Operator · Builder · Theorist

I work inside the machine I keep warning you about.

Twenty years in enterprise software: a paper route at nine, the executive floor eventually, deploying enterprise AI before it was a headline. I know how that sounds. Stay with me here, because it’s the whole reason I have anything worth saying: the clearest view of how technological power actually gets distributed is from inside the room where it happens. I’ve been in that room my entire working life.

01 The Operator

Yes, I still work there. That’s the point.

For two decades I’ve worked inside the optimization machine (customer experience, technical account management, enterprise digital transformation), and today I lead a technical-account team at Zendesk, guiding large enterprises through the shift to AI-first support.

Here’s the part most people skip. My employer isn’t one of the handful of companies that own this. We pay them. A lot. Almost everyone is a tenant; that’s the point. The concentration reaches up and taxes even the winners.

The guy selling the stuff is warning you about the stuff. I know. But I’ve read the invoices, and that’s exactly why I can see it.

So I name my own position first, every time: I’m part of this. And the critique isn’t aimed at the thousands of companies and millions of people downstream who rent the capability. It’s aimed at the handful who own it.

Receipts: Six Sigma Black Belt · 15+ years leading distributed teams
Wrote The Technical Account Manager’s Playbook · views are my own, not my employer’s.

02 The Theorist

The Inversion

This is the paper my whole education was training me to write: communication and media studies, development economics, and the AI frontier, finally pointed at one question.

For two hundred years, every major information technology eventually leaked the power to create back out to ordinary people. The press, the telephone, the internet: monopolized for a while, then the capability diffused. I argue artificial intelligence is the first to run the other way, because its core input, frontier compute, is scarce, capital-intensive, and owned by a handful of firms. Cheap access to use a model is not possession of the means to make one.

The argument is built to be proven wrong: it names the exact conditions that would do it. If I’m wrong, I want to know exactly when, and how, and I’d honestly be relieved.

→ Read the working paper → The open questions

03 The Builder

Then I go home and build the alternative.

The thesis says capability is being concentrated, and that the way out has to be built rather than waited for. So that’s my nights and weekends. Chump is a self-hosted AI coding agent with persistent memory and bounded autonomy. It runs on local models, on your own rails: your keys, your data, your compute. No tenancy required. (I built agent platforms five times before this one worked. Named it Chump on purpose.)

The full workshop (Chump’s architecture, the open-source work, the rest of it) lives at the builder’s site.

→ jeffadkins.dev

04 Writing

Connective Intelligence

I write wherever the thinking goes: the craft of technical account management on the days it’s that, the political economy of AI on the days it’s that, and late-night quantum rabbit holes on the days I really should have gone to bed.

→ Read at Connective Intelligence