Executives have long relied on simple categories to frame how technology fits into organizations: Tools automate tasks, people make decisions, and strategy determines how the two work together. That framing is no longer sufficient. A new class of systems — agentic AI — complicates these boundaries. These systems can plan, act, and learn on their own. They are not just tools to be operated or assistants waiting for instructions. Increasingly, they behave like autonomous teammates, capable of executing multistep processes and adapting as they go. Notably, 76% of respondents to our global executive survey say they view agentic AI as more like a coworker than a tool.

For strategists, agentic AI’s dual nature as both a tool and coworker creates new dilemmas. A single agent might take over a routine step, support a human expert with analysis, and collaborate across workflows in ways that shift decision-making authority. This tool-coworker duality breaks down traditional management logic, which assumes that technology either substitutes or complements, automates or augments, is labor or capital, or is a tool or a worker, but not all at once. Organizations now face an unprecedented challenge: managing a single system that demands both human resource approaches and asset management techniques.

The separation of technology and strategy inside most organizations exacerbates this challenge.  — Read More

Three kinds of companies come to me for help with AI. While they are all in different places on their AI-path, they all have the same underlying challenge: how to effectively understand and manage business risk for systems that contain AI-based components.

The first kind of company is on the outside looking in. … The other two kinds of companies are already on the inside, with different problems. One built a working proof of concept, [t]he other already crossed that bridge, shipped something real, and got burned. … None of these companies can see their actual business risk surface clearly enough to make decisions about it. — Read More

Open models, AI models where you can download the weights online, are generally not as capable as the best closed models (models only available through an API), but how large is the gap, and how does it change over time? We try to answer this question by using data from 17 selected benchmarks (8 private, 9 public, ~110 datapoints) measuring various capabilities. All the data and code needed to reproduce this can be found on github.

We find that, as of today, on private benchmarks, where the data is not publicly accessible, open models are roughly 8-10 months behind the closed frontier, while for public benchmarks the gap is roughly 4-6 months. We also find that the gap was smallest around the time of DeepSeek R1, in Jan 2025, and since then the gap has been growing. — Read More

Anthropic are strongly rumored to be about to have their first profitable quarter. Stories are circulating of companies surprised at how expensive their LLM bills are becoming from usage by their staff. I think this is because OpenAI and Anthropic have both found product-market fit. — Read More

#strategy

The question I keep getting from founders and prospective employees: is there any AI application layer left to build, or are OpenAI and Anthropic going to kill everything?

There’s a particular flavor of AI psychosis behind the question. Some people have concluded the only durable places to avoid the permanent underclass are inside a big lab or out on the frontier building in robotics, hardtech, or similar – theoretically anything “the labs can’t touch.” If every piece of software is about to be eaten, either by Codex or Claude absorbing the work directly, or by a future model that will make whatever you’ve built unnecessary, then run!

… The Yellow Brick Road is our shorthand for the path the labs are walking, where they’re committing extraordinary resources. The reason the labs are best-suited for problems like code generation, writing, or image-creation is because these problems improve with raw model capability: every dollar spent on pre-training and post-training improves product quality. Meanwhile, the rest of Oz is inhabited by more complex, often vertical problems, that aren’t as simple as giving a business user a horizontal tool with access to standard tools and computer use. The value comes less from the underlying model’s raw capability (though that’s still important!) than from the scaffolding around it that makes the output trustworthy, compliant, and operational inside a specific industry. — Read More

It’s unclear right now how AI is going to play out for most companies, and I don’t think anyone has a clean answer yet, including me. But there’s a pattern I keep coming back to, and it has less to do with what AI eventually becomes and more to do with what it can already do.

I don’t think the capability curve breaks at some single moment we’d call AGI. It just keeps climbing. Each release adds capability somewhere, and we don’t need to reach the top of the curve for the bottom of it to start reshaping things.

This past Tuesday at Google I/O, Antigravity 2.0 built a functioning operating system from scratch in twelve hours. …Take the staging with whatever grain of salt you want. The point underneath is what “good enough” looks like in mid 2026. … Not because of where it ends up, but because of what it can already do.

A capable agent loop, called many times in parallel, with reasonable cost and reasonable latency, is enough to recreate most of what the application layer of software currently sells. The curve keeps going from here. The question that follows is which kinds of companies sit downstream of that engine and which don’t. — Read More

Amid rapidly growing adoption of enterprise-level AI agents, there’s a disconnect emerging between ambition and execution.

Although 85% of organizations say they want to be agentic within the next three years, 76% say their current operations and infrastructure can’t support that change. They cite a lack of readiness across people, processes, and workflows.  — Read More

Last week, Andrej Karpathy stood in front of a room at Sequoia‘s AI Ascent event and told everyone that vibe coding (a term he invented and made popular) was already obsolete. The future, he said, is agentic engineering. He went on to list exactly what agentic engineering actually involves: 

preserving quality
writing design specsvsupervising plans
inspecting diffs
writing tests
building evaluation loops
managing permissions

… [S]trip the engineer-specific vocabulary, and you have product management. — Read More

At any given time, technology does two things to employment: It replaces traditional jobs, and it creates new lines of work. Machines replace farmers, but enable, say, aeronautical engineers to exist. So, if tech creates new jobs, who gets them? How well do they pay? How long do new jobs remain new, before they become just another common task any worker can do?

A new study of U.S. employment led by MIT labor economist David Autor sheds light on all these matters. In the postwar U.S., as Autor and his colleagues show in granular detail, new forms of work have tended to benefit college graduates under 30 more than anyone else. 

… The paper, “What Makes New Work Different from More Work?” is forthcoming in the Annual Review of Economics.  — Read More

The easiest mistake to make about Anthropic is to treat it as “OpenAI, but with a conscience.”

That story is emotionally satisfying but overly simplistic. OpenAI was the first-mover, ostensibly motivated by “AGI for the benefit of all humanity” which, at the time of writing, is still in their charter. On the surface, this sounds very similar to Anthropic’s morality.

After all, Sam Altman is easy to read, what you see is what you get. He wants money, power, control, and the largest possible seat at the table. There is nothing particularly mysterious about that. Throw a dart blindfolded in the Bay Area and you’re bound to hit a founder with the same goals.

Anthropic is harder to read because it speaks in a different register. They talk about safety and alignment, but they go beyond the CBRN and cybersecurity risks that OpenAI focuses on. Anthropic adds in x-risk, and lately, geopolitical dominance as its top-of-mind concerns. They don’t just want to win the AI race on business terms, they believe they have a personal mission to save humanity. 

That is precisely why they deserve more scrutiny. — Read More