The Underwater Internet

Picture the web as an ocean.

On the surface: browsers, feeds, dashboards, the familiar theater of clicking and scrolling. Humans still swim there. We will keep swimming there for a long time — for taste, for art, for conversation, for the pleasure of looking.

Underwater is where the volume moves.

Most internet use will not be you opening tabs. It will be almost single-use software — personal systems with a job — reading APIs, syncing state, filtering noise, writing journals, asking for your attention only when a decision deserves it. The human stays at the surface. The work happens below.

Surface and depth

Today we still design as if the primary client of the internet is a person with eyes.

That was true when the cost of integration was high. You visited the bank. You visited the broker. You visited the newsletter. Every service wanted your attention because attention was how software got used.

Personal systems invert the client.

Your investment observer does not “visit” the market the way you do. It syncs. It filters. It admits a few candidates. It may never paint a candle chart you will stare at. Your creative OS does not browse every tool’s homepage; it calls the adapters it owns. Your intake agent does not scroll a knowledge base for fun; it retrieves what the next question needs.

The surface becomes exception handling and desire. The depth becomes routine cognition.

This is not a claim that humans disappear from the web. It is a claim that human browsing stops being the median packet. Most requests will be machine-to-machine, policy-bound, memory-backed — issued by systems that exist for one person and one job.

Underwater is not invisible chaos

“Underwater” sounds opaque. In a good design it is the opposite of wild scraping.

Depth needs:

Clear contracts — APIs, permissions, rate limits, audit trails
Owned memory — what was fetched, what was decided, what was refused
Human gates — especially where stakes are real
Quiet as a valid outcome — most underwater cycles should end with nothing to surface

If the deep layer is just unsupervised agents clicking around, you have not built a future. You have built liability with latency. The pattern only holds if personal systems are aligned before they are autonomous — and if chat is not the system of record for what they did on your behalf.

AI is already standardizing the internet

Something quieter is already underway. AI is standardizing how the internet is produced and consumed.

Not by decree. By pattern.

Models prefer structure: headings, schemas, consistent navigation, predictable forms, machine-readable metadata. Publishers optimize for retrieval and citation. Tools emit the same shapes because the same models sit in the middle. Agents learn “how a settings page works,” “how an OAuth dance looks,” “how a docs site is organized” — and those conventions reinforce themselves.

Patterns form. Then they compound.

A layout that agents parse reliably gets copied.
An API that agents call safely becomes the default integration path.
A content shape that summarizes cleanly becomes the house style of half the web.
A failure mode that burns tokens gets designed out of the next generation of UIs.

The internet starts to look less like a wilderness of idiosyncratic sites and more like a shared grammar that humans and machines can both inhabit — with machines increasingly fluent in the deep currents.

Standardization here is not boring bureaucracy. It is the ocean becoming navigable at depth. Without shared patterns, personal systems cannot swim. With them, half-day systems can plug into the world without each builder inventing the ocean from scratch.

When specialization becomes cheap, you do not need one mega-portal. You need many small systems — and a web regular enough that those systems can operate underwater without drowning in exception handling.

What changes if this is right

For people: less living inside other people’s dashboards. More living with a small shelf of systems that know your job. You surface when something matters — or when you want to wander for its own sake.

For builders: the primary user of your service may be an agent with a policy file, not a person with a mouse. Design for contracts, idempotency, and explainable state — not only for screenshots in a pitch deck. Code is not only for humans anymore; neither is the web.

For platforms: the prize may shift from “own the session” to “be a reliable depth layer” — data, identity, execution rails — while personal systems hold judgment and memory closer to the human. A thin shared environment; intelligence at the edges.

For culture: the open web does not die. It becomes a place humans still go for meaning — while logistics, monitoring, and routine decisions migrate below the waterline.

A shelf of systems, an ocean of patterns

The companion claim to almost single-use software is this:

Personal systems are how you specialize.
Standardization is how the internet becomes swimmable for them.

Together they sketch a future where most use of the network is quiet, continuous, and owned — underwater — and where the surface is reserved for attention that is actually worth spending.

That future is not evenly distributed yet. Parts of it are already here: sync jobs, agent browsers, structured docs, APIs that assume a machine is listening. Taking it seriously means building as if the median client of the web will soon be a system that works for one person, on one job, and only knocks when the deep current changes.

The insight: the internet’s volume will move underwater — used by personal systems — while AI standardizes the surface into patterns that compound until depth becomes the default way the web is used.