Air Traffic Control

Keith Adams, welcome to Still Burning, where we talk to geeks who still care and are still doing something about it. And I can't think of anybody geekier than you who cares more and is still doing so much about it.

It's a huge pleasure being here with you. We have a little bit of history, and this isn't the first time we talked in front of a fire at dusk about software. And we'll see if we can keep it PG-13 this time.

So when did you first know that you were a geek?

Oh wow. Well, so I wasn't exposed to geekdom — I was brought up in Reagan's America, so for me, geek was a term of abuse. I probably found out I was a geek from all the people telling me I was a geek as I was being shoved into lockers and held upside down by my ankles. So I think it was a little bit thrust upon me. But I think I've liked computers ever since I knew they were a thing. My dad was an officer in the Marine Corps. He bought an Apple IIe in 1983 — take-home pay for his grade was $30,000 and the sticker price was $3,000. And if you're an artillery officer, there's not that many things an Apple IIe does for your job. But he just saw the potential of these things and was excited about them. As soon as that thing was in the house, I kind of wanted to know everything about it.

It seemed like this portal that directly connected the world of things — it's obviously this hot object that makes noise in the room and does stuff — and the world of ideas. The software part of it was almost the pure stuff of thought in a strange way. I kind of temperamentally ended up being a systems person for most of my career.

Talk a little bit about what you're doing now and how that transition happened.

I'm about four years into a career transition into venture capital. I was really fortunate to visit all my dreams as a software guy. I was a relatively early person at VMware — there from 2000 to 2009, got to do a bunch of cool hardware-software interface stuff. Was relatively early at Facebook where we met in the early teens — built HHVM there, the JIT compiler that was running the application for the big blue site at the time. And then was part of the generation of people who got fired up about deep learning in 2013 because of what was going on in the ImageNet competition — the first rumblings of the AI revolution we're living through. I brought a little demo to Zuck's calendar, and to his credit he understood what we were doing and gave us a ton of resources to build out what became the AI lab with Yann LeCun. Then I was at Slack for about four years as chief architect, Slack went public, and I left to figure out what was next. That turned into spending time with early-stage founders doing something with a technical insight, some connection between fields or a trick nobody else had figured out.

So the SaaS playbook era. This was a long, stable period where a set of metrics fully characterized a software business — LTV, CAC, churn — and you could build an organization that executed that playbook without much technical risk.

Exactly. And in that world, taking on technical risk was often irrational. If you could just execute on a proven model, why take the chance? The playbook was stable enough that execution beat innovation. That era is over now. The playbook went blank.

So when did that happen?

The revolution I'm referring to started in undeniable terms in November of 2025 with the release of Opus 4.5. Prior to that, you still saw people who were resisting the use of coding agents and could kind of justify it. After that, I think for people of a serious disposition — if you're ignoring those tools entirely post November of 2025 — I think you're fooling yourself.

Edward Yarden in 1994 wrote "The Decline and Fall of the American Programmer." He was right about the trends and wrong about the consequences. The programming profession has survived and expanded through multiple waves of "this will replace programmers." How do you think about that pattern now?

The honest answer is I don't know. Software is more like roads than like coal — build more roads and you generate more travel, not less. The classic Jevons paradox. And we've seen that play out over and over. Offshoring didn't kill the profession, it expanded it. Better languages didn't kill it. And the counterintuitive thing is that all of this said — everybody I know, myself included — the more you use these tools, the busier you seem to be. That sort of suggests there's not a lump of software out there we just need to churn through.

Thirty years worth of "yeah, it's probably not worth it" is all in play again.

That's a really good way to put it. And I'm realizing how ossified that circuit of my own brain is — how pre-verbally, before I even fully form the thought of the thing I'd like to try, that restricting loop fires: "That's too hard. You're never going to — it would take you a thousand hours to even find out. Don't even worry about it." Realizing that you can relax that muscle has been incredibly liberating.

I think it's incredibly invigorating to realize we get to live through a time where all of this stuff I wasn't sure would ever be built is going to be built for sure. We have a blank sheet of paper collectively as a profession right now to figure out how we do business.

Is AI leveling the field or widening the gap?

I thought it would democratize programming — raise the floor for everyone. What I'm actually observing is the opposite. The people who I know who are most engaged in a creative loop right now are the people who are virtuoso creators already. It seems like if anything this has been more of an amplifier of differences in facility than a leveler. The billion-dollar lever is going to people who were already building things at speed. I want to be wrong about this. I hope the democratizing effect catches up. But right now the gap is widening, not narrowing.

What does "code legibility for alien minds" mean?

Programming language design has always been applied psychology — how do you design notation that fits the way human minds reason? Now the question is what makes a codebase legible to something that doesn't reason the way we do. One possibility I find interesting: proof-carrying code. Code that comes annotated with formal proofs of its properties rather than readable source that humans have to interpret. If the agent can verify correctness against properties, the whole notion of "readable code" may become less important than "property-annotated code."

What you're suggesting is there is a moat and it's gigawatts.

That's right. A Pebblebed intern found CVEs in the Linux kernel by just running LLM-based security analysis continuously — with no apparent diminishing returns. If more compute always finds more bugs, software quality becomes proof-of-work. And proof-of-work concentrates advantage at whoever can afford the most compute. That's a concerning dynamic.

Programming doesn't have the flow state that it used to have anymore. The agent world feels more like being an air traffic controller — very interrupt-driven, trying to get productivity through breadth instead of through that kind of deep immersion. You're managing planes, not flying them. That's a genuinely different relationship with the work. Many people found the deep immersion to be an intrinsic reward. I'm not sure the air traffic control version produces that same reward, even as it produces more output.

I'm reading the output of eight agents at once while live streaming. It's like plate spinning. Every time one agent stops, I throw it something new to do.

That's exactly what it feels like. And for many of us who became programmers because of the deep, focused immersion in a problem, this is a real loss. It's a phenomenological cost of the transition that I don't think we talk about enough.

Last question. What about this situation scares you? What wakes you up at night?

For me it's bioterrorism. AI synthesis knowledge becoming broadly accessible at the same time as synthetic biology capabilities are improving. That strikes me as the most likely near-term catastrophe. The thing that worries me isn't rogue AI — it's AI in the hands of a small number of bad actors with access to synthesis capabilities.

Our society has a bunch of big problems to address. I don't worry about those. I worry about the meta-problem: we have lost the ability to address big problems. There are people who are willing to make a buck on the continuance of the problem. The economic incentives are leading to irreversible consequences. And we've lost the mechanisms — social shame, responsibility, religion, community, legacy — that used to stop people from creating irreversible harm just to make another buck. Those mechanisms are now broken and I don't know how we get them back.

Maybe this is old age talking, but I think the most desirable social characteristic became sort of "cool" instead of seriousness or gravity. The impulse to distrust authority — which is important and part of our intellectual tradition — became reflexive to the point where any exercise of authority seems illegitimate. And so if anyone amassing power to solve a big problem is subject to infinite critique, it does make it really hard to actually solve things. Keith, it's great talking to you.

Such a pleasure, man. Take care, buddy.