I wasted my one kid-free morning with my wife at a coffee shop explaining what Gas Town is.
To be fair it seemed like a suitable addendum to our conversation the night before in the car about the actual mechanics of LLMs and the attraction of the 'magic' factor, which undoubtedly exists (it's not all hype).
Something bothered me about Gas Town that I couldn't put a finger on. Haunted me, maybe. There's a spectacle to it, an awe factor. It's not something a regular person would even think to build, and not something an irregular person would be able to build, until now (assuming it really works).
I was searching, through our conversation, for the source of the uneasy feeling I had about the existence of this... abomination. Why did I feel that way? Was it personal bias, or am I being carried along by an anti-AI community sentiment which which I'm sympathetic? Or is my intuition picking up on something?
I think at least part of the answer lies in how projects like Gas Town frame the act of creation, and the ideal that they strive for. It's boss shit. Pejoratively.
God Complex
The reason Gas Town's creator, Steve Yegge, believes it's the future is because it will allow "one creator [to] make a product [...] that would have taken a whole company before." (Not linking to the source of this quote since Yegge has stated visibility personally contributes to his house stake in the gambling occurring on his idea, and I've benefitted him enough on that account already just by referring to it. Side note: really loving the attention economy; this is going great!) (Update: now that the pump-and-dump moved to the dump phase, I am indeed linking the post and also this chaser so you can share in some schadenfreude).
This is the dominant promise behind most AI pitches: Become a Creator. In Yegge's view, the mechanism behind this will be to assume the role of Overseer to direct a whole hierarchy of virtual drones. You, the Creator, are the one with the vision. The drones do the work. At last, in 2026, the Idea Guy will take his rightful throne.
Here's what I think the ick intuition I had about all this was getting at. These AI Creator projects are reusing the same framing of work and industry as the worst archetypal bosses. Well before the introduction of virtual serfs, our cult of industry pictured the CEO as the visionary genius, and the corporation as his will made manifest. (You might say it's the echo of feudalism and slavery, still reverberating, looking for the right material to resonate again. Maybe that seems too uncharitable to you, just musing).
It is the handful of bosses who decide what problems get solved, and how. The role of the worker is to implement. To enshrine this hierarchy we note that production requires capital, and capital is privately owned; therefore the handful who accumulate this capital by any fashion get to be the bosses.
The problem is not that AI Creator types didn't identify this as problematic (well, some of them at least). It's more that they pinpointed the problem as "handful of bosses" and went full speed ahead trying to turn that into "everyone can be a boss." When in fact this was already as true as it probably ever will be. Anyone can be a Big Boss, provided they possess the will to dominate, extract, and coerce; provided they have enough hubris to believe they know what's best; and provided they're quite lucky. I'm not convinced that technology is going to meaningfully change that equation, as long as it lives under the dome of capitalism.
Petit Boss
I think the real fallacy here is believing that this boss model works at all. I would, rather, suggest that the effectiveness of industry - even in the throes of the greatest Boss Cult waves - comes from the skill of workers. Almost like workers produce value. It rings a bell. I dunno.
Having been a worker all my professional life, I've seen this enough firsthand. I've had the enjoyable experience of working for good managers and bosses, who took a collaborative and bidirectional role alongside their teams. Experiencing that trust and agency helped me deliver work that exceeded all of our horizons. Good leadership helped direct me on the what and why of building. But along the way of deciding how, the what (and even why) were open to influence by me, too. This led in amazing and unexpected directions, for everyone.
I don't think bosses shouldn't exist, necessarily (though I'm open to systems which imagine this). Rather I think that the position of "boss" works most effectively when it exists in a cohesive web of feedback across the whole team. It's when the boss begins to see themselves as the only agent - the only person in the equation, that things begin to fall apart. And AI Creator mindset is the distillation of that. You are, in fact, the only person left in Gas Town (maybe this is what made me describe it as 'haunting'). Why is it you think this is going to go so much better than that time your terrible boss demanded you shut up and build what he asked you to build? Just because you're the boss, now?
We have to stop thinking of bosses as a caste of people. When you step into the role of Lord Overseer in your serf village federated agentic mesh, I'm not sure (like I'm actually not sure) you get to keep your "worker" badge anymore. Maybe this is why the "Creator" label has been appropriated, to obscure this transformation. You are putting down your tools and becoming a Boss.
