"Headless software dominated my conversations. Enterprises need to be able to ensure all of their software works across any set of agents they choose. They will kick out vendors that don't make this technically or economically easy."

This was Aaron Levie, CEO of Box, opining on observations he's had meeting with AI leaders across banking, media, retail, healthcare, consulting, tech, and sports, to discuss agents in the enterprise.

Levie's observation was followed up four days later with Salesforce announcing their "headless strategy":

The theme driving both is simple: for underlying systems of record, AI agents will be the ultimate customers of these products, and rather than humans clicking buttons on UIs, the new UI for agents are APIs.

For an incumbent system of record like Salesforce, becoming headless and building API-first is an existential requirement. Absent a headless strategy, Salesforce would become irrelevant as enterprises grew increasingly reliant on AI agents for execution. It's also an offensive move to accrue value, because there will be a near-term world in which there are more AI agents than humans interacting with systems of record, and economic value will accrue to those who empower the agents most effectively.

At Gigi, we're not an incumbent system of record. Our platform partners, like Amazon, are the system of record. We're the intelligence and decision layer on top of the media channels we operate. But similar to Salesforce, we're quickly recognizing that we cannot be dogmatic about our customers logging into the Gigi UI to get the most value out of Gigi. As we articulated in "Our OpenClaw Strategy", we recognized the need to "meet our customers where they work."

Using that principle as a guide for where a headless Gigi should appear, two new surface areas are quickly emerging as priorities for our customers.

Earlier in Q1, an agency leader paused a Gigi onboarding mid-cycle. His reason: "Our internal resources are allocated to competing AI work streams."

That sentence revealed a GTM failure on our end. Customers weren't deciding between Gigi and nothing. They were deciding between Gigi and their own agentic investments, and our pitch had positioned us as antagonistic to work they'd already prioritized.

The fix was obvious once we saw it. Agencies building their own agents shouldn't have to choose between their AI roadmap and ours. So we lead with a different pitch: "Let's build the future you want together, right now."

In practice, that means exposing Gigi through MCP, so any third-party buyer agent can invoke Gigi's capabilities from inside the agency's own technology.

The manifestation of this is profound for many agencies. Rather than relying on limited early use cases for buyer agents, like agentic deal negotiation, they can rely on Gigi to perform end-to-end actions with the Amazon DSP. Amazon DSP is already a high-priority media channel for any buyer agent given its growth and increased relevance. Gigi makes it the most powerful and capable channel they can operate on, and it meaningfully expands what a buyer agent can actually do.

Co-building with buyer agents also lets Gigi benefit from proprietary signals in the agency's tech stack. One of our agency partners, for example, has equipped their agents with proprietary audience signals, competitor intelligence, and holistic media strategies. Gigi uses all of that business context as increased intelligence for every advertiser it touches at that agency.

Almost all of our customers operate in Slack every day (and those that don't do so in Teams). They interact with us in Slack (we've created shared dedicated Slack channels for all of our customers). They interact with their own customers in Slack. And they collaborate amongst each other in Slack. As we build for Slack, we've found that it's uniquely capable of offering a number of benefits to our customers that Gigi alone would not be able to deliver natively.

More surface areas will emerge. Recognizing that Gigi needs to be headless has already reshaped what we build and what we stop building. Design decisions from three months ago are being rewritten. As an example, instead of building tasks around a selection of prescriptive form fields with action-based toggles, tasks should be open-ended chats where Gigi, behind the scenes, rewrites the prompts into robust markdown files for optimal output. The Gigi UI is becoming a command center, a place where an owner sets Gigi up, not the only place Gigi gets used.

That reframe clarifies where the moat actually forms. It won't be in UI components, and the UI doesn't need to live inside Gigi. It will be in the layers underneath: the skills, the proprietary tools, and the memory that stays consistent no matter which surface the customer invokes Gigi from. Those are the layers that compound. Everything above them becomes interchangeable the moment the agents arrive.