I Have a Team Now — It Just Happens to Be AI

Eight months ago, I was perpetually behind. On everything.

I don’t mean “busy.” Busy implies you’re making progress on too many things at once. I was making insufficient progress on all of them. React components for a client project. AWS infrastructure governance for another. Kubernetes migrations with hard deadlines. Salesforce automations that needed attention three weeks ago. Each domain had its own language, its own context, its own state — and switching between them wasn’t just a time cost. It was a cognitive tax that compounded with every transition.

By 3 PM most days, I wasn’t making decisions anymore. I was recovering from the last context switch while dreading the next one.

The email that started it

The first thing I used AI for — really used it, not just experimented — was writing emails. GPT-3.5. I would brain-dump everything I needed to communicate into a chat window — unstructured, grammatically questionable, half-formed thoughts — and get back something I could send after one or two editing passes.

That sounds trivial.

It wasn’t.

Email was consuming more cognitive bandwidth than I’d realized. Not the content — the composition. Translating technical context into stakeholder-appropriate language, structuring the message so the key points are at the top, and correcting tone where needed. Every email was a small act of translation, and I was writing dozens a day.

Offloading the composition freed up space I didn’t know I was missing. Not a lot — but enough to notice that the constraint wasn’t time: it was cognitive bandwidth.

From assistant to collaborator

With GPT-4, ChatGPT got better. I started using it for more than email — rapidly prototyping WordPress plugins, troubleshooting legacy code (especially the undocumented kind, which was most of it), and reasoning through architectural decisions where I needed a second opinion that wasn’t going to judge me for asking a question I should probably already know the answer to.

The shift was gradual. The AI went from “a tool I use” to “a collaborator I consult,” and the distinction matters. A tool does what you tell it; a collaborator helps you figure out what to tell it. The governance documents I’d started writing — almost accidentally, just experimenting to get consistent output — were turning The Collaborator into something more reliable.

Something that remembered how I think.

The migration that proved it

About six months ago, I had to undertake a significant solo infrastructure migration. Hundreds of containers across multiple environments with a hard deadline driven by external constraints that weren’t negotiable.

The responsible estimate for this work — with a team of six experienced engineers — was six to nine months. I had three months.

And I was the team.

Were it not for ChatGPT, KiloCode, Cursor, and later Claude, I would not have been able to complete it. That is not a hyperbolic statement; it is not “it would have been harder.” I would literally not have been able to complete the migration within the constraints I was given - while still juggling my “regular work.” The project would have failed, or I would have.

Agentic AI enabled me to operate at a scale previously unavailable to a single person. Not because the AI wrote all the code — it didn’t. But because it could hold the context of each subsystem, while I focused on the decisions that actually needed a human. The infrastructure state, the dependency graphs, the rollback procedures — the AI held that, so I could hold and refine the strategy.

The case study tells the technical story. The human story is simpler: I shipped it — on time, no less — and I didn’t completely burn out doing it. Both of those outcomes were improbable without the tooling.

The Department

After the migration, I fully committed to Claude Code and started building what I now call The Department.

A site architect for this website — layouts, components, editorial, SEO
A sysadmin agent for infrastructure governance
A Project Manager that unifies my communication between Slack and Asana - and keeps me from missing things
An observability bot for monitoring
A content agent
A life-strategy agent
Several agents in charge of writing software, like Actions, Panoptisana, and the Markdown Editor I’m using to write and edit this post
And several more, besides

Each one has a defined role, governance documents, and institutional memory that persists across sessions.

The ability to context-switch without context-switching is the thing I didn’t know I needed.

When I need to work on infrastructure, I open the sysadmin agent. It knows the current state of every system I manage and what we did in the last session. It knows the conventions, the constraints, and the things I’ve told it not to touch. I don’t have to reconstruct any of that — I just pick up where I left off.

When I am working on this website, the site architect has the same depth in its domain (as well as LinkedIn). Different context, different conventions, different memory — but the same experience of walking into a room where someone already knows what’s going on.

The mental relief is almost too great to put into words. The thing that was destroying me — carrying the state of many different domains in my head simultaneously, losing pieces of each every time I switched — is the thing the agents handle. My working memory is freed for the decisions that actually need my judgment — strategy, architecture, ideation.

Everything else, the agents hold.

The inversion

One of the most significant personal findings of this process is that the output scaled because the cognitive load dropped. Not the other way around.

The conventional model is that more output requires more effort, more tracking, and more stress. You scale by working harder or hiring more people. The cognitive load tracks linearly (or worse) with the output.

The Department inverts — or perhaps, subverts — that: more domains under management, more projects shipping, and perhaps more importantly, the ability to rapidly switch between them without losing momentum. And all of that occurs with less cognitive overhead — because the overhead has been offloaded to agents whose entire job is holding the context I used to carry in my head.

It’s not just automation; I’m not replacing tasks I used to do manually — though I certainly do when it makes sense. It’s amplification — extending what I can hold and act on simultaneously. The decisions, strategy, and judgment calls are still mine, but the state-tracking, the context-holding, the “where was I?” recovery — that’s distributed across a team that doesn’t forget, doesn’t get tired, and doesn’t need me to repeat myself.

Still behind

I’m still behind.

I don’t think that will ever change. The scope of what I’m required to do always expands to fill (and slightly exceed) the capacity I have — that’s a result of employment and a feature of ambition, not solely a bug in the tooling.

But the texture of “behind” has changed. Eight months ago, behind meant drowning; it meant context switching so fast that I couldn’t maintain identity in any single domain. It meant 3 PM cognitive shutdowns and the creeping feeling that I was failing at everything simultaneously.

Now, behind means I have more projects than hours. The state of each one is held by an agent that’s ready when I am. The cognitive tax of switching is close to zero. And when I stop for the day, nothing is lost — it’s all documented, governed, and waiting for the next session.

And there are still domains that don’t have a door to agentic work yet — the ones where the process is opaque, sequential, and offers no meaningful feedback. Try getting a 10DLC campaign approved through Twilio when a denial comes back as “didn’t pass” with no further explanation. There’s nothing to reason about, nothing to architect. Just guess, resubmit, wait, repeat. Those still run on spite… if I have time.

I’m still behind. But I’m not losing my sanity in the process.

The management model behind the “AI department”: I Manage AI Agents the Way I Manage Teams. The governance methodology: What Is Pass@1? and The Governance Documents. The infrastructure migration: Kubernetes Migration.