Where's the line?
I know the framing is that Agentic AI is the new tool. I think this is just the part that I don't buy. Because it's also framed as a factory. It can't be both. And, personally, one of the great joys of my life is to use good tools (and make them), and I know what that feels like. It's not the same. Agentic AI is delegation, not participation (in contrast, non-agentic LLMs do feel more tool-shaped to me; perhaps this is bias or contradiction? I'm not above it.)
This is, in fact, why I do find a use every now and again for Agentic AI. There are certain mechanical tasks which I feel comfortable delegating. Ones which bring neither joy nor insight; things which I have done a million times before and just haven't found a way to abstract for reuse. Things I might pass on to a more junior team member instead if I'm working on a team (there is still something in it for them to learn and gain experience from).
This serves as an excellent foothold to counter-argue this essay. Let's briefly sketch that out.
Why (you might say) can't we extrapolate that into higher and higher levels of building? Perhaps I call on an Agentic AI to create a component which formats the localized time, for use in a blog post header. Solved problem. Delegated. Isn't there also a point, in my journey as a maker, where it's time to delegate the creation of that header, too? What about the entire blog structure?
Well, what about the blog posts, themselves?
Is there a point in my journey as a maker, where I graduate from making things to being a person who has made things? Where I have so mastered every lesson of the making process that I no longer need to make? Where I am entitled to simply speak, "let there be a blog," and behold, there is a blog?
To speak a word and, by some obscured and delegated work, bring forth a creation - my argument is that this is no longer the process of making. At some point it tips the scale. I'm a boss now.
To put it another way, I am not advocating that everyone be a pure artisan, a Worker par excellence. Delegation is a component of practical and effective work. But maxing out on delegation turns you into a boss. That's just what a boss is. A new technology of delegation doesn't change that, it just lets more people become bosses.
Like I said, I don't think bosses shouldn't exist. I just don't think "boss" is the final evolution of "craftsperson." And it seems like that's what people are assuming here. If we convert all craftspeople to bosses, we may find real production and innovation grinds to a halt. Perhaps I'll have to argue that point elsewhere, but for now I'll just assert that craft should continue to be a thing humans practice.
What, then, is the path of the craftsperson? How can I continue to level up as a worker without branching into the Boss skill tree? This isn't a new question at all. And there are two dimensions, at least, to explore.
Expert
First, I think the actual depth of craft as such is much deeper than people realize. And reaching for delegation before grasping this is a sure way to miss that entire space. If you've worked as a craftsperson under a boss before, you've probably seen a version of "this blindness." Your boss says something like, "oh, and it needs to do this," and you spend the next hour digging into specification of how this can be done. Perhaps they dive in with you and follow along in curiosity as you examine the approaches and trade-offs (good job, boss). Perhaps they shrug and say, "just figure it out." Either way you, as the worker, are responsible for understanding a fractal detail which the boss will continue to abstract as this when they conceptualize the output after they leave the room.
The fallacy here is thinking that this blindness won't happen to you, too.
As workers we have, historically, been forced not to brush all those thises aside. It's our job to figure out how to do this. This is the engine that drives expertise. A skilled generalist will learn how to do a couple thises and compose the work of specialists for the ones they don't understand (this might be more on the "boss" skill track). A specialist may spend a lifetime on a single this and its sub-thises.
Every single one holds the opportunity for innovation. New insight. Digging deeper into a problem than anyone has before. Stuff that's not in the training data. Combining two problems into one solution which no one thought of. New trade-offs, new behaviors, new variables. All of that can be lost if you see a forest and not the trees.
Spending time to carefully craft, iterate, fail, and finally produce a working system is a function with two outputs: the system, and expertise. When people claim LLMs can produce the same output, they seem to ignore the latter. I guess the assumption is you can offload it back to the AI.
So the first way to level up as a craftsperson is to dig in. Refuse the call to back away and delegate. Treat each this as a chance to deepen your expertise. Not just to deliver a product, but to shape who you are.
Collaborator
But that's not the only thing. I've noted the upward of delegation and the downward of specialization and expertise. There's another dimension: reaching outward in collaboration.
The key point is, collaboration is not delegation. Collaboration is an invitation to diversified thought. A delegate follows orders, a collaborator reinvents, re-interprets, and even defies them.
