The Silent Intelligence

10/13/2026

On Sunday, Andrej Karpathy posted some napkin math about how differently people now experience AI. By his estimate, roughly 75% of the world's population has barely come in contact with these models at all. Another 20% or so treat free chatbots "a bit like a better Google search" or a writing aid. And then there is a thin sliver, professionals using frontier models for math and code: "Somewhere around 20M people (0.2%) see first-hand that large, complex projects that used to take them weeks/months can now be completed by agents with a prompt."

If you happen to be in that 0.2%, or even visiting it as a tourist, you know how addictive this feels. A friend of mine who is a musician by trade has been building an application with Claude Code that would put some developers to shame.

Yet every time I watch a demo like that, the same thought stays with me: the build was the easy half. When building gets cheap, the hard part does not disappear. It moves.

This week's piece is about where it moves: understanding the work you're building for, the user interface, and change management.

https://read.silentintel.com/p/the-easy-half-and-the-difficult-half?utm_source=facebook&utm_medium=social&utm_campaign=p10&utm_content=page

What was the hardest part of your last AI project, once the build was done?

Or, what a very good Claude Code or Codex prototype still has not proven

10/06/2026

I work with AI agents for a good part of every day: Claude, Codex, Gemini. And for a while I kept searching across all three for my own documents. That, more than anything, is the reason I ended up building my own AI infrastructure.

If I had to build my own infrastructure just to find my documents, imagine how it feels to everyone else.

I hear it from owners and operators often. They have ChatGPT. Somebody swears by Claude. They set up Projects and skills, or bought Copilot licenses for everyone. A few things got faster. And then: I feel like I'm just scratching the surface. What am I missing, and where do I start?

The numbers back them up. In the U.S. Chamber of Commerce's new survey, two-thirds of the small businesses it surveyed now use AI, and the share of owners who call themselves knowledgeable about AI went down, from 74% to 68%. More tools, less certainty.

My answer, in this week's piece: if you are asking, you probably started a while ago. Now figure out where your business gets stuck. The evidence has been sitting in your email threads all along.

https://read.silentintel.com/p/ai-tools-are-overwhelming-where-do?utm_source=facebook&utm_medium=social&utm_campaign=p9&utm_content=page

Which workflow in your business would you look at first?

Or, why you still might be just scratching the surface

09/29/2026

Moscow, late 1980s. I was a teenager, allowed into a real recording studio on the understanding that I would not draw too much attention to myself. When the band went out for a smoke, I went straight to the keyboards they had left on — a Yamaha DX7, a Roland, a Korg.

My school band was drums, bass, a guitar or two, a singer, and sometimes a keyboard player on horrible-sounding East German electric keys. In that room, one or two boxes seemed to contain the whole band.

And I was overwhelmed within minutes. I had no idea about sequencing, programming a drum machine or layering. The instruments were all there. The craft was not.

Every so often we get superpowers delivered to us. In the 1980s it was synthesizers, for musicians and bands. Now it is AI agents, for pretty much everyone, for pretty much everything. But the synthesizers could not, by themselves, produce a masterpiece or a single listener. Neither can the agents.

I wrote about what that room taught me, and why the audience is still the hard part:
https://read.silentintel.com/p/the-synthesizer-moment-in-ai?utm_source=facebook&utm_medium=social&utm_campaign=p8&utm_content=page

What did the first genuinely powerful tool you got your hands on teach you about the gap between capability and craft?

The tools became abundant. The audience did not.

09/22/2026

If the clients are not asking you this already they will pretty soon:

Imagine a twelve-week project that AI helps you genuinely finish in two. The deliverables are good. The client is happy. And you bill twelve weeks, because that is what the proposal said.

Then the client figures it out and asks.

"Should we still pay you the same, now that you are working with AI?"

It is not an unreasonable question. And there are two extreme reactions, both of them wrong.

Say nothing, keep the old model, and quietly absorb the difference for as long as you can. That one is becoming indefensible, and in at least one profession it is explicitly against the rules.

Or hand every saved hour back as a discount. That one prices the wrong thing. The real value was never the hours.

This week I am writing about considerations and ways to think about your business model for professional services in the age of AI. Eight in ten people using AI say it has made them more productive. Only 37% of organizations can point to any effect on their profit at all.

AI made knowledge work more efficient before the market learned how to price it

09/15/2026

A few weeks ago I told Claude to stop thinking like a human.

I run an LLM planning session every Monday across my workspaces, calendar and inbox. This time it came back and told me — politely, with excellent reasoning — that I did not have enough hours in the week to get everything done.

It wasn't wrong. But it had filled my calendar with two-hour blocks for drafting and research, and reserved a Tuesday morning for a task that didn't require me to be awake.

So I typed: you are thinking too much like a human, and I need you to think like AI.

Pause there. I was telling an AI to stop thinking like a human and start thinking like an AI.

Its reply is the part I keep coming back to:

"I did schedule this like a human calendar: I gave you a block for every stage, including stages that don't need you. That's the wrong shape, and it's the thing we'd charge a client to fix."

Then it proposed a new workflow that saved me several hours a week.

Nothing got smarter. No tool was added. One assumption changed — that a task has to occupy a block of a specific person's time to count as done.

These models are mostly trained on data from the past, so they inherit how we used to work. Which means the old thinking doesn't announce itself. It shows up disguised as a perfectly reasonable schedule.

Companies make the same mistake, at considerably greater expense, when they roll out AI department by department.

This week's piece is about why the org chart is the wrong map for that work.

09/08/2026

This week's piece starts on the floor of CES in January, looking for a phrase that was not there.

For about ten years "Internet of Things" was printed across hundreds of booths. This year it had essentially gone. What replaced it was a new vocabulary - agentic AI, physical AI, ambient everything - and my first reaction was to wonder whether the thing I had spent years arguing for had simply lost.

Then I looked at the products, and they were all still doing exactly what they had been doing. The clearest version of it is on my mother-in-law's arm: a glucose monitor the size of a bottle cap that she thinks of as a thing that helps, not as technology.

A technology has won when nobody says its name anymore.

09/03/2026

This week's piece is about something most companies think they have already solved: standard operating procedures.

Almost every company has them written down somewhere. The interesting question is not whether they exist - it is how closely they match what actually happens on a Tuesday afternoon. Usually there is a gap, and usually the gap is where the money is.

What changed my mind on this was watching the plain text files that run my own work get read and executed every single day. A document that gets used daily cannot quietly go out of date. Something breaks visibly, that day, in front of you.

Every company writes procedures. How close to reality are yours?

08/25/2026

A Case for Reviving the Doctrine of Completed Staff Work.

A friend who still approves every hiring offer personally, a one-page memo the US Army wrote in 1942, and why "just delegate" is the wrong advice.