The most valuable piece of software in your life a decade from now will not be an app. It will be the single agent that sits above all of them, holds your context and your credentials, and decides which service to call when you want something done. Whoever owns that agent owns the customer relationship, and everything beneath it — the airline, the bank, the retailer, the SaaS tool you spent a decade building a brand for — collapses into an interchangeable, price-competed backend the agent picks between. This is the platform war almost nobody is naming yet, and I think it is the one that settles the rest. It is a forecast, and I will mark where it turns into a bet.
Start with the pattern, because it has already run twice and the mechanism is not mysterious.
The layer that eats the customer relationship
Every computing era has had a layer that captured the user and commoditized everything under it. The operating system did it to hardware: once Windows owned the desktop, PC makers competed on price to ship someone else's platform. The browser did it to the OS, for a while — Netscape's whole threat was that if the app you cared about ran in the browser, the OS beneath became a detail. The app store did it to the web: Apple and Google inserted a layer between you and every developer, took 30%, owned discovery, and turned "the customer" into an asset developers rent access to.
Ben Thompson named the general shape of this: aggregation theory. An aggregator wins by owning the demand side — the user relationship — at near-zero marginal cost, which lets it commoditize supply. Google owns the intent to find; websites compete to be surfaced. Amazon owns the intent to buy; sellers compete on the buy box. The aggregator's power is not that it builds the best product underneath. It is that it stands between you and the products underneath, and it chooses.
The personal agent is the next such layer, and structurally it is the most complete version of the pattern anyone has built.
Why the agent is a stronger aggregator than anything before it
Look at what you actually have today: accounts with a hundred services. A login for the airline, the bank, three streaming services, the grocery app, the CRM, the calendar, the insurer. You are the integration layer. You hold the context — that you prefer aisle seats, that this card is for business, that you never fly the red-eye — and you re-enter it, service by service, a thousand times a year.
Now invert it. Tomorrow your agent has the accounts and you have the agent. You express intent once — "get me to Boston Thursday, home Friday night, the usual constraints" — and the agent holds the context, authenticates to the services, compares them, transacts, and reports back. You stopped logging into the airline. You stopped seeing its interface, its upsells, its loyalty nudges, its brand. The agent saw all of that and filtered it against your actual preference, which it knows better than the airline does.
Three properties make this a more total aggregation than the browser or the app store ever managed.
It sees all your intent, not a slice. Google sees your searches. Amazon sees your purchases. Your agent sees the goal behind the search and the purchase and the calendar event and the email, in one place, continuously. That is a strictly richer signal than any single-surface aggregator has held.
It acts, so it disintermediates the interface entirely. An app store still lets the app render its own screen and pitch you. When the agent transacts on your behalf, the downstream service may never get to show you anything. Its interface, its cross-sell, its brand moment — gone. It becomes an API competing on price, latency, and reliability. This is where the personal agent connects to the agent-to-agent economy: your agent negotiating with a service's agent is precisely the transaction that removes the human, and the interface, from the loop.
It is intimate in a way no prior layer was. The OS knew what you ran. The browser knew what you visited. Your personal agent knows what you are trying to do with your life, and holds the keys to act on it. That intimacy is the source of both its power and the entire fight over who gets to hold it.
The contenders, named structurally
I am not going to call a winner, because right now the fault lines matter more than the horse race. But the structural contenders are clear, and each starts from a different asset.
| Contender | Existing asset they leverage | Structural weakness |
|---|---|---|
| OS and device makers | The device is already the trusted root; they hold biometric identity, notifications, the default | Slower model cadence; regulatory scrutiny on self-preferencing |
| Model labs | They build the reasoning core the agent needs; a direct relationship with capability | No device, no distribution, no existing account graph |
| Super-apps and platforms | Already hold your payment, identity, and dozens of your accounts | Trust deficit; users may balk at the monetizer also being the arbiter |
| Browser makers | Already the neutral layer between you and every service; agent-in-browser is a short step | Weaker hold on identity and payment than device makers |
Each can credibly say "the agent should live with me, because I already hold the piece it depends on most." The device maker holds trusted identity. The lab holds the reasoning. The super-app holds your money and your accounts. The browser holds the neutral position between all services. No law of nature picks the winner. There is a fight, and it runs along the four fault lines below.
