From Emacs to Agents

Vivek Haldar

May 21, 2026

From Emacs to Agents

I used to be a heavy Emacs user. Note the past tense.

Over the last few months my Emacs usage has fallen off a cliff, and I want to talk about why, because there’s a general lesson in it.

My Emacs Commits Flatlined

Like every other Emacs geek, I used to spend inordinate amounts of time chiseling away at my config. That activity has died down over the last couple of years. My last commit to that repo was October 2025.

You can probably guess the reason: agents and AI. But to understand why a heavy Emacs user would walk away, it helps to remember what Emacs was giving me in the first place.

Why Emacs Was My Computing Cockpit

The original reason I was so into Emacs — other than its sheer elegance — was that I was constantly fashioning it into my first-class computing cockpit. I used it as my primary interface for most of my work, because as a developer most of my work was text-based. Occasionally I’d break out into an IDE or a browser, but I tried to do as much as I could through Emacs.

The reason that worked is that Emacs had:

A nice interaction model.

Emacs Lisp for extensibility.

With those two, you can shape the tool to fit your needs and implement your custom workflows entirely inside Emacs. Then there are the tools layered on top — Magit, a far nicer Git interface than the command line, and elfeed, the RSS reader I still use.

Agents Became My New Entry Point

The big change is that my usage shifted to agents. My personalized Claude and Codex setup is now my primary entry point to the computer. I delegate as much as possible to the agent, and the fraction of my work it can absorb keeps growing.

When I need a UI or visual output, I’ll have the agent create a single-page HTML file or a throwaway app tailored to that one query. Everything is getting subsumed into agents. Over time I’ve built up a growing collection of skills that capture the workflows I run frequently.

Output Parabolic, Tool Count Shrinking

Over the same period, my output — measured by GitHub commits — has gone parabolic. I suspect this is a pattern for many people: output going parabolic, and yet their tool count has shrunk over the last few months in favor of channeling everything through agents.

That’s the interesting inversion. The Emacs era was about expanding one tool to subsume everything. The agent era is about collapsing every tool into one interface — but the interface is a conversation, not a buffer. (Chat isn’t the ideal interface for everything, but that’s a separate post.)

This Won’t Stay Just a Developer Thing

This is mostly a developer phenomenon today, but most of knowledge work is heading the same way. Signs of it are already visible. As generative UI matures — models getting faster and better at producing throwaway, highly customized, one-time-use UIs — the shift will accelerate.

It’ll reduce tool sprawl — the number of interfaces and SaaS subscriptions people maintain — in favor of agents with their highly customized throwaway generative UI. Why pay for a tool when the agent can render exactly the surface you need for the next ten minutes, then throw it away?

Nostalgia, But the Times Are Changing

I feel a tinge of nostalgia. Emacs was great. I loved using it. I loved tweaking it. I loved procrastinating with it.

But the times, they are a-changing.