Stop Asking AI to Do Your Old Job Faster

I came up through tools that reward throughput. Claude Code in the terminal. Cursor in the editor. Agents that refactor, open PRs, and clear the backlog. That is real leverage. It is also, increasingly, table stakes.

When a frontier model lands and the coding Twitter timeline splits into "this is genius" and "this is unusable," that is not only noise. It is a signal. The model may be uneven on the tasks you already automated. Meanwhile it quietly unlocks surfaces you never put on the roadmap: interactive demos, proposal sandboxes, short documentaries, practice sims, 3D walkthroughs.

If you only score the release on "did my existing coding loop get 12% faster," you will miss the plot.

Efficiency AI vs opportunity AI

Efficiency AI helps you do the job you already have: faster, cheaper, with fewer handoffs. Draft the email. Summarise the meeting. Scaffold the CRUD. Close the ticket.

Opportunity AI changes the menu. It makes viable the project that used to need a specialist, a budget, or three vendors.

Efficiency compounds into expectation. Customers and teammates will assume you can move at agent speed. Opportunity is where category shifts happen, and where a solo builder or a small team can punch above headcount.

Neither wins by killing the other. The trap is treating efficiency as the whole strategy.

The blank page problem

"Go find new opportunities with AI" is useless advice. Humans do not walk around with a complete inventory of inventions. We carry a short list shaped by our role, our tools, and what our peers celebrate.

So opportunity work needs thought starters: borrowed moves from other crafts, ported into your world.

A useful shortcut: watch what another profession does that feels like magic, then ask what the equivalent would be in your product, practice, or team.

Twelve ports for builders (steal freely)

These are not product requirements. They are prompts for experiments.

1. Playable marketing — A five-minute game that makes your thesis felt, not claimed. Example: coordination cost as a delivery game for operators.
2. Video as a default output — Not "hire a studio." Script, then phone A-roll, then an agent-built pipeline for captions, motif, cuts.
3. Explorable demos — Let the buyer navigate by their question (security, setup, cost) instead of your slide order.
4. Shapeable proposals — Scope, time, and cost as toggles the client can stress-test before the reply-all begins.
5. Decision simulators — Force shared assumptions. Make "what if demand is half" a button, not an argument.
6. 3D where spatial beats slides — Walkthroughs, layouts, physical product intuition.
7. Customer stories as short films — Interview plus screenshots plus before/after, a 60-90 second documentary, not a case-study PDF.
8. Judgment gyms — Practice hard conversations and calls with feedback mid-stream.
9. Prototype the physical — Form factors, kits, teaching objects. Treat conditions as designable.
10. Browser-native utilities — Small features that used to need a full product team.
11. Synthetic test crews — Agents that attack your UX, onboarding, and edge cases before users do.
12. Expertise as a product — Package the repeatable part of your judgment so others can work through it without booking you.

Notice the pattern: many of these are what-if machines. They externalise trade-offs that used to live only in your head.

How I am applying this as a builder

I build in public-ish loops: training software (RivaFlow), a photography site (The Room), and a personal AI harness that has to survive real life. The efficiency stack is already there: agents, vault, calendars, drafts-not-sends.

The opportunity question is sharper:

• What would I ship if video production were "record on a phone and drop a folder"?
• What would sales or onboarding look like if the demo were explorable?
• Which piece of my own judgment should become a small product others can run?

I am not abandoning Cursor agents for homework. I am refusing to grade every model release only on whether it refactors TypeScript more politely.

A practical rule for the next 30 days

Each week, run one efficiency loop and one opportunity loop.

• Efficiency: something you already do. Make it 2× cheaper or faster with an agent.
• Opportunity: something you would not have attempted last year. Ship an ugly v0 in under four hours.

If the opportunity experiment fails, you learned the shape of the blank page. If it works, you just changed your product surface.

Closing

Broad model capability is high enough that "better at the old thing" is no longer the only story. Sometimes the model will feel like a regression on your favourite workflow and a breakthrough on a workflow you never owned.

Builders who only chase efficiency will be very fast at a job the market has already repriced. Builders who also chase opportunity will look, from the outside, like they have a bigger team than they do.

That is the point.

---

Inspired by a long-form AI Daily Brief on opportunity AI. Reworked here for builders who live in agents, repos, and shipped product, not slideware.