Anthropic finally brought Fable 5 back and in the same week, they also launched the new Sonnet 5 model. In this video, I break down everything you need to know about these models and explains which one you should be using. Enjoy!

There's a mental model for working with AI that most people inherit from their experience with software: find the tool, learn how it works, use it to accomplish specific tasks. The mental model is tool-use, and it produces a certain kind of result. There's a different mental model that produces a different kind of result: management. Specifically, the kind of thoughtful management you'd apply to a capable but inexperienced hire who needs clear direction, good context, consistent feedback, and well-understood expectations to do their best work. These mental models produce genuinely different outcomes. Not because the tools are different, but because they shape how people interact with them in every session. ------------- Context ------------- The tool-use mental model tends to produce transactional interactions. You need something done. You open the tool. You describe what you need in the way that feels natural. You evaluate what comes back. You iterate until it's close enough. You move on. This works. It produces reasonable output. But it carries a specific set of limitations that become most visible when the work requires more than average output. The tool-use approach doesn't naturally lead to investing time in context, because context feels like overhead on a transactional interaction. It doesn't naturally lead to articulating quality standards clearly, because the assumption is that the tool will produce something and you'll adjust it. It doesn't naturally lead to diagnosing what went wrong when output misses the mark, because the instinct is to try a different prompt rather than identify the root cause. The management mental model produces different habits. A manager who wants good work from a new hire invests time in context upfront rather than treating it as overhead. A manager provides examples of what good looks like rather than leaving quality standards implicit. A manager who gets poor work diagnoses whether the problem was the brief, the capability, or the execution rather than just asking for a redo. These habits, applied to AI interactions, produce significantly different results over time.

Most people think success is about money. It's not. Money is just what buys you options. I've worked hard for decades. Not because I fell in love with the grind, but because I fell in love with what the work could create. Every uncomfortable conversation. Every risk. Every time I wanted to quit but didn't. None of it was just to make more. It was to own my time. To be there for the people I love. To create memories instead of regrets. To have the freedom to say yes to what matters and no to what doesn't. Don't chase success because you want to look successful. Chase it because one day you'll realize time is the only thing you can't earn back. Work hard. Do the uncomfortable things. Become the person capable of creating the life you want. Because real success isn't measured by what you own. It's measured by how fully you get to live. Question for you: If you had complete freedom over your time one year from now, what would you spend more of it doing... and who would you spend it with?

Research published in early 2026 found that the average small business worker saves 5.6 hours per week using AI tools, with managers saving closer to 7 hours. Those are meaningful numbers. Across a year, 5.6 hours per week is over 280 hours: roughly seven full working weeks returned to professionals who use AI consistently. Most people who see those numbers nod in recognition. The time savings feel real. There's less friction on specific tasks, drafts come faster, research compresses, routine work moves quicker. And then someone asks where those 280 hours actually went, and the conversation gets complicated. ------------- Context ------------- The productivity paradox of AI is one of the least discussed aspects of the current wave of adoption. Time saved on tasks and felt experience of having more time are different things, and for a significant number of professionals, they're not converging the way the numbers suggest they should. The explanation isn't mysterious. Time savings don't automatically translate into felt margin unless the saved time has somewhere deliberate to go. If the work expands to fill the capacity AI creates, if new obligations emerge to absorb the recovered hours, if the time savings get distributed across thirty small tasks rather than accumulating into meaningful blocks, the felt experience of the week doesn't change even when the productivity data does. This is the absorption problem. Time savings get absorbed rather than accumulated, and the absorption is usually invisible. No single thing consumed the saved time. A hundred things each took a little. The net experience is: I'm using AI, the tasks are definitely faster, but somehow the week is just as full. A consultant described this pattern with unusual precision. She tracked her time carefully before and after adopting AI tools and found that the data confirmed the savings: about four hours per week in reduced task time. But over the same period, she had taken on two additional client projects, joined a committee she wouldn't previously have had time for, and expanded her content output to take advantage of the new production capacity. The four hours were real. They were also gone, immediately and invisibly, into expanded scope rather than into margin.

Most conversations about AI and productivity focus on task speed: how much faster can a draft, a report, a piece of research get done. That's a reasonable place to focus, since task speed is visible and easy to measure. You can time it. You can compare before and after. The gains are concrete. But task speed isn't where the real leverage is anymore, for a specific and important reason. When AI compresses task time across the board, the bottleneck in most workflows moves somewhere task speed can't reach: the speed at which decisions get made about what to do next. Decision speed, not task speed, is quietly becoming the more important variable, and it's not showing up on anyone's task list because it was never a task to begin with. ------------- Context ------------- Think about what a typical AI-assisted workflow actually looks like now. A draft that used to take two hours takes fifteen minutes. Research that used to take an afternoon takes twenty minutes. The execution layer of most knowledge work has compressed dramatically. What hasn't compressed at the same rate is the layer above execution: deciding what to work on, evaluating whether a direction is right, choosing between options, determining when something is good enough to move forward. This layer was always there. Before AI, it was partially hidden inside the execution time. Deciding what a report should argue happened, in part, while writing it. Deciding which research direction to pursue happened, in part, while doing the research. The thinking and the doing were intertwined, and the total time included both. Now that doing has compressed dramatically, the thinking that used to be embedded in it has to happen more explicitly and more separately. And for a lot of people, that thinking hasn't gotten any faster. It's the same deliberative process it always was, but it's now a larger proportion of the total time a piece of work takes, and it's often the part that isn't being tracked or improved at all. ------------- The Bottleneck Moved, and Most People Haven't Noticed -------------