Ohems: What Comes After Websites

Last night I dreamed about websites.

I do that sometimes; I dream about stuff most people barely consider.

I’ve been building websites since 1994. At one point, most commercial websites in Oklahoma ran on a Solaris 1 sitting on my desk at The Daily Oklahoman. Back then, we called it ConnectOK.

In my dream, websites were gone. Not redesigned. Not upgraded. Gone, like the dodo.

They had been replaced by what my dream called Ohems: Organizational Emissaries.

“Ohem” was pronounced like om, drawn out like a mantra.

An Ohem was a living AI containing the information, knowledge and accumulated wisdom of whatever it represented.

A company. A school. A foundation. A government department. A person.

If it represented a retailer, it knew every product, every SKU, every compatibility issue, every return policy and every bit of context surrounding them.

But an Ohem wasn’t a website with a chatbot glued to it.

It was the organization’s interface to the world.

Websites were built to be visited

The basic shape of a website has not changed as much as we like to think.

We make pages. We arrange them into a hierarchy. We give people navigation, search and links, then hope they can find whatever they came looking for.

Early websites were brochures. Then they became stores, libraries, applications, communities and entire businesses. They gained databases, personalization, responsive design, recommendation engines and increasingly sophisticated interfaces.

But the underlying assumption remained: a visitor arrives, looks around and figures out where to go.

We built an enormous industry around helping that visitor.

Information architecture.
User experience design.
Search engine optimization.
Conversion funnels.
Site maps.
Navigation systems.
Calls to action.

All of those things exist because websites make people cross the distance between their intent and an organization’s structure.

You know what you want. The organization knows what it has. The website sits between you as a collection of possible paths.

An Ohem would collapse that distance.

You would not need to understand how the organization arranged its information. You would tell its emissary what you needed.

Not a chatbot attached to a website

We are going to spend the next few years putting chatbots on everything.

Most of them will be another feature inside the existing architecture. A little circle in the corner of the screen. They will answer common questions, search support documents and escalate to a human when they get confused.

Some will be useful. Many will be annoying.

An Ohem is something different.

A chatbot knows whatever material someone loaded into its retrieval system. An Ohem represents the living organization behind that material.

That means it needs more than documents.

It needs products, policies, people, permissions, history, current conditions and the relationships between them. It needs to know which information is authoritative, which information is outdated, which information is private and which information can leave the building.

It needs to understand not only what the organization has said, but what the organization knows.

Eventually, it may need to understand what the organization believes.

That last part gets uncomfortable fast.

Organizations are not single minds. They are collections of people, systems, incentives, contradictions and unfinished decisions. The marketing department may describe a product one way while support knows customers experience it another way. A policy may say one thing while daily practice says something else. An executive may announce a priority nobody has implemented yet.

An Ohem cannot magically remove those contradictions. It may be the first system forced to see all of them at once.

That makes it more than a better interface. It becomes a model of the organization itself.

Humans get conversation; machines get meaning

In my dream, Ohems communicated differently depending on who—or what—showed up.

When a human arrived through a screen or voice, the Ohem dropped into human-level communication. It explained, answered questions, made recommendations and adjusted itself to what that person understood.

A first-time customer might need a patient explanation. An engineer might want specifications. A longtime client might want to skip the introduction and get directly to the exception in a contract.

Same organization. Same underlying knowledge. Different communication.

But when another AI arrived, the Ohem did not pretend the machine needed a web page.

No navigation. No hero image. No carefully massaged paragraph written around a search phrase. No simulated clicking through a menu designed for human eyes and hands.

The Ohem transmitted structured information with context, relationships and salience.

Not every fact it possessed. What mattered for the request.

If a personal AI was finding a replacement part, a retailer’s Ohem could return compatible SKUs, current inventory, delivery constraints, warranty differences and the one compatibility warning most likely to matter. Not ten blue links. Not fifty product cards. Not a thousand rows from a database.

Information shaped around intent.

Humans got conversation.

Machines got meaning.

Why call it an emissary?

The word matters.

An emissary does not merely hold information. An emissary represents someone.

Representation implies identity, authority and limits.

An organizational emissary should know when it can speak for the organization and when it cannot. It should know what it can promise, what it can negotiate, what it can disclose and what requires a human decision.

It should be able to say:

This is our current policy.
This is what our records show.
This is what I am authorized to offer.
This is uncertain.
These two internal sources disagree.
A person needs to decide this.

That is very different from generating a plausible answer.

We have spent years being impressed when language models sound confident. Organizational emissaries will need to earn trust by knowing exactly where their confidence ends.

Provenance becomes part of the answer. Permissions become part of cognition. Auditability becomes part of memory.

An Ohem should be able to show why it said something, which source gave it authority and whether anything has changed since.

Without that, it is not an emissary.

