Keeping up
It has never been easier to feel behind. Every week there's a new model, a new terminal, a new agentic tool, a new workflow builder. I stopped treating that feeling as a reading problem. For me, keeping up is a building problem.
What I believe is simple: sit down at night and make something. Set it up, break it, set it up again. Go to bed knowing a little more than I did that morning. Not consuming the feed. On most days, deliberately, getting a bit more capable.
But to be honest, my evenings weren't always like that. People who know me know this: I was over 120 kilos, and my blood work had peaked too. By the end of the day I didn't have the energy to stay on my feet. I had plenty of ideas I could build and far too many products and technologies I wanted to try, but I was in no shape to sit down at the desk in the evening. For a long time I called it "no time"; what was actually missing wasn't time, it was energy. For the last two months, with doctor check-ins, treatment and a workout routine, I've started finding myself again, and today I spend far more of my evenings building than I used to. The nice part is that the process also left me with a really good app idea, but that's a story for another post.
I'm not saying this as the guy lying on the couch anymore. I work on a DGX Spark: I stand up open-weight models, try different quantization options, and see the memory and throughput limits with my own eyes. I've written before about why I spend that time on local AI: control, insurance, not getting locked in. In short, what I still believe is this: speed can now be bought; anyone can rent the fastest model. But nobody can rent you the meaning of what you're building or the ownership of what you've set up. Those only become yours by building and running them yourself.

If I really want to learn something, I have to sit down and use it myself; I only learn how it behaves and where it struggles by trying it again and again. Reading release notes isn't the same as feeling a model be wrong. I keep up because I sit with the tools until they do real work, or until I get to a point where I can name why they broke.
But constant trying has a trap too. Most of the hours that used to go to tabs and feeds now go to something that looks more serious: prompt, wait, look, fix, again. At first I was really excited. Then I noticed the day getting narrower. The output comes back almost ready, so one more round always feels earned. It looks like practice. Most of the time it isn't.
I've seen many times how fragile knowledge is when I memorize it without understanding; it falls apart at the first different question. That's the risk of the loop. You can ship the almost-ready thing and still not be able to say why it works. The loop can hold you; but it was also what pulled me back into building after a long break (Filling the gaps). What I do about it is simple. I tie the reward to an outcome verified outside, not to how many rounds I've spun: a user, a payment, a decision. I set hours when the tool is off. Before I pick up the phone, I name why.
I've accepted that learning something deeply is slow. Rushing to get a head start usually leaves the learning on the surface. I don't need the newest model on day one. I need to understand what's on the table.
Let me get more concrete. In the first week after the iPhone Duo came out, I only studied it in the simulator (thanks, Apple, for shipping the first version broken) and looked at the design principles of Apple's own apps. How should a design that works on one screen behave when it opens across two? But I couldn't learn that just by looking. I started opening Claude and Figma side by side and trying different design practices: drawing the same screen in several directions, thinking in adaptive layouts from the start instead of fixed screen sizes, throwing out what I didn't like and starting over. In one evening I can see as many variations as used to take me a week. And what I really learned isn't the tool; it's noticing faster what I don't like and why.

Something similar to the Duo happened with Grok Bot. I use Grok Bot actively every day, and after a while I started thinking an organization should have a Grok Bot running locally inside it. While I was building something along those lines, Open Bot came out. I found myself going through its setup steps. The onboarding was genuinely easy. But a bot I'd installed locally talking to Open Bot's own systems made me uneasy. I didn't dig too deep; still, the moment I tried to install it on our internal, offline server, it immediately wanted to reach an outside network. So I decided to try it on a separate Mac instead, and went back to building my own. What I took from it: in what I'm building, working offline can't be an option. It has to be the default.
For me the real point isn't developing the software; it's knowing how things are done. Keeping up isn't a reading list; it's keeping the energy and curiosity to sit down in the evening, break something and build it again. For a product-minded developer, I think that's what matters most for keeping up: trying, trying, and trying again.