From user to fleet operator — the four phases of working with AI

A founder showed me their AI setup recently. They had a notes doc open — a collection of carefully crafted prompts they’d built up over months. Copy this one for emails. Paste that one for summaries. A few variations for different clients.

“I’ve been doing this for over a year,” they said. “I probably save ten hours a week.”

That’s true. It’s also Phase 1.

Most founders are here. Using AI seriously, getting real value, but still the ones doing the work — just with a faster pen. The AI is a tool. The founder is the operator.

There are four distinct phases in how people work with AI. They’re not about technical skill. They’re about where you sit relative to the work.

Phase 1: User

You prompt. You review. You use it.

AI is a sophisticated input-output machine you’ve learned to operate. You know how to get decent results. You’ve built a library of prompts that work. You’re saving time on things that used to take longer.

The tell: you describe your AI usage with verbs like “I asked it to…” and “I used it to…” You’re the subject. The AI is the object.

The bottleneck: you. Everything still runs through your attention. You’ve made the execution faster, but nothing has moved off your plate.

Phase 2: Power User

You’ve gone deeper. Multi-step workflows. Chained requests. Knowing when to push back on a weak output, how to guide toward better results, when to switch approaches.

You’ve stopped treating it like a search engine and started treating it like a capable, if sometimes frustrating, collaborator. You can brief it properly. You know its failure modes.

The tell: you think about prompting as a craft. You’ve read things about it. You have opinions. You’ve taught colleagues how to do it better.

The bottleneck: still you, but less visibly. You’re not doing the manual work anymore — you’re supervising intelligent work. Still, every task still touches your attention. You’re a more efficient individual contributor.

Phase 3: System Architect

This is the first real phase shift.

You’ve stopped prompting for individual tasks and started building systems that prompt themselves. Recurring work gets automated. A new article draft triggers a workflow. A meeting note turns into action items without you opening a chat window. A weekly report assembles itself.

The AI isn’t being asked anymore. It’s been configured.

The tell: you use words like “workflow,” “trigger,” and “it just handles that.” You’ve written instructions files, built templates, set up automations. A growing slice of your recurring work runs without your involvement.

The bottleneck: design quality. The systems you build are as good as the thinking you put into them upfront. Poorly designed workflows break quietly. This phase rewards people who can think in systems.

Phase 4: Fleet Operator

Most people never get here. It looks strange from the outside.

You’re running multiple AI agents as a team. Different agents for different functions — one handles research, one handles drafting, one handles publishing, one coordinates the others. They communicate, hand off work, flag blockers, report results. You read summaries in the morning of what ran overnight.

Your job is judgment and direction, not execution. You set the agenda, review the outputs that need reviewing, make the calls that require human judgment. The AI handles the rest.

The tell: you describe your AI usage in terms of “my team” or “it ran while I was away.” You’re the operator at the top of an org chart where most of the org is AI.

The bottleneck: your ability to design governance. These systems need oversight structures, not babysitting. The question isn’t “can I get Claude to do this?” — it’s “what does this workflow need to be trustworthy without my constant attention?”

Where are you?

Be honest about it.

Most founders who think they’re at Phase 3 are at Phase 2. They’ve learned to use AI well, but they’re still in the loop for every output. Nothing runs without them.

The diagnostic isn’t about sophistication — it’s about whether things happen when you’re not looking. Phase 3 work outputs things you didn’t manually trigger. Phase 4 runs entire workflows while you sleep.

The move to the next phase

1 → 2: Stop prompting for outputs. Start prompting for drafts you edit, not drafts you accept. The shift is from using AI as a faster version of yourself to using it as a capable first-pass.

2 → 3: Pick one recurring task you do every week. Automate it fully — not semi-automate, automate. The first one is the hardest. After that you see the pattern.

3 → 4: Stop automating tasks and start designing workflows. A task is a single thing. A workflow is a sequence with handoffs, conditions, and output standards. Build the first one around something you currently supervise closely — so you can spot when it breaks.

The honest version

Most founders don’t get past Phase 2 because Phase 2 already feels good. Real productivity gains, visible time savings, something worth mentioning at dinner. The incentive to go further is less obvious.

But Phase 2 is still about doing faster. Phase 3 is about doing less. Phase 4 is about running an organisation that does things for you.

The jump isn’t about technical ability. It’s about whether you’re willing to invest design time upfront in exchange for attention freed permanently. That’s a different calculation than productivity.

The question isn’t what phase you’re capable of. It’s what phase you’re willing to design for.