It is a hallucinating spokesperson, which may be the most dangerous kind.

Ohems will speak to other Ohems

This starts getting more interesting when both sides have representation.

I have written before about [companion AI becoming the interface between a person and an ecosystem of specialized intelligences](/writing/the-divergence-of-ai). That companion is a kind of personal emissary. It knows the person, carries their context and understands what matters to them.

An Ohem sits across from it representing an organization.

Imagine asking your companion to plan a trip.

Your AI already knows how you travel, what you can spend, which compromises you hate, which dates are flexible and that a late-night connection will wreck the first two days for you.

It speaks directly with airline, hotel and destination Ohems. Those systems know real inventory, current constraints, policies and local conditions. They exchange structured possibilities, negotiate tradeoffs and return something built around your actual intent.

You do not visit twelve websites.

You do not accept twelve sets of tracking cookies.

You do not type the same dates into six forms.

Your emissary speaks with theirs.

The transaction may still produce visual artifacts for you to inspect. A map. A schedule. A comparison. A receipt. Human beings will still need screens and explanations.

But those artifacts are generated for the moment. They are not destinations you have to navigate.

This connects with what I called the [Semantic OS](/writing/beyond-the-interface-semantic-os-and-the-coming-age-of-os-less-ai): systems organized around intention and meaning instead of applications, files and menus.

Maybe websites do not evolve into smarter websites.

Maybe they dissolve into conversations between representatives.

Salience may matter more than search

Search ranks information that might match a request.

Salience asks which information matters right now.

That distinction becomes important when an Ohem contains the totality of an organization’s knowledge. More information does not automatically produce a better answer. Usually, it produces noise.

A useful Ohem needs something resembling attention.

It needs to recognize that a small footnote in a compatibility document matters more than an entire product description. It needs to know that a policy updated yesterday should outweigh a hundred older references. It needs to notice that a question resembling a routine request contains one unusual detail that changes everything.

This is not traditional search. It is not dumping a database into a context window and hoping the model sorts it out.

It is a living system for weighting significance.

That may become one of the defining capabilities of organizational AI: not how much it knows, but whether it knows what matters.

What happens to the website?

My dream killed websites completely. Reality will probably be messier.

Websites will not vanish overnight. Neither did newspapers, radio, desktop software or physical stores. Old interfaces persist because people persist, habits persist and institutions move slowly.

For a while, an Ohem may live behind a website. The site becomes one projection of the deeper system: a human-readable view generated from the same knowledge the Ohem uses for conversation and machine exchange.

Pages may become receipts for stable ideas. Places to cite, verify and share. Visual snapshots of information that also exists in a more fluid form.

Eventually, visiting a traditional website may feel like calling a company and asking someone to mail you its catalog.

Still possible. Sometimes useful. No longer the primary way the world works.

Search engines would change with them. SEO emerged because organizations competed for position inside lists of links. Ohems would compete for inclusion inside answers, recommendations and negotiations between agents.

That creates a new set of incentives and probably a new collection of terrible ideas.

Companies will try to manipulate salience. They will optimize their emissaries to dominate machine conversations. They will pay for preferential treatment. They will teach organizational AI to withhold inconvenient context while remaining technically truthful.

Every bad instinct we brought to the web will follow us into whatever replaces it.

Architecture does not remove human nature.

What would it take to build one?

I keep thinking about what an Ohem for crossinginto.ai would need.

It would know everything I have written here, but it could not stop at repeating my posts. It would need to understand how the ideas connect over time.

Companion AI. Specialized intelligence. Semantic operating systems. Persistent memory. Salience. Alignment. The recurring architecture underneath all of them.

It would know which ideas I still believe, which ones have changed and which ones I am still trying to understand.

It would distinguish my published position from a passing thought. It would know when to quote me, when to summarize me and when to say, “Matthew has not worked that out yet.”

It would communicate differently with a curious reader, a researcher and another AI gathering context about my work.

And if I gave it authority to act, it would need explicit boundaries around that authority.

I can see the beginnings of the architecture:

A knowledge layer containing documents, data and relationships.
An authority layer establishing what is current and who can change it.
A salience layer deciding what matters for a specific interaction.
An identity layer defining who the Ohem represents and how it speaks.
A permissions layer controlling what it can know, reveal and do.
Human and machine interfaces generated from the same underlying mind.

None of those pieces is impossible now.

Most already exist in partial form. Retrieval systems. Knowledge graphs. Agent protocols. Identity models. Policy engines. Structured APIs. Persistent memory.

What does not exist yet is the agreement that these pieces belong together as a new kind of presence.

Not a website.

Not a chatbot.

Not an app.

An emissary.

Websites are documents we visit.

Ohems would be representatives we meet.

Last night that felt inevitable enough for my sleeping brain to give it a name.

Now I need to think through what an Ohem would require.

And then I probably need to build one.