactly what it pays for: quiet hoarding.
The companies that capture the value will be the ones that redesign around these people making sure some of the upside flows back to the person who created it, whether as scope, pay, or autonomy.
The technology is already here. What’s missing is the org chart.
Addendum, October 23, 2026
The same day this went up, Anthropic published How employment is changing at Anthropic . It’s a look at what happens inside a company when the “one person doing the work of three” scenario stops being hypothetical. Commits per engineer were up roughly 5x in six months. Agents logged about six hours for every human hour, and human attention became the bottleneck.
Then they ask the question this piece is about: if the gains are that large, how do employees share in them?
“Three features stand out,” they write, and the first is the one I’d been circling:
First, widespread stock ownership gives employees a strong interest in the company’s success.
Solutions are starting to show up. So far it’s mostly at the companies that can most easily afford it.