Ridgeline

Writing this from 35,000 feet, using Starlink satellites, giving direction to AI agents across multiple projects in parallel. I asked ChatGPT how many people might also be on a United flight doing the same thing at any given moment. Its guess: 20. The future is here. It is just not evenly distributed. But the longer I live in this new world, the more I notice it is not just the technology that is unevenly distributed. It is our capacity to absorb it. Getting off the plane at O'Hare, I walked through Terminal 2. It has not been updated in decades. You can still see where the telephone booths once were. Spend a few hours working with AI, then try to go back to the old way. The contrast is jarring. The old world has not disappeared just because the new one arrived. Knowledge workers are running two operating systems at once: the existing job they are still accountable for, and the AI-native workflow they suspect will eventually evolve it. Energizing and exhausting in equal measure. This is the transition tax. Not the cost of the technology. The cost of reorganizing around it. The question worth asking right now is not how do we use AI to do the old work faster? But what would this work look like if AI were assumed from the beginning? The advantage goes to whoever is willing to redesign. And pay the transition tax! Full edition in the comments below!

The crown jewels in software have historically been the source code. Code became the product, and the workflow around that code became the business. AI complicates that story. If the model is rented from OpenAI, Anthropic, or Google, AI-native applications must prove they own more than just access to intelligence. That is the question I keep coming back to: where is the IP in AI products? The current best answer is that the harness is the product. Think of it as an exoskeleton for the LLM. The harness gives the model memory, tools, integrations, permissions, proprietary data, reusable skills, and evals. It turns raw intelligence into something that can deliver reliable outcomes. But not every harness is equally defensible. If the product is just a chat interface and a few saved prompts, it may be thinner than we want to admit. Microsoft can launch a Word feature and crater you. Customers may decide they can build it themselves. The more interesting answer is that the real IP lives underneath the visible application layer. It lives in the accumulated understanding of how work actually gets done: the tacit knowledge that comes from repeatedly doing the work. In SaaS, the product was the software. In AI-native companies, the product may be the harness. But the "business" of AI may be in delivering the outcome. There is an old adage: “People don’t want to buy a quarter-inch drill. They want a quarter-inch hole.” That feels increasingly relevant to AI. A law firm does not need another copilot, it needs trusted work product. A CFO does not need another accounting assistant, she needs the books closed. An insurer does not need another claims tool, it needs claims processed accurately. The winners may be the companies that turn rented intelligence into repeatable, trusted, and governed outcomes. Rent the model, harness it, and generate outcomes for customers. The IP may live in the system that connects all three. The full story, including how this is playing out in legal AI with Harvey and Legora, is in today’s full edition.

The next killer application for AI may sound like a bad idea: give it access to your computer. The machine where knowledge work happens. Your browser, files, Excel, Word, inbox, and calendar. That sounds reckless. It may also be the most logical next step toward making AI useful for real knowledge work. It’s called AI computer use. The AI uses a keyboard, mouse, browser, and operating system just like humans do. If AI is going to automate knowledge work, it has to get “hands on” with the knowledge worker’s primary tools. The AI is moving out of the chat window and into the environments where real work happens. We have already seen this start to work in agentic coding: codebases, terminals and GitHub. Knowledge work is messier. It does not have the more structured world of software development. The first wave of products are here. Claude Cowork, Perplexity Computer, Factory’s Droid Computers, OpenClaw, and Hermes. Different products, same idea. Move the AI into an “always on” position and into the place where work happens. By now, you’ve probably formulated an opinion on whether you would give AI access to your computer. My hunch is that most readers are a “no.” I get it. You could buy a new computer and give the AI its own. That’s what I’m doing with my Mac mini, which arrives this month. Giving AI access to your computer is not a small permission. It is like giving a junior employee your laptop and telling them to log in and help you get things done. That employee may be capable. They may also misunderstand the task and post private photos to LinkedIn. That is funny until it is your LinkedIn. In the human world, we manage that risk with roles, permissions, and supervision. Zero Trust starts from a similar premise: do not assume trust just because something is inside the system. Verify access. AI needs this version of trust machinery. It cannot just use the computer. It has to use the computer as "someone" with a specific role, with permissions and boundaries on what it can and cannot do. For decades, knowledge work was: learn the software, push the buttons, produce the work, share it. The new model may be different: describe the outcome, watch the system work, approve what matters, share it. From pushing the buttons yourself to supervising the work. We may want to give the AI access to a computer to get there.

