Agency is not about agents

I am tired of the AI content economy. Every feed right now is full of posts about how to use agents, how to prompt better, how some workflow will change your life, usually illustrated with an AI-generated image and capped with a link to a course. A lot of people are trying to sell you a better way to use AI.

Here is the funny part: the people who actually use AI well don't need any of it. They just try things until they figure them out. The models change monthly, so every "definitive guide" is stale by the time it arrives. These posts aren't education. They're clickbait. That's why the real way to learn something is to play, explore, figure it out, and communicate with other people, not pay someone for a course who probably isn't even qualified to tell you what AI can actually do.

I say this as someone who has spent a career applying ML/AI technology to real-world applications, not commenting on it. I built an AI-powered transit signal-priority and emergency-vehicle management system at LYT. I patented an AI power-management system for buildings. Right now I'm helping a company build foundation models that derive meaningful inferences from neural signals in plants, which is some of the most interesting applied AI I've seen in years. I run agents for hours at a time in my own work. I am not a skeptic of the technology. I am a skeptic of the theater around it.

The tool is not the craftsman

Agency is not about agents. It's about a person's ability to orchestrate software, and that is not new. It's what we've been doing for decades. I like to joke that the best engineers are lazy; they will go to great lengths to make things easier.  

Mastery of Premiere versus fumbling through Final Cut can be the difference between a six-day edit and a three-week one. Give someone who has never touched wood a fully equipped woodshop, and they'll produce junk for a long time. Give a woodworker with ten years of experience that same shop and they'll build something remarkable in a day. The shop was never the point. The tool isn't amazing. The craftsman is.

AI is the best-equipped woodshop ever built, handed to everyone at once. That's exactly why the output of most people using it looks sloppy.

What the tool actually does

In my own work, roughly half of what AI hands me is wrong in some way. If I weren't a domain expert, or didn't spend the time cross-referencing and actually reading the sources it cites, I would be publishing the same confident nonsense I'm complaining about.

Take fundraising, since "let my AI team fundraise for you" is a pitch I keep seeing, usually from people who have never raised a dollar. Raising money is not about automated emails landing in inboxes. It's warm introductions and relationships. If you can't get a warm intro to a top investor, you are probably not getting a check, even with exceptional metrics. Anyone who believes AI has fundamentally changed that doesn't understand how trust works between humans.

A CEO recently argued on a podcast that the future belongs to 10 times as many people who can multiply their output a hundredfold with AI. I think that's roughly right and describes the idea of a 10x engineer. But notice what it requires: the foundational knowledge and experience have to already be there. AI doesn't make me smarter. It clears the busywork so my existing judgment goes further. It multiplies output for people qualified to filter it, and it multiplies garbage for everyone else. Same tool, opposite results.

We have seen this movie

I've watched more than ten years of self-driving car promises, built by genuinely brilliant engineers grinding on some of the hardest software and hardware problems on the planet. The research goes back decades, to labs in Europe and at UC Berkeley, and we are still only on the precipice. Transformative technology is real, and it is also slow. Those two facts coexist.

The same will likely be true for AGI: I have yet to see anything that suggests sentience. What we have is a statistical model, an enormous file of weights mapping inputs to outputs through gradient descent. Extraordinarily useful. Not a mind. My hunch is that anything genuinely close to general intelligence will look more biological than anything we're building today. That is a longer conversation for another post.

A friend of mine put it well recently: innovation theater has its pros and cons. As a founder, I understand the value of making noise to get attention. It attracts talent, capital, and customers, all needed for companies to grow. But there's a difference between making noise about what you're building and lying about what the tool can do to sell a lie or a half-hearted course on A.I. to capture people who are afraid they are being left behind. The wild part is that AI has a bad wrap because most people worldwide are afraid it will replace them. 

The layoff excuse

Which brings me to the claim that AI is replacing jobs. Mostly, it isn't. It's a convenient explanation for layoffs that were already going to happen, one that lets executives skip the harder conversation about why. If AI were genuinely the driver, we'd be seeing it in productivity data before we saw it in press releases.

What AI actually does for most workers is simpler and less flattering: it increases the volume of what they produce without improving the quality. More documents, more code, more posts, more slop. That's the pattern underneath every hype cycle, but it takes a long time, lots of hard work, and attention to detail for everything to come together to build truly lasting impact at scale.

So let's talk about AI the way we talk about every transformative tool in history, from software to the internet: honestly. It is a genuine step change, similar to a compiler but way more generalized. I won't discredit that. But it rewards people who already know what they're doing in specific niches, and it exposes everyone else trying to find a shortcut.

For what it's worth, I used AI to help edit this essay. That's the whole point. The tool helped me say this faster. It didn't decide what was worth saying.

Volume is not quality, and quality is not output; there may be a point in time when physical agents and AI will make us reconsider society as a whole, and that is teh more interesting conversation.