Customer intelligence only creates value when teams can act on it. For most of our history, using Treasure AI meant opening a screen: log in, build an audience or journey, activate it, then leave. The interface was the head; customer context and execution lived behind it.
That made sense when the user was always a person, and today it mostly still is: by our own count, 95% or more of the activity on the platform comes from a human driving a UI (user interface). We expect that to flip. Agents don't click through an audience studio / journey builder, and they won't watch a webinar to learn where we moved the segment button.
So we made Treasure AI ‘Headless’ Agentic Experience Platform through supported connections, AI agents can access customer data, segments, journeys, and activation, with skills that teach them how to use the platform well. Over the next 12 to 15 months, there will be no segment or journey created manually, while people set direction, review work, and approve customer-facing actions.
Why now?
AI is changing the economics of work. Inside our own product, we have seen some queries fall from fifteen minutes to seconds in the last 12 months. As intelligence becomes less expensive, more workflows become practical to automate, freeing teams to focus on judgment, creativity, and outcomes.
What stays scarce is not the model; anyone can rent one. The scarce asset is governed customer context: who this person is, what they consented to, what they did an hour ago, under rules your team can defend. We have spent fifteen years building that layer. A headless experience platform makes that context usable in more workflows, so each interaction can inform the next decision.
The date is not a vibe, either. Our own engineers ran the experiment first. Hand-writing code is now getting optional for them; the job moved to guidance and review. That shift is spreading to every department, and marketing is where it shows next: when an assistant builds the segment, the team launches more campaigns, and more campaigns make more money.
What headless means here
Same data, same permissions, same audit trail, more doors. Two are open, and the know-how ships through both.
The tools you already use
Customers told us they did not want yet another destination to check. So at Agentic World 2026 we announced Agent Gateway. Connectors for Slack and Teams will come first. A Remote MCP server, reachable from Claude and ChatGPT Enterprise, follows. If your team lives in Slack, the CDP comes to Slack.
The same goes for the tools people write code in. Claude Code, OpenAI Codex, VS Code with GitHub Copilot, and Cursor each have a documented path: install the CLI, run one auth command, and the platform shows up as an MCP server beside the editor. If you embed Vercel AI SDK into your apps, you can call all of our platform capabilities there.
Your growth marketer and your data engineer end up driving the same platform from different rooms.
The terminal
Treasure Code, the CLI we shipped in February, drives the whole platform in plain English. Behind it sit more than seventy microservices and hundreds of API endpoints. One side effect surprised us: agents find features that humans never do. The product has a lot of switches, and nobody reads the documents. An agent reads the documents via mcp.
The 50+ skills learned from 400+ CDP implementations
A connection is not knowledge. An agent with a key can call the API; it still will not know which journey pattern converts, or which join resolves the identity graph. It sees every switch in the cockpit and has no idea which one to flip.
That is why we pair connections with skills informed by more than 400 CDP implementations: journey and email design, landing pages, SQL optimization, platform configuration, and SDK integration. APIs tell agents what they can do. Skills guide them on how to do it well.
This is what makes "openness" practical. Point Claude Code or Codex at Treasure AI, download the skills, and it learns how to use the platform right away.
Nobody has to sit through a webinar. And the more agents that know how to use the platform, the more people in your organization can actually reach your customer data.
What headless does not mean
Headless does not mean the end of the UI. Treasure AI Studio remains the visual workspace for people; headless access adds another way to work with the same governed platform.
It does not mean the end of judgment, either. An agent works within the permissions and guardrails you give it. Each conversation runs in its own isolated container, its network connections and file access traced, and every action lands in the audit log. Anything customer-facing, like a send, still waits for explicit human approval. If an agent acts on your behalf, you remain accountable for it. That has not changed, and it should not.
A Tuesday, not a keynote
Demos are fun. The thesis lives or dies on a Tuesday.
It is 11pm. An agent notices a repeat-purchaser cohort going quiet, checks the last three campaigns, drafts a win-back variant, and stops to wait. In the morning, a marketer reads the reasoning, changes two lines, and approves. The send goes out, the results feed back, and the next decision starts from a better place. Nobody opened the dashboard at 11pm.
This is an illustrative composite, not a customer story. It shows the operating model we are building toward.
The part that stays yours
A UI serves the people who log in. A headless platform also makes customer context and approved actions accessible to agents working on their behalf. The 11 p.m. example is one possibility; the larger opportunity is helping teams run more governed customer workflows without requiring every person to stay in the interface.
None of that diminishes the marketer's job. It removes the part that was never the job. Clicking through a segment builder was always the toll you paid to reach the real work: deciding what to want, for whom, and why. Within fifteen months, the toll is gone. The real work stays yours.
Open to any AI
Most platforms want AI to run on their platform. Bring the work to their agent, inside their walls, and everything you learn stays there too.
We are building for a different choice. We will keep making our own AI better; Treasure AI's agents are not going anywhere. But they will not be the only way in. Whatever AI your team chooses (Claude, ChatGPT, Copilot, or one you build yourself) should be able to reach your customer data with the same context, under the same rules, and with the know-how to use it well.
The reasoning goes back to what is scarce. The model is the part anyone can rent, and it gets replaced every few months. Governed context is what lasts. Locking that context behind one AI, even ours, would make the scarce thing depend on the replaceable one. So we build the platform open, and let our AI earn its place by being the best one at the table, not the only one allowed in.
Summary
Treasure AI is now ‘Headless’ Agentic Experience Platform beyond the UI, so supported AI tools can work with governed customer context, segments, journeys, and activation. The platform pairs governed access with skills shaped by 400+ CDP implementations, helping agents use it effectively from day one.
Treasure Code, our CLI, is available today and works from Claude Code, Codex, Copilot, and Cursor. Agent Gateway with a remote MCP server are rolling out now.
Agents operate within defined permissions, and customer-facing sends require explicit human approval. Our goal over the next 12 to 15 months is to make agent-assisted segment and journey creation the norm—while people continue to set direction, approve actions, and remain accountable for outcomes.
Let’s prepare for that future together in the NOW.
