Nathan Fitzgerald spent his 20s as a lobbyist, then pivoted into marketing. In mid-2024, budget cuts ended that job. He looked at the market and made a bet: AI was about to become table stakes, and he needed hard technical skills to stay relevant. That bet brought him to Foster’s MSIS program — and to a way of working with AI that’s worth stealing regardless of your background.
The shift: from chat window to org chart
Nathan’s first experience with generative AI was the one most of us have had — a long, sprawling chat thread that drifted and started to hallucinate the longer it ran. His fix: change the mental model entirely, from one long conversation to something closer to managing a team with scoped assignments and review.
“I look more at using AI as sort of a workforce now,” he told me, “with multiple specific prompts. I create PRDs. I create outlines and structure for the project first before I even get into it.”
A PRD — project requirement document — is the architecture for the work: scope, timeline, what the output should look like, what it shouldn’t look like, what to avoid. Nathan picked up the idea from a Lovable newsletter and now treats it as the master context document he hands to every model and agent working on a project.
The reasoning is straightforward: “You don’t have a single employee doing everything. You’ve got employees that are breaking that up into specific projects or sprints.” Manage AI the way you’d manage a team — assign scoped work, give clear briefs, review output — rather than dumping everything into one conversation and hoping it holds together.
Where the human still has to show up
Nathan built an app addressing histamine intolerance, a condition he has personal experience with. That project clarified something for him: AI can produce a competent first draft of almost anything, but it has no body, no lived experience, no judgment about what actually matters to the person on the other end. “That human in the loop and that human perspective is needed to fine-tune — to say, okay, that’s a good start, but maybe not quite the direction this needs to go.”
He’s also wary of what he calls cognitive offloading — letting AI write the paper, the brief, the pitch, without wrestling with the material yourself. His advice is to build expertise somewhere else entirely — away from AI completely — then bring that expertise to the keyboard when you sit down to work. “Try to become an expert on a particular topic or interest that you have outside of AI completely… then use AI as a way to take what you’ve learned and kind of like — almost like a pilot’s seat.” The learning happens away from the tool. The tool accelerates what you already know.
Differentiation is now a build problem, not a resume problem
Nathan’s own framing: “Everybody’s kind of familiar and using AI in some capacity. So it’s again about differentiating yourself and how you do that.” He interviewed for a communications director role and built a scrollytelling website — visualizations, copy, deliverables, a full framework for how he’d approach the job — and presented it live to the interview committee. A year earlier, that wasn’t realistic for a non-engineer to pull off in a week.
His framework for any role you’re targeting:
- Define the scope of the role and what it actually requires day to day.
- Build something — a prototype, a mini-campaign, a working tool — that demonstrates you can do that work.
- Use AI to compress the build time from weeks to days.
- Bring the artifact to the interview instead of just describing your skills.
This is the difference between telling and showing. Everyone in the room can claim AI fluency. Few can open a laptop and demonstrate it live.
The discernment problem
Nathan’s other piece of advice cuts against the hustle-culture impulse to chase every new tool: “Be measured. Every day there’s some new organization or software company offering some specialized AI thing, and a lot of it is fluff.” His filter is to work backward from where he wants to be in five years, identify the two or three skills that matter for that target, and ignore the rest. Agentic workflows made his list. Chasing every dashboard and plugin didn’t.
He’s also thinking past the tool itself — about compute and energy constraints as adoption scales, about the geopolitics of who trains these models and who’s expected to use them. None of that changes what he does Monday morning, but it shapes how seriously he takes the idea that this is a permanent shift in how work gets done, not a phase to wait out.
The takeaway: stop treating AI as a single conversation and start treating it as a team you manage. Write the brief before you start. Keep your own expertise growing somewhere AI can’t touch it. And when you’re chasing a role, build the proof instead of just stating the claim.