It's honestly frustrating that LLMs have captured the word "agent" because a collaborator is an agent in the way AI is specifically engineered not to be. A collaborator's first instinct on encountering your idea is not to enthusiastically praise it as "totally right." The best collaborators are invested skeptics.
A good collaboration is built on mutual respect. Your collaborator will see your approach and ask, "have you tried..." and they actually mean it as an open question. The person who asks that and secretly wants you to just go with their idea is a boss in disguise. The collaborator wants to see how their idea takes shape in your mind. They can't do that without you, and you can't do it without them. That's the magic.
A collaboration is not a competition of bosses trying to delegate to one another. Its goal is to leverage diversity of perspective, experience, and knowledge to produce a transcendent outcome. The work becomes a collective effort of which no one person could reasonably claim the title of Creator.
The reason I'm belaboring these distinctions is because the language of collaboration has been co-opted by bosses to talk about delegation. I just want to recall what we're really talking about here. My best bosses were actually collaborators: curious listeners, challenging me to broaden my perspective and try new approaches without ever mandating how I spend my time and effort.
Really, my biggest concern about the Agentic Engineering Revolution is that we lose track of real collaboration. That it will be such a huge buff to delegation that collaboration looks like a losing strategy. Why deal with the complexity and give-and-take of true, personal collaboration... when mechanically, you can go much faster with the power of our new delegation machines?
Why go together when you could go so much faster alone? Gosh, feels familiar, is there some kind of... aphorism... about that...
The future of...
So I suppose I should sum it up. I don't like being a hardline nay-sayer about any particular thing. That's not how I think about the world. My convictions are based in context. I'm more interested in saying this is the wrong tool for that instead of this tool is evil.
So maybe what I'd like to say is this: it's best to mentally downgrade claims like "Agentic AI is the future of work." This is too broad a claim. We don't have to attack and refute the whole enterprise to respond to it. We are not stuck in a dichotomy.
The mental downgrade is more or less to: "Agentic AI may be the future of going fast alone." There are lots of times you want to go fast alone! Assuming other problems with this model (cough, energy reliability security safety) are eventually resolved, we may be left with something genuinely quite useful, when it comes to going alone. But there is a reason you don't always go alone. And just because going alone might potentially now scale up to the work output of an entire company, does not mean that it will transcend the intrinsic equation of going alone. Can one person really ship a good product, alone? Not just mechanically, but holistically? Maybe some people can. But it's worth meditating on what usually happens to, say, the solo projects of band leaders. Or when popular authors finally shake off their editors.
So before buying into "the future," think about what model of production this future is enshrining, and consider whether perfecting that model is actually perfecting production itself. This is also a great opportunity to think about what different models we could strive toward instead.
And maybe sit with that a bit before dumping your savings into $GAS or NVDA or whatever.
The real conclusion
But I forgot why I was writing this in the first place. I was not just trying to make a practical observation about not believing the hype, I wanted to say something about work itself and our relationship to it.
I believe effort is important. Didn't you hear? In 2026 we're friction-maxing. But really, encountering obstacles and overcoming them is so key to self-formation. I changed my mind while writing this essay. It's a big part of why I write essays! Not just to be someone who wrote an essay. Not even mostly that.
And self-formation never stops. I don't think it's an accident that all the richest guys are sociopaths. Lots of people think it's because being a sociopath gives you money-accumulation superpowers. That's possible, but also, having all that money - all that delegation power - makes it so easy to lose touch with the real labor of doing things. In fact I worry that excessive delegation is, itself, a kind of spiritual rot. Making delegation power more accessible to the masses might not be empowering good makers to go change the world - it might in fact be corrupting good makers into bad bosses.
With our current wealth and influence, workers of the world are forced to specialize, and to collaborate. This is how we deliver, how we grow, how we find our niche and thrive. Suppose they really succeed at building each of us a personal drone army to delegate to, instead. When we gain that power, what do we lose? No longer driven, by sheer physical need, to work together, how will we not become more alienated? More myopic? More competitive? More factional? More mistrusting? Not rhetorical questions - if this really is inevitable, please let's actually answer them. Jury's still out on that though.
Don't lose sight of your values. Even if they have a sledgehammer to smash the current broken systems, ask what will be built on the rubble before taking a swing. Worse worlds are possible. But better ones are, too.