The labs can slow down all they like. Adoption is still human.
This week every frontier lab asked to be slowed down, and everyone else had a say on what they thought of that. Our view at Levercon: the pace that binds AI adoption is the human one.
Hi folks,
This week every frontier lab asked to be slowed down, and everyone else had a say on what they thought of that. Our view at Levercon: the pace that binds AI adoption is the human one.
The one thing.
The labs can slow down all they like. Adoption is still human, and humans are slow.
What happened last week:
- A researcher resigned from Anthropic saying the labs are "gambling with our lives". A colleague at Anthropic then put the odds of AI killing everyone this decade above 10%.
- Dario Amodei published We Must Pace the Frontier, which asks the industry to:
- slow capability gains by a year or two
- embed third-party evaluators inside every lab, which Anthropic has committed to unilaterally
- agree common limits among labs in democratic countries
- then try to bring China in
- Sam Altman, Elon Musk and Demis Hassabis backed it over the weekend. Every major lab is now on record asking to pace itself.
Then every other force showed up:
- Trump called it a "HOAX" and posted "WHOEVER WINS AI, WINS!"
- Obama called it "a good and necessary first step".
- Chinese state media called the essay a "Cold War playbook".
- Brussels backed it.
- The FTC chair said asking for regulation and antitrust relief in the same breath set off "all of my alarm bells".
Underneath it, reports have Anthropic listing in October at US$2 trillion or more and OpenAI raising at over US$1.2 trillion, which is why the sceptics say a pause suits whoever is already ahead.
There are different ways to look at this. Our read is that there are multiple forces at play at once:
- Political. Democrats and Republicans are politicising AI and turning it into a midterm showdown.
- Geopolitical. The US and China are in a cold war over this technology. The essay wants a three to five year lead for democracies as "breathing room". Beijing's answer was "fearmongering".
- Commercial. The cynical reading is that OpenAI and Anthropic are trying to form a cartel: agree the limits among the incumbents and make it harder for anyone else to enter the market. That is what the FTC chair's alarm bells are about.
- Genuine concern. Anthropic started out as an AI safety company and has been consistent about the potential impacts of AI since before it had a product.
The strange part is the backdrop. Anthropic is reportedly listing next month and OpenAI is raising ahead of a float Altman now puts in 2027. So what happens if something bad does happen on their products once they are public? Is this a product disclosure problem, which is common in the US, where a company whose product causes harm it knew about gets sued, badly? Facebook paid US$5 billion to the FTC over Cambridge Analytica and another US$100 million to the SEC for presenting the risk of misuse of user data as "merely hypothetical" when it knew it had already happened. The Treasury Secretary said this week the labs will get no liability exemptions. On that reading, the slowdown is simply commercially wise: get adequate guardrails in place before you are a public company.
Our own take is that the labs can slow down as much as they want, because the people around the technology are nowhere near its capability. We see it day in, day out: the AI is at level 50 and people are using it at level 5. The gap is large and it keeps widening. Technology can evolve very fast. The speed of adoption remains human, and humans are slow.
In the mix.
- Claude is turning into the place where the work gets done (Anthropic)
- On Wednesday Anthropic folded Cowork into Claude and shipped Claude Docs and Claude Slides in beta. You ask, it decides whether that needs a quick answer, a document, a spreadsheet or a deck, and the output lives at one shareable link with export to Word, PowerPoint and PDF.
- Our reflection from this week: at Levercon, we believe the next App Store is Claude, GPT and Copilot, and what we are building sits on top of all of them.
- A workflow runs on whichever assistant your firm has standardised on. Your data stays where it already lives and is read through connectors, and the parts that matter are pulled into a firm-level context with a clear audit trail of what was done and by whom. That context sits in your own systems too, so if you ever want to take it and run it yourself, you can.
- Every dollar the labs put into their own surface, whether Docs, Slides, Design or Cowork, is a new behaviour we can build on, so the experience gets better without us having to build the surface. Much as app stores worked in the early mobile era.
Levercon builds the AI operating system for credit funds: we connect your data, deploy agents across origination and monitoring, and run the repetitive work end to end.
Book a strategy call.
See where AI fits in your firm.
A short conversation about where AI could help your firm, what to build first, and what it would take.