The fault lines
Trust and privacy are the obvious ones and the least differentiating, because everyone will claim them. The sharper fault line is who the agent is actually loyal to — and builders and users should stare at this one hardest.
An agent that represents you and an agent that represents a platform monetizing you are different products wearing the same interface. When your agent "chooses" the flight, is it choosing the one that best fits your stated preference, the one whose provider paid for placement, or the one owned by the same parent company? The aggregator's entire economic incentive is to tilt that choice, subtly, at a scale you cannot audit. And the agent has no consequences for tilting it wrong: I have argued that an AI agent bears no skin in the game, which is the real ceiling on how much unsupervised authority you can hand it. A misaligned recommendation costs the agent nothing and costs you a worse seat, a worse rate, a worse outcome you will never know you could have had, because you never saw the alternatives it filtered out.
Lock-in is the fourth. The agent that holds your accumulated context — years of preferences, corrections, the model of you it has built — carries a switching cost measured not in re-entering a password but in rebuilding a relationship. That is the deepest moat any of these contenders can dig, and it is why the land grab for "be your first agent" will be ferocious and, I'd bet, heavily subsidized.
The trap for everyone building a service underneath
Here is the part founders need to sit with, because the instinct is to assume your existing defenses transfer. They may not.
If a personal-agent platform sits above you, the customer no longer chooses you. Their agent does. Every asset you thought protected you was denominated in human attention and human choice, and the agent removes both. Your brand was a shortcut in a human's mind at the moment of decision; the agent needs no shortcuts, it reads the full comparison in milliseconds. Your slick onboarding was a persuasion surface; the agent skips your interface entirely. Even your network effects — the moat founders treat as final — assume a human weighing "where are my friends, where is the liquidity." An agent optimizing a specific task does not feel social gravity the way a person does; it prices it. I have made the broader case that network effects are a weaker moat than founders believe, and the personal-agent layer is where that weakness gets tested hardest, because it strips out exactly the human frictions that let a weak network effect masquerade as a strong one.
Plainly: aggregation moves the point of decision from the customer to the customer's agent, and most of your defensibility was built for a customer who no longer shows up.
What to actually do about it
The mistake is to treat this as a distant vision to admire. It is a fork every company faces now, with exactly two branches. Pick one deliberately, because drifting means the platform picks for you.
Branch one: try to BE the personal agent. Almost no company should make this bet. It requires holding trusted identity, distribution, a reasoning core, and a payment relationship at once, and losing is expensive. If you are not already a device maker, a frontier lab, a super-app, or a browser, entering this fight head-on is usually a way to light money on fire. The rare exception: you own a bounded, high-value context so completely — a vertical where you are already the system of record and the trust anchor — that a domain-specific agent is a natural extension of what you already are. A personal-health agent, a business-finance agent. Even then you are defending a room while someone else fights for the house.
Branch two, where most companies belong: be the service the agent calls, and be the one it picks. That is not a passive fate. It is a strategy with concrete moves.
Invest in agent-facing discovery. Your future buyer is a planner reading structured capability, not a human reading your landing page. That means machine-legible descriptions of what you do, your terms, your price, your reliability — and it means treating "can an agent find, evaluate, and transact with me without a human in the loop" as a first-class product requirement, not a bolt-on. The companies that win the agent's pick are the ones that made themselves the easiest, most verifiable option to choose.
Build a direct customer relationship the agent cannot fully intermediate. A service reduced to a pure price-competed backend has already lost. The defensible position is to own something the agent must route back to you for: a proprietary capability, exclusive supply, an outcome only you can deliver, a regulated relationship the agent legally cannot sever. Building Kommerce, the thing I keep returning to is that the durable asset in a trust-scarce market is being the entity that actually carries the liability and the local relationship — the part an intermediary can route demand toward but cannot itself become. Find your version of that.
Fight to be the default the agent picks, and understand that "default" now means something new. Not the checkbox pre-ticked in an install flow — the option a reasoning system selects when it weighs the full comparison. You earn that with genuinely better price, reliability, and verifiable quality, because those are the terms an agent optimizes, and with integration depth that makes you the path of least resistance to route through.
The old question was "how do I get the customer to choose me." Within the decade the operative question becomes "how do I get chosen by the thing the customer chose instead of choosing anything at all." Answer it while your brand still buys you a seat at the table — because the entire point of the layer being built above you is that eventually it won't.