Two businesses. Both the same size. Both using broadly similar AI tools. One is seeing compounding returns: each month, more capability, less overhead, better output. The other is seeing incremental convenience. Useful, but not life-changing. Same tools. Different results. The difference almost never comes down to which specific AI platforms they chose. It comes down to how they think about what AI is in the first place. One business treats AI as a collection of tools you use when something needs doing. The other treats AI as infrastructure: something designed into how the business operates at a foundational level. That distinction produces entirely different outcomes, and understanding it is probably the most useful frame shift available right now for anyone trying to build something durable with AI. ------------- Context ------------- Most people encounter AI as tools first. ChatGPT for writing. An AI transcription app for meetings. An image generator for creative work. A research assistant for information gathering. Each tool is adopted to solve a specific problem, and each one delivers its own set of gains. This is a perfectly rational way to start, and it produces real value. But tool-thinking has a ceiling. When AI is a collection of tools you pick up for specific tasks, each new task requires deciding which tool to use, setting up the context for that tool, getting the output, and integrating it back into whatever else is happening. The overhead of that process repeats with every task. The gains from each tool are real but isolated. They don't accumulate into something larger than their individual parts. Infrastructure-thinking is different. It starts from the question: if AI is going to be involved in how this business operates, what does it need to know, what processes does it need to run inside of, and how does it need to connect to everything else? The answer to those questions produces systems: shared context documents, documented workflows, standard operating procedures that include AI as a participant rather than a visitor.

There are two ways people get privacy wrong with ChatGPT. Some share everything without thinking, pasting in client names, exact numbers, and private details. Others get so nervous about privacy that they barely use the tool at all. Both come from the same place: never having decided how they actually want to use it. Think about the last time you opened ChatGPT to help with something real. Maybe you were drafting a proposal, working through a client situation, or organizing your thoughts on a business decision. You probably typed in whatever context felt useful in the moment, the names, the numbers, the specifics, without stopping to think about where that information goes or whether the tool was set up to keep it private. Or you did the opposite. You held back the details that would have made the answer genuinely useful, because you weren't sure what was safe, so you settled for a vaguer, weaker response. Either way, you were guessing. And you were almost certainly running on the default settings you never chose. ---------- THE REAL PROBLEM ---------- The problem is not "ChatGPT isn't private." The problem is "I've never reviewed the controls or decided what's safe to share, so I'm leaving it to chance." By default, unless you're on a Business or Enterprise plan, your conversations can be used to help improve OpenAI's models. Most people never change that, not because they decided to leave it on, but because they never opened the setting. That's not a tool problem. It's a control problem. And control problems get solved by making a few deliberate choices, not by avoiding the tool or hoping for the best. ---------- WHY THIS MATTERS ---------- When you use AI without clear settings or a clear rule, you end up in one of two costly positions. The first is overexposure. You share specifics that point directly to real people, real clients, or real numbers, on default settings you never reviewed. You may never have a problem. But you've handed over information you can't take back, and you did it without choosing to.

For most of the people we talk to, writing used to be the bottleneck. The blank page, the slow start, the draft that took two hours to get to a point where it felt workable. That bottleneck is largely gone now. A capable AI model can produce a usable first draft in under two minutes. The hard part of writing: getting words on the page, has become nearly effortless. What nobody warned us about is what happens next. When first drafts are cheap, editing becomes the job. And most people's editing process was designed for a world where drafts were expensive and rare, not fast and abundant. The result is a growing backlog of AI-generated content that's good enough to feel like it almost works, but not quite good enough to use without significant revision. A growing awareness is setting in that the revision is taking longer than the writing used to. ------------- Context ------------- The economics of writing have flipped. Before AI, time was heavily front-loaded. Research, outlining, drafting: these consumed the majority of hours, with editing as a finishing step. A piece of content that took three hours might have involved two and a half hours of creation and thirty minutes of editing. Now the ratio has inverted. A draft that takes two minutes to generate might need forty-five minutes of editing to reach a standard worth publishing. The total time is still less than before, but the distribution has changed, and the nature of the work has changed with it. Editing is harder than drafting in one important respect: it requires holding the standard for quality in your head while simultaneously evaluating whether what's in front of you meets it. Drafting lets you externalize thinking. Editing requires you to internalize a clear picture of what good looks like and apply it consistently to every sentence, paragraph, and argument in the piece. Most people haven't developed that capacity deliberately, because most people haven't needed to. The drafting process used to do a lot of the thinking work. The act of writing was also the act of figuring out what you were trying to say. AI drafting removes that process, which means the thinking has to happen somewhere else, usually in the editing phase, which is why AI-assisted editing often takes longer than it seems like it should.

In this video, I break down the new releases from Anthropic: Claude Fable 5 and Mythos 5. While Mythos 5 is still not available to the public, Fable is (note: it's been removed and hopefully be back up soon), and I show you exactly what it's capable of right now. Enjoy!

There's a particular kind of productive-feeling procrastination that the AI era has made very easy to fall into. It involves tabs. Usually many tabs. Reviews, comparisons, Reddit threads, YouTube walkthroughs, LinkedIn posts from people using tools you've never heard of. "Is this better than what I'm using? What is everyone else using? Am I falling behind?" An hour passes. No work has been done. But it feels like work, because you're learning about AI. The gap between what you're currently doing and what's theoretically possible feels like a problem you need to solve before you can get back to the actual work. This is one of the most underacknowledged time costs in AI adoption right now, not the failure to use AI, but the consumption of working hours by the pursuit of better AI. ------------- Context ------------- The pace of AI development creates a genuine psychological pressure. New tools are released constantly. Capabilities improve on timescales of months, not years. The thing that was cutting-edge in January can feel ordinary by April. For anyone paying attention, there's a persistent sense that the current setup might be suboptimal, that somewhere out there is a tool or a workflow that would produce meaningfully better results, if only you could identify it. That sense isn't entirely wrong. The tools are genuinely improving. New options really do appear regularly. Some of them are meaningfully better than what came before. But the question worth examining is what the search for those tools is actually costing. Every hour spent researching, evaluating, and switching tools is an hour not spent doing the work that makes a business run. And the time cost of staying current with the AI tool landscape has grown dramatically alongside the number of options available. A straightforward audit shows what this costs at scale. If someone spends ninety minutes per week researching AI tools, reading about new capabilities, watching demos, and evaluating potential switches, that's about 75 hours per year, nearly two full work weeks. Two weeks invested in understanding what's possible. Two weeks not spent executing.