The Pure Utility

The first piece in this series argued that a billion people will pay $100 a month for ambient intelligence by 2030. The second argued that the $100 replaces the subscription stack rather than adding to it, and that every person becomes their own only customer. This one is about who they pay. My answer is that most people will have a direct, one-to-one relationship with a frontier company, and that the crowd now assembling to stand in between will mostly be gone by 2030.

I want to be fair to the crowd, because it is large, well funded, and not stupid. So let me state their case before I take it apart.

Everyone who wants to sit between you and the frontier has a reason you should let them. The reasons are real, in the sense that each one is true today about some product. Here they are, as their sellers would put them.

• Cheaper tokensWe buy in bulk, we route to the cheapest model that can do the job, and we pass the savings to you.
• Model independenceBuild a skill once and run it on any model. Never be locked in. Switch when a better one ships.
• OptimizationWe cache, compress, batch, and route. Faster answers, fewer tokens, better uptime than going direct.
• Spread the loadSplit your work across providers so no one outage or one rate limit stops you.
• Governance and complianceWe add the audit log, the data-residency guarantee, the approved-vendor status your company requires.
• The familiar placeYou already use our software. Get the AI inside the thing you already open every day.

Investors like these pitches. OpenRouter’s reported $1.3 billion valuation is one data point. The whole enterprise “AI platform” category is another. The largest example of all is Microsoft, which has spent years and a reported thirteen billion dollars becoming the biggest reseller of OpenAI’s intelligence on earth. If the middle can be defended, Microsoft is the company that can defend it. Hold that thought.

Three things kill the intermediary, and they get stronger every year, not weaker.

The lag

New models reach the resellers late. Third-party trackers put the gap between Anthropic’s direct release and the same model on Amazon Bedrock at one to four weeks, and models are the easy part, because a model is a file that can be copied. The capabilities that actually matter to the billion people in the first piece are not files. Scheduled tasks, memory that carries from one day to the next, skills, plugins, the ability to act inside your own machine: these live in the frontier company’s own app. Anthropic’s scheduled tasks run in the Claude desktop app and are not available through the API at all. A router that reaches 400 models cannot reach that, because it is not a model. It is a relationship, and relationships do not resell.

So the lag is not weeks. For the things that make intelligence ambient, the lag is infinite. The intermediary can only ever sell you the part of the product that was designed to be sold through intermediaries, and that part shrinks every quarter.

The terms

Every layer between you and the source is a party with leverage. It has its own price, its own terms of service, its own outages, its own support queue, and its own board deciding what you may do. The second piece described the rule I run my own life by: the test of a dependency is not whether it is a third party, it is whether it holds leverage over you. A reseller holds all of it. When the frontier company changes its price, the reseller decides whether to pass that on. When the frontier company ships a capability, the reseller decides whether you get it. You did not buy intelligence. You bought a subscription to someone else’s decisions about intelligence.

The feeling

This is the one the analysts miss, and it is the strongest of the three. When you go direct, the AI is yours. Claude, on my machine, on my account, on my bill, is on my team and nobody else’s. I know exactly who I am dealing with and I know they have no one else to answer to in the transaction but me. When an intermediary is involved that feeling never arrives. There is always a sense that the intelligence is on loan, that it is really someone else’s, configured for someone else’s reasons, and that I am a tenant in it. Nobody feels that way about their phone, and nobody will tolerate feeling that way about their AI once they have felt the alternative.

Nobody wants a value-added reseller of the same intelligence with terms added. They want the source, and they want it to feel like theirs.

Back to the company best positioned to defend the middle. Microsoft resells OpenAI’s intelligence as Copilot, inside software that a billion people already open every day, with the deepest partnership in the industry behind it. If “the familiar place” and “governance” and “we already have your data” were enough, Copilot would have won by now.

Instead, in August 2026, Microsoft merged its consumer and business Copilot apps into one and retired a list of features for consumers: Deep Research, group chats, AI-generated podcasts, the Labs area for experiments, the animated assistant. The reporting on the merger was blunt about why: keeping two separate products made it hard to hold up against ChatGPT, Claude, and Gemini, each of which offers one assistant, direct. The reseller with every structural advantage on earth is consolidating and cutting, while the sources are shipping. That is the whole future of the intermediary layer, run early, at scale, by the strongest possible candidate.

We have watched this before. In the 1990s AOL, Prodigy, and CompuServe stood between people and the internet with curated content, keywords, and a friendlier interface, and for a while that was worth paying for. Then people discovered they could go directly to the thing, and the friendlier interface became a cage. The lesson was not that curation is worthless. It was that nobody stays inside a walled garden once the open field is one click away and everything new appears in the field first.

None of this means nobody makes a living near the frontier companies. It means the living is made beside the person, not between the person and the source. Two kinds of work survive.

The first is the records layer the second piece described: the company as a database, a server its people’s AIs can call, and a governance layer over who may touch what. The clouds have a real role there, hosting the ledger and the permissions. But notice that in that design the cloud is not between you and your AI. Your AI is yours, direct; the cloud is where the company’s records live. That is a fine business. It is not an intermediary.

The second is the person who sets you up. Connectors, skills, plugins, scheduled tasks, one household or one owner at a time. That person does not resell intelligence and does not sit in the path of it. They put your direct relationship in place and then get out of the way, the way the person who wired your house does not stand between you and the power company. That is what we do at Charleston AI, and the reason I am comfortable saying the middle evaporates is that we are not in it.

Claim: by 2030 most people pay a frontier company directly, and most intermediaries are gone

Already true

The capabilities that make intelligence ambient ship first, and sometimes only, in the frontier companies’ own apps. Model availability through resellers lags by weeks. The strongest reseller in the world is consolidating and cutting consumer features while the sources ship. People already describe going direct as “mine” and going through a layer as “theirs.”

What has to happen

The frontier companies have to keep shipping capabilities faster than the middle can copy them, which is the trend, and they have to keep selling direct to individuals at a price an individual will pay, which the $20, $100, and $200 tiers already do. And the announcement-day reflex has to stay strong: people have to keep wanting the new thing the day it exists. Every phone launch for fifteen years says they will.

Where I am probably wrong

Regulation could force a middle into existence, mandating audit layers or national gateways that everyone has to route through. A frontier company could become the intermediary itself by buying the reseller, which is a different outcome from the reseller surviving. And the companies people work for could win the fight over whose AI you use at work, so that the direct relationship exists at home and a governed layer exists at the office. That last one is the version I would bet against myself on: it is exactly the split people rejected with their phones, but companies have more reasons to fight for it here.

The bill goes to the source. The feeling of it being yours only exists at the source. And everything new appears at the source first. That is three reasons, and one is enough.