We Now Have the Two Most Powerful AI Models Ever Built. So Why Can't You Use Them at Work?

Hello,

This is Simon with the latest edition of The Weekly. In these updates, I share key AI related stories from this week's news, list upcoming events, and share any longer form articles posted on the website.

P.S. — sharing this issue with a couple of people gets you a free guide. More on that partway down.

Anthropic released its most powerful Claude model, Fable 5, to the public on 9 June 2026, and within days, the US government restricted its availability outside the US due to its enormous capabilities and the potential damage that could be done with it in the wrong hands. After a few weeks of negotiation, the restriction was lifted, and Fable 5 was reopened to the public across the globe on 1 July.

Despite Fable 5 now being widely available, I'm aware that many companies on Enterprise plans still haven't opened it up for use by their workforce, and I suspect there are two reasons behind this. The first is the aforementioned immense capability of this model, which, with some loose instructions, could take unwanted actions on a worker's machine. The second is that it's also very expensive. Companies are already trying to put the toothpaste back in the tube when it comes to their teams using AI tools, with costs going up dramatically over the last few months. I'm writing about this now because OpenAI, the maker of the most widely known AI chat tool, ChatGPT, recently launched their most advanced model, GPT-6 Astra. I won't go into the details of what makes these latest LLM models so powerful, but they really do have the potential to vastly change how we do things in the world of work, and our social lives.

So we have two incredibly powerful options now at our disposal, but it's interesting to see that some businesses have put a cap on this, making them out of reach for most of the workforce. I'll be intrigued to see whether some people are granted access with elevated privileges based on role and seniority, or whether, over time, once the hype has calmed down, these businesses will actually make them available again once they realise the gains they can make in the office. It's hard to say, obviously, but have we reached the ceiling of AI capability for the everyday business user?

"When I think of revenue, I think of Attio." - Shreman Shrestha, Head of Business at Granola

Here's what that adds up to:

Start Your Trial

On September 2nd, Uber announced that it is cutting roughly 10% of its global workforce, which equates to around 3,300 roles. These are mainly middle managers, in what CEO Dara Khosrowshahi described as "removing layers and simplifying team structures," according to CNBC. AI was notably absent from the official explanation, yet earlier this year Khosrowshahi disclosed that AI agents are already writing about 10% of Uber's code, and the company had blown through its entire annual AI tools budget in just a few months. The cuts are the largest since the pandemic.

Why it matters: A broader pattern is emerging: 50% of major layoff events announced so far in 2026 cite AI as a contributing factor, yet analysts note widespread "AI washing". For business leaders, this makes it harder to have honest internal conversations about where AI is genuinely changing headcount and where cost pressure is simply finding a convenient label.

The UK government has launched a consultation on regulating workplace monitoring technologies, which is open until 30 September 2026. It covers AI tools used to track productivity, monitor communications, and assess employee behaviour, and is seeking input on three possible policy approaches ranging from voluntary codes to new statutory obligations. The consultation is open to employers, employees, and representative bodies.

Why it matters: This is a direct opportunity for UK businesses and HR teams to shape how these tools will be regulated before rules are set. The direction of travel is clearly towards greater worker protections around AI monitoring, which has to be a good thing. Companies that get ahead of this, rather than waiting for legislation, are in a stronger position.

MIT's Project NANDA reviewed more than 300 enterprise AI deployments and found that roughly 95% delivered no measurable impact on the profit-and-loss statement, with only around one in twenty reaching production with real value, according to reporting across multiple outlets citing the study. Separately, Gartner found that more than 50% of generative AI proof-of-concept projects are abandoned before reaching production. The abandonment rate for AI initiatives among companies surveyed rose from 17% in 2024 to 42% in 2025.

Why it matters: The gap between AI investment and AI return is now being quantified at scale, with five major research organisations — MIT, Gartner, RAND, BCG, and McKinsey — all saying the same thing: the majority of enterprise AI projects fail to deliver expected value.

Thanks for reading, and see you next Thursday.

Simon,

Was this email shared with you? If so subscribe here to get your own edition every Thursday.

You can also follow me on social media: