We went the opposite direction from Salesforce - and here's why

Strip Headless 360 of the announcement theatre and the architectural argument is this:

Agents are a new kind of user. They don't click. They don't navigate. They consume context, select tools, and act. If your platform requires a graphical interface to be useful, it's built for a constraint that no longer exists. So expose everything - data, workflows, business logic, compliance controls - as programmable endpoints. Let any agent, in any tool, reach into your platform from anywhere. The GUI was a convenience for humans. Agents don't need it. Remove the friction.

Some work, Salesforce concedes, still needs a visual component. So they built the Agentforce Experience Layer: a service that renders cards, workflows, and decision tiles inside Slack, Teams, ChatGPT, Claude, Gemini, mobile, voice, WhatsApp. The agent's intent is decoupled from its presentation. Build once, render everywhere your people already work.

This is, if nothing else, intellectually coherent. It takes a position. It commits. And it correctly identifies the structural question every enterprise platform is now forced to answer: in a world where agents can reason, plan, and execute, do you still need a CRM with a graphical interface?

Our answer is the opposite of theirs. And in regulated verticals, our answer is right.

Headless 360 is optimised for one question: how do we make enterprise software legible to agents? It is a brilliant answer to that question.

But there's a different question that's much more important to anyone actually running a regulated business: how do we make our operations legible to ourselves?

These are not the same question, and conflating them is where the Salesforce bet breaks down.

Agents acting through chat hides process, not complexity

When a claims handler asks a chatbot to "update the case and raise a search," a lot happens. The agent reasons about intent. It selects tools. It calls APIs. It writes to records. It triggers workflows. It sends emails. The user gets a cheerful confirmation: "Done. I've raised the environmental search with Birmingham LA."

The user sees two things: their request, and a response. What they don't see is the process. They don't see the shape of the work. They can't tell you at a glance which cases are blocked, why, and what needs human judgement next. They can't look at the business and understand how it's running - because the business isn't on a screen. It's in a transcript.

Transcripts are a terrible medium for operational awareness. Ask anyone who's ever tried to audit a Slack channel after a system incident.

The "Experience Layer" is just a UI with a new name

Salesforce's own admission is instructive. Even they concede some work needs more than conversation: approvals, workflows, decision tiles, data layouts. So they built the Experience Layer to render these across six different surfaces - Slack, Teams, ChatGPT, Claude, Gemini, WhatsApp.

Which means: to run your business on Headless 360, your staff now work across six chat surfaces, each rendering ad-hoc cards generated per-interaction, with the state of the operation scattered across transcripts in each of them. That's not simplification. That's a new kind of fragmentation - the same disease under different branding.

And critically: these are not persistent operational surfaces. They are ephemeral rendered responses to conversational turns. The card you saw yesterday is not the card your colleague sees today. The picture of the business is assembled, per user, per moment, from a model's reasoning - and then dissolved.

The regulatory failure mode

This is where the argument doesn't just get weaker. It gets structurally broken for the businesses we serve.

When a regulator asks a conveyancing firm to explain why a specific decision was made on a specific case - a real, routine request - the firm needs to reconstruct exactly what happened. Not an approximation. Not a summary. The actual state the operator was looking at, the exact action they took, and the chain of reasoning and authority that led there.

In an architecture where interactions happen through ephemeral chat responses across six different surfaces, reconstruction is a forensic exercise. You're stitching together transcripts, tool call logs, model outputs, and state changes after the fact. You're hoping nothing was lost. You're hoping the model's summary matches what actually happened. You're hoping the rendering logic that produced the card the operator saw yesterday is still the same logic today.

In an architecture where every interaction happens on a persistent, versioned, audit-captured surface, reconstruction is the default. You look at what the surface looked like at 14:37 on April 14th. You see what was presented to the operator. You see what action they took. You see which agent acted, with what authority, on what entity state. There is nothing to reconstruct because nothing was transient in the first place.

Here's the comparison without marketing varnish on either side.

There's no objectively correct answer here. These are two different bets, each coherent in its own logic, each strong for different buyers.

Salesforce is betting that agent-first conversational enterprise is the winning architecture for the mass market - CRM, customer service, marketing automation, commerce. That's a defensible bet. Conversational interfaces work well for customer-facing workflows where the question is "help me do X" and the correctness bar is "did it get done."

Our bet is that for regulated operational businesses - conveyancing, property management, financial services, accountancy, healthcare back-office, insurance claims - the correctness bar is fundamentally different. It's not "did it get done." It's "can we explain, reconstruct, and improve exactly how it got done, for every case, on demand, under regulatory scrutiny." That's a different architectural requirement. It needs surfaces, not transcripts.

If you run a marketing team, Headless 360 might genuinely be the better architecture for you. An agent that spins up a campaign brief in Slack, renders an approval card, executes the spend, and reports back - that's an excellent workflow. The measurement model is clear (did the campaign perform?). The audit requirement is low.

If you run a conveyancing practice, the calculus is completely different.

You have a regulator (SRA or CLC) who can request complete reconstruction of any case at any time. You have clients who trust you with the largest transaction of their lives. You have staff who need to know, at a glance, which cases are blocked and why. You have compliance obligations that are not satisfied by "the model decided." You have outcomes - exchange times, error rates, client complaints - that you need to measure consistently across hundreds of cases, year after year.

None of this is served by conversational agents acting on your behalf across six chat surfaces. It is served by operational surfaces where every case is visible, every action is captured, every decision is traceable, and every outcome is measured against the same rubric.

SurfaceOS is an agentic operating system for regulated business. It has two parts - a kernel and a UI layer - just like any operating system.

The kernel is entities (purpose-built to the vertical - cases, parties, chains, certificates, tenancies), agents (autonomous, scoped, authority-bounded), a safeguard layer (evaluating every write action and external communication before execution), and execution capture (every action, every model call, every state change, persisted as a permanent structured record).

The UI layer is surfaces. Not dashboards. Not chat interfaces. Persistent, composable, agent-native interfaces where the business actually runs. Staff surfaces. Client surfaces. Agent surfaces. All working off the same entity state. All captured in the same audit trail. All measured against the same operational outcomes.

Every surface is the process itself. When a case handler looks at the case surface, what they see is the case. Not a report about the case. When an agent updates entity state, the surface updates with it. When an operator makes a decision, the surface records it. When a regulator asks what happened, the surface history is the answer.

This is not a better CRM with AI features. It is a different architectural response to the same shift Salesforce is responding to - just from the opposite end.

Headless 360 is a serious piece of architecture from a serious company. It's not wrong. For the right buyer - horizontal, conversational, high-velocity - it may well be exactly right.

But it is built on an assumption that does not hold universally: that the UI is overhead to be removed, rather than the operating surface itself. For regulated operators - the lawyers, the property managers, the accountants, the insurers, the clinical operations teams, the financial services back offices - that assumption is wrong.

The UI is not overhead. It is how your business exists as an observable, measurable, improvable thing. Remove it and you don't get efficiency. You get a business that runs on transcripts and hopes no-one asks for reconstruction.

We went the opposite direction from Salesforce because, in the verticals we serve, the opposite direction is the right one.

Matt Todd - Founder, FluentForward - Building SurfaceOS, the agentic operating system for regulated business.