Published: Nov 16, 2026

Agentic Programming is Software Engineering

I haven’t written anything here since my end of 2025 post, and certainly a lot has happened. My daughter was born on New Year’s day, which is a fantastic birthdate to celebrate every year. And I left my job in June, which has given me an opportunity to really dive into agentic programming in a big way. Ideally I’d like to turn this experience into a startup, but barring that I’m actually interested in going back to an IC role as this has renewed a love of programming that I hadn’t felt in a long time.

Programming was never an art form

A lot has been written elsewhere lamenting the loss of something due to LLMs writing code. I certainly understand the existential dread a lot of people feel since these tools really are a fundamental upheaval of many people’s careers; where decades of knowledege and expertise are depreciating quickly in some ways that makes the future uncertain.

But the claim that programming is really an art form that captures human expression is just bullshit, and is especially bullshit when regarding code written for work. I’ve worked with a lot of fantastic engineers in the past 20 years, and I have never seen a PR that stunned me with its beauty or made me consider its deeper meaning. The best I could muster during code reviews was a quick glance for obvious bugs, and if I didn’t find them our production alarms generally would. There’s nothing deep to explore here, most products don’t involve blazing a trail of new research, and almost all code ever written is a regurgitation of existing patterns and ideas. Sure LLMs write derivative code, with shitty abstractions, and sprinkled with bugs; but pretending that it was any different before AI came on the scene is just dishonest.

So why is Agentic Programming so exciting?

We’ve established the LLMs have the same faults as people, but I’m fascinated by agentic programming because of the scale of it. Ultimately the reason I went into engineering management was because I wanted to amplify my impact. I’m a good engineer, but I’ll never be as good as a dozen good engineers; so I wanted to find a way to have as much influence as possible over technical decision making even if it meant an abstraction of implementation that I’d rarely see behind. LLMs let me have that same model of development with the output of 100 people, and with none of the other tedious aspects of management. I get to be an IC again, but with unlimited technical influence and output.

But it’s more than that, I’m a huge advocate for “vibe coding” because it kicks the gates open to vastly more people being able to bring their own ideas to life. The scribes had nothing good to say about the printing press, but making books available to everyone was objectively fantastic for humanity.

I feel incredibly fortunate in this moment because my career path has made me especially well prepared to leverage agentic programming, mostly because I really did touch every part of the stack (frontend, backend, infra and devtools), and I’ve worked on all kinds of products at different scales (telecom backbone, broadcast voip, e-commerce, mobile, sensor-fusion AI systems, enterprise SaaS, etc). It turns out that if you want to build really big and complex things with agents you need to have strong opinions on patterns/contracts and the knowledge to verify them to keep the system aligned.

How is this Software Engineering?

Software Engineering as a discipline has always been about complex systems, specifications, and outcomes; and LLMs accelerate that kind of development in some amazing ways. Most programmers derive satisfaction from point solutions that are often narrow in scope but deep on technical understanding. In the world of agentic programming the details are more abstract: complexity, resilience, scaling, extensibility, and all of the other challenges that come with building big systems. The technical understanding required is wider and in many ways even more important since the consequences of those decisions have broader impact.

Like I mentioned above I feel like I have a 100 person dev team working for me now, and I’ve easily generated 10x more code in the past year than the rest of my career combined (and that’s if I’m being charitable to the rest of my career). Keeping that output coherent is a different kind of technical challenge, and it’s a similar level of abstraction to being a software architect but with a much tighter iteration loop. I can look at the code but I rarely do, instead I look at API contracts and Postgres schemas and Kubernetes configs. I’m far more interested in the boundaries of the system than the internals (and where to define these boundaries is a separate design challenge). It’s fun to imagine the largest and most complex system you’ve ever worked on and think of ways you could build it from scratch, by yourself, with these tools.

I’ve experimented enough to find some useful patterns for agentic programming, perhaps I’ll write about them at some point. But regardless of that I hope you discover the same sense of wonder with these tools as I have, and do not mourn the programming discipline of 5 years ago. The profession is heading in an exciting new direction, and the empowerment that comes from these tools is extremely exciting!