If you are worried you're falling behind on learning to use AI. Don't be. Gallup says half of U.S. employees now use AI in their role. But only 28% use it frequently, and just 13% use it daily. It is still early. Very early. Most people are still using AI like a slightly better version of Google. That alone is useful. I do it too. This morning I asked, “How many days until Mother’s Day?” But it gets much more interesting when AI starts taking real work off your plate. That is why I like starting with a to-do list. There is an old story about a man who walked up to J.P. Morgan and said, “I have the secret to success in this envelope, but it will cost you $10,000.” Morgan asked to see it first. He opened it. Inside was a single line: “Make a list and do it.” Morgan paid the man. That is usually where my day begins. Pen, legal pad, and a list of what is actually on my mind. Then I take a picture of the list, drop it into ChatGPT, and say: “Turn this into a plan. Do the first pass on anything you can.” That is the human-to-AI handoff. One item becomes a research brief. Another becomes a prioritization engine. Another becomes a cleanup job I’ve been putting off. And if the workflow is worth repeating, I move from ChatGPT into Claude Code and build something more durable. The answer is not just which AI tool to use. It is time to think differently about your work and how to get the job done. Start with your list. Then ask a simple question: What can the AI do for you? In today’s full edition, I walk through five real examples from my day, plus prompts, best practices, and a few tips that have helped me get more out of these tools. Link in the comments.

My first-grader and I have been building software that helps her learn to spell and type. She is six and wields a whiteboard and Claude Code. What fascinates me, as her dad, is the combination. A six-year-old’s imagination is already wild. Claude turns that imagination into something tangible. Today, we have a Trolls-centric, Sing 2-themed spelling game with different modes, levels, and mini-game rewards, including her own version of Doodle Jump. And while we wait for Claude to write the code, we spell the words with pencil and paper. Again and again. She has been getting good marks on her spelling tests. When a close friend asked what I am teaching my kids about AI, I came back to this story. The new edge may belong less to the strongest or smartest, and more to the children who learn to respond to change. That is what I want to reinforce at home. Not fear. Not doomerism. I want them to build. Ask questions. Speak clearly. Treat people well. Stay curious long enough to notice what others miss, and resilient enough to keep going when the first version does not work. I want to give them enough agency to feel the consequences of their decisions while the stakes are still manageable. That is how judgment gets built. I want them to understand that they are more than any one skill the market prizes in a given moment. Titles change. Tools change. Status changes. The deeper work is learning to adapt without losing yourself. That is what I am teaching my kids about AI: a responsible relationship with powerful tools, character, curiosity, communication, resilience, judgment, and the instinct to ask better questions. It is a new form of street smarts for an AI-native world. Lots more in this week’s full edition of Reading Ambitiously in the comments below.

First, the ChatGPT era. Now, the Claude Code and agentic coding era. Up next? My hunch is always-on intelligence. There is an iPhone-caliber product to be built around this idea. You read an article on Tuesday. By Thursday, it connects to a conversation you are having. You know you saved it. And when you need it, you cannot find it. What if you had an executive assistant monitoring not just that one article, but every input across your knowledge work? An assistant with memory, an understanding of your goals, and the ability to help you move toward outcomes before you even ask? We are starting to see early versions of this with OpenClaw and Claude Cowork. In a prototype of AmbitiousOS, Claude maintains a living wiki on top of the source material. New inputs update a growing knowledge base on companies, people, and concepts. It feels like hiring your own librarian to build a private Library of Alexandria around your work. Over time, your accumulated knowledge starts to look less like a pile of inputs and more like a living asset. Somebody is going to productize this. And win big. This may be part of what Jony Ive is now working on with OpenAI. Maybe it is something even bigger. This week’s Reading Ambitiously includes a demo of a prototype with these capabilities. The video edition and full edition are in the comments below.