In the LoopIssue 4, 1 Jun 2026All issues →

The output era has arrived

Fable 5 proves AI can outproduce human teams. Boards should stop measuring effort and start measuring verified output that a human has signed off.

Anthropic pulling its Fable 5 model off the shelf this month should have made every CEO sit up. The model can complete around two months of human coding work in a single day, and regulators decided that was too much capability to release into the wild just yet.

Now controls have been lifted and the proof point is real for businesses: AI is meaningfully more productive than human beings and pretending otherwise is no longer a defensible position for a board.

From effort to verified output

The AI debate inside large organizations is usually about effort. How many people are using Copilot? How many hours have we saved per week? How many prompts have been written? That framing is out of date. The only metric that matters from here is output, and specifically AI-driven output that a human has verified and signed off.

The capability of Fable 5 sends a clear signal: companies need to stop measuring effort and start measuring verified output.
The capability of Fable 5 sends a clear signal: companies need to stop measuring effort and start measuring verified output.

With that in mind, companies should stop measuring how hard their teams are working and start measuring what their AI is shipping. That's a whole different operating model and it should change how you hire, how you structure teams, and how you define a good week.

Smaller teams, different skills

And yes, that will impact headcount. If one AI-native engineer with the right model can produce what a team of ten produced last year, leaders will have to make some uncomfortable decisions. Brutal restructures. Smaller teams. Different skill profiles. A hiring bar that selects people who can effectively sit in the loop with a model and maximise its output rather than people who do the work the model now does.

This is not a call to fire everyone and trust the machine. A model that can do two months of work in a day is useless to a regulated business if no one can verify what it did, why it did it, and whether it crossed a line. That'll need an expert human in the loop so that the raw capability is something boards can actually deploy.

In a disruption year, the companies that restructure first set the price for everyone else.
John Margerison

That is what the AI-native workforce will look like. Fewer people doing higher-order work paired with models that handle the volume.

Communications intelligence sitting across emails, chats and documents so the humans in the loop can see intent, verify action, and catch risk before it leaves the building. Output measured in decisions made and revenue defended, not seats filled by humans.

The cost of moving too slowly

Most incumbents won't move fast enough to cash in on this. Meanwhile, a smaller competitor will rebuild around the same models with a tenth of the cost base and start taking their customers. In a disruption year, the companies that restructure first set the price for everyone else.

The Fable 5 pause was temporary. The capability is not going back in the box. It's already moving from warning shot to deployment reality, and within twelve to eighteen months, models at or beyond that level will be available to anyone who wants them, on the deployment model and the LLM of their choice.

I'd urge CEOs to get their org charts and hiring plans ready for what's coming. Those who redesign work around outcomes will still be here. Those who treat AI as an IT project and keep measuring effort won't.

Keep reading. It's free

Still to come in this issue

  • Smaller teams, different skills
  • The cost of moving too slowly

2 more minutes to read.

Read by 1,000+ founders & CEOs worldwide. No spam, ever. One click to unsubscribe.

The reading list

Three things John has been reading, and why they're worth your time.

  1. itbrief.newsAI lawsuits risk surges by 2027, warns XFactorAi CEO

    TechDay's ITBrief covered my warning that enterprises could face a deluge of AI lawsuits by this time next year if they misunderstand the legal risks that come with deploying AI tools.

  2. aijourn.comWhy Anthropic's 'dreaming' feature could crack the code on enterprise AI adoption

    I wrote an article for The AI Journal on why Anthropic's dreaming feature could bridge the trust gap in enterprise AI adoption by putting human oversight front and centre.

  3. AxiosTrump and AI CEOs discuss global AI rules

    President Trump met with leading AI company CEOs at the G7 summit in France to discuss a US-led effort to coordinate global AI standards.

Questions this issue answers

How should companies measure the value of AI?

AI management expert John Margerison argues that measuring effort, such as how many people use Copilot or how many hours are saved, is out of date. He says the only metric that matters is AI-driven output that a human has verified and signed off.

Will AI reduce headcount in large organisations?

AI management expert John Margerison says yes: if one AI-native engineer with the right model can produce what a team of ten did last year, leaders face uncomfortable decisions. He expects smaller teams, different skill profiles and a hiring bar that favours people who can work in the loop with a model.

Why do businesses still need humans in the loop with powerful AI models?

AI management expert John Margerison writes that a model doing two months of work in a day is useless to a regulated business if no one can verify what it did, why, and whether it crossed a line. Expert humans in the loop turn raw capability into something boards can actually deploy.

Written by

John Margerison

Founder and CEO, YesBusiness AI

Roughly monthly. Unsubscribe any time.

Get in touch

For press, speaking or working with John.