The New Kind of Insomnia: When Agentic AI Turns Technical Debt Into Idea Overload

For nearly eleven years as a web developer at UC Berkeley, I've navigated the familiar terrain of technical constraints. Python, Django, Wagtail — these were the tools, and the question was always: How do I build this?

Agentic AI has fundamentally rewritten that question to: What should I build?

The "how" has become almost trivial. The real challenge now is crafting the prompt — the refined, detailed articulation of what you want the AI to create. It's not about whether you can build something anymore. It's about whether you can describe it clearly enough.

Technical debt used to be the bottleneck. Limited time, limited resources, limited bandwidth to maintain and extend systems. You learned to say no to good ideas because you couldn't execute them all.

Now the bottleneck is completely different: idea overload.

I have more viable project concepts than I have hours in the day. Each one is technically feasible. Each one could be prototyped in an afternoon. The constraint isn't capability — it's clarity and prioritization.

Last week, I was handed a statement of work for a new project at Berkeley. Normally, this would kick off weeks of planning, architecture meetings, and careful scaffolding before a single line of code got written.

Instead, I fed the SOW to my agentic AI assistant.

By the time the kickoff meeting happened, I had a working prototype.

Not a mockup. Not a proof-of-concept diagram. A functioning application that demonstrated the core features, complete with database schema, user authentication, and API endpoints.

The kickoff meeting became a demo instead of a planning session. We skipped straight to refinement and iteration. The team could see it, touch it, interact with it — and suggest improvements based on reality instead of imagination.

That's the power shift: from "let me explain what this might look like" to "here, try it out and tell me what you think."

This shift demands new skills. Not coding skills — those are still useful, but less central. The new core competency is prompt architecture.

Can you articulate:

- The problem clearly enough?

- The user stories precisely enough?

- The edge cases thoroughly enough?

- The architectural principles explicitly enough?

A well-crafted prompt is like a well-written specification, but more powerful — because it doesn't just describe what to build, it actually builds it.

I'm learning to think in layers:

1. High-level vision — What problem are we solving?
2. User experience — How should this feel to use?
3. Technical guardrails — What constraints matter?
4. Integration points — How does this fit with existing systems?
5. Quality standards — What does "done" look like?

The better I get at structuring these prompts, the better the output. It's almost fractal — clarity begets clarity.

So yes, I'm still up at night. But instead of worrying about technical debt, I'm wrestling with:

- Which of these fifteen project ideas deserves my attention first?

- How can I describe this complex workflow clearly enough for AI to nail it?

- What am I not building because I haven't imagined it yet?

- How do I keep up with the pace of what's now possible?

It's a better kind of insomnia. The "good problem to have" kind.

But it's still insomnia.

We've solved the scarcity problem. Technical execution is no longer the limiting factor. But abundance creates its own challenges.

When you can build anything, what should you build?

When prototypes take hours instead of weeks, how do you avoid chasing every shiny idea?

When the tools are this powerful, how do you maintain focus?

These aren't technical questions. They're strategic, philosophical, even psychological questions.

If you're a developer and you're not experimenting with agentic AI yet, you're not just missing a tool — you're missing a fundamental shift in what the job is.

The role is evolving from:

- "Person who writes code" → "Person who architects solutions"

- "Technical implementer" → "Vision translator"

- "Builder constrained by time" → "Curator constrained by attention"

You'll still need to understand code — to review it, to refine it, to know what's possible. But your primary value shifts to knowing what to build and why, not just how to build it.

In this environment, the differentiator isn't coding speed. It's:

1. Domain expertise — Understanding the problem space deeply
2. User empathy — Knowing what people actually need
3. Strategic thinking — Choosing the right battles
4. Communication skills — Translating vision into actionable prompts
5. Iteration discipline — Refining rapidly without losing focus

The people who thrive won't be the ones who can code the fastest. They'll be the ones who can think the clearest.

The strangest part of this transition? I'm my own bottleneck now.

The AI can build faster than I can decide what to build. It can prototype faster than I can evaluate whether the prototype is worth pursuing. It can generate options faster than I can choose between them.

I'm learning to:

- Timebox exploration — Give yourself permission to experiment, but set limits

- Document decisions — Write down why you chose one path over another

- Version your thinking — Your prompt is your spec; treat it like code

- Share early — Get feedback on prototypes before you fall in love with them

For all this automation, one thing hasn't changed: judgment.

AI can build what you describe. It can even suggest improvements. But it can't decide what matters to your users, what aligns with your organization's mission, or what's worth your finite attention.

That's still on you.

And honestly? That's the part that keeps me up at night — in the best possible way.

Technical debt used to be my nightmare. Now it's idea overload.

I'll take this problem over the old one any day. But let's not pretend it's not a problem.

The challenge has shifted from execution to curation. From building to choosing. From how to what.

And if you're not feeling a little overwhelmed by the possibilities right now, you're probably not paying attention.

Welcome to the age of abundance. Try to get some sleep.

James Farris is a web developer at UC Berkeley, where he's been building web applications for almost eleven years. He's currently trying to figure out which of his seventeen project ideas to actually finish. Connect with him on LinkedIn to watch him struggle with idea overload in real time.