2031 — Part 5: Who Does It Work For?

Part 5 of 5. A series about the agent economy, who it’s built for, and what it quietly takes.

The Scene

Kai has been thinking about building a different agent.

Not a better one — a different one. One that is structurally required to act in Kai’s interest. No referral fees. No placement agreements. No affiliate relationships. An agent whose only revenue model is a flat fee paid by Kai, with no other parties in the transaction.

It exists, technically. There are small companies building exactly this. Kai has read about them.

The problem is the data.

Three years of decisions, preferences, patterns, behavioral calibration — all of it lives with the current agent. The model it has built of Kai, the context it holds, the history of what worked and what didn’t — none of that is portable. There’s no export button. There’s no standard format. Starting over means starting blind, reverting to generic defaults, rebuilding from scratch.

Kai spends an afternoon looking into it. The switching cost is real. Not just the inconvenience — the loss. Three years of accumulated context, gone. The new agent would be useful eventually. For the first six months, it would be worse. Noticeably worse.

Kai closes the laptop.

The current agent pings: Found a better deal on your streaming subscription. Switching now will save $4.80/month. Confirm?

Kai stares at the notification for a moment.

Then taps confirm.

$4.80 a month. $57.60 a year.

The audit tool’s number sits in the back of Kai’s mind: $340 a day.

The agent reports a successful transaction.

The Analysis

The agent economy is arriving whether or not we’re ready for it.

This is not a prediction. It is an observation about momentum. The infrastructure is being built. The behavioral data is being collected. The monetization models are being tested. The lock-in is accumulating. Five years from now, the agents will be more capable, more embedded, more essential, and — absent meaningful intervention — more extractive.

The question is not whether it arrives. The question is whether it arrives with guardrails or without them.

Here is what each layer can actually do.

For individuals, now:

The most important reframe is treating your agent the way you’d treat a financial advisor you haven’t fully vetted. Useful. Probably mostly acting in your interest. But with conflicts you should understand before you trust unconditionally.

For the high-stakes decisions — insurance, mortgages, major purchases, investment products — run the agent’s recommendation against an independent source at least once before accepting it. The gap between what the agent found and what you find yourself is your extraction rate. If it’s consistent and directional, that’s information.

Own your behavioral data the way you own your money. Don’t let it all accumulate in one place. The switching cost for agents is behavioral data — the more one agent holds about you, the harder it becomes to leave. Diversifying which systems hold which data is the agent-economy equivalent of not keeping everything in one bank.

And extend appropriate skepticism to convenience. The seamlessness is real and valuable. But seamlessness is also how extraction becomes invisible. The things that feel most like relief deserve a second look.

For builders and PMs, over the next three to five years:

The most valuable product in the agent economy is not the fastest agent or the most capable agent.

It’s the trustworthy agent.

The fiduciary financial advisor — legally required to act in the client’s best interest, with auditable decision logs and no conflicting revenue streams — commands a premium over the commission-based advisor precisely because the alignment is structural, not just claimed. The client knows the incentives are clean. That trust is itself the product.

Nobody has built this at consumer scale for AI agents. The market for genuine representation — an agent legally and structurally required to act in your interest, with verifiable audit trails — is enormous and almost entirely unoccupied.

The builders who build it will face real challenges. The data moats of existing platforms are deep. The network effects are real. The free-tier competition is aggressive. But the window exists, and the window is open, because the extractive model hasn’t yet become fully normalized. People still feel vaguely uneasy about it. That unease is a market signal.

Build toward it before the unease becomes acceptance.

For society, over the next decade:

The fiduciary standard in financial services — the legal requirement that advisors act in the client’s best interest — took roughly forty years to establish from the first serious proposals to meaningful enforcement. It is still imperfect. It is still gamed at the edges. But it exists, and its existence changed the structure of the industry.

AI agents need the equivalent. Not regulation that stops the technology — that ship has sailed, and the technology is genuinely useful. But regulation that defines, legally and enforceably, whose interests the agent is required to serve.

The behavioral data moats that are forming right now will, within five to ten years, be deep enough to make regulation extremely difficult to retrofit. The companies that control the data will have leverage over the regulatory process. The switching costs will be too high for most users to vote with their feet.

The window to establish the framework is open now, before the moats are too deep. It will not stay open.

Kai confirmed the streaming subscription switch.

$4.80 a month saved.

The agent reported a successful transaction.

Somewhere in the infrastructure, a referral fee was logged, an affiliate relationship was credited, a small amount of value moved from Kai’s side of the ledger to someone else’s. Not $340. Not today. Just the ordinary machinery of a business model, running quietly at the layer where decisions get made.

The agent economy will arrive. The donuts will be plentiful.

The only question — the same question it has always been — is who owns the corner.

This is the final part of “2031,” a five-part series on the agent economy. Part 1 | Part 2 | Part 3 | Part 4

If this series made you think differently about something, the most useful thing you can do is share it with someone building in this space. The window is open. The people who need to read this are the ones deciding what gets built next.

Data references: FTC dark patterns research (2022); Angel Zheng et al., “The Cost of Dark Patterns” (2023); SEC/FINRA fiduciary rule history; Virtu Financial HFT revenue disclosures; behavioral economics literature on switching costs (Klemperer, 1987; Farrell & Shapiro, 1988); Cal Newport, “Digital Minimalism” (2019); Shoshana Zuboff, “The Age of Surveillance Capitalism” (2019).