There’s a particular textbox in popular HR software Workday I want you to picture. It’s the one for performance review comments. Three lines deep, eighty characters wide, with a tiny scrollbar that fights you when you try to write anything longer than a haiku.
Sunday night, laptop
Year of work, three-line textbox
“Meets expectations”
A whole year of someone’s work and growth gets squeezed through that mailslot, on a Sunday evening, by a manager uttering a heavy, heavy sigh.
Now picture the same rectangle in your CRM, your expense tool, your HRIS, the ITSM form the help desk wants filled in before they’ll talk to you. Different vendors, same rectangle. The interface tax that systems of record charge their occasional visitors has been remarkably consistent for two decades. We’ve all just gotten used to paying it.
But wait! Salesforce recently announced Headless 360 and underneath the marketing it’s a quiet admission. The screen is not the product. The model is the product, the data, the workflows, the guardrails… all of it exposed through APIs and agent interfaces, available to be consumed by whatever front end you want to put in front of it. And it’s more than just a quick API against the backend, it’s thoughtful packaging. Salesforce is far from the only one moving in this direction. Workday, ServiceNow, the major HRIS and ERP vendors all have headless or agent-callable variants in flight or in pilot. The industry has finally noticed that the rectangle is the part their users loathe.
The Occasional User Problem
Most people don’t live inside these tools. They visit. A line manager opens Workday twice a year for reviews and once for promotions. A relationship banker logs into the CRM when a deal forces them to. A clinician touches the HRIS when something administrative breaks. The interface was designed for power users, the small minority who live there all day, and it punishes everyone else. The result is the worst of both worlds. Casual users either skip the system entirely or feed it bad data, and leadership ends up running dashboards on signal that was never properly captured in the first place.
The Composable Promise, Finally
Gartner coined “composable architecture” around 2020, and for a few years it was a term that lived mostly in slide decks. The idea was that organizations would stitch together best-of-breed components, swap them out as needs evolved, and avoid getting locked into any single vendor’s ecosystem. It was a beautiful diagram and a frustrating reality. Every system insisted on being its own universe, with its own front door, its own UX patterns, its own assumptions about who the user was and what they wanted.
Headless changes the math. When the system of record is content to be a service, composing across systems stops being a vendor-relations problem and becomes a plumbing problem. The conversational layer, or the workflow layer, or whatever the user actually touches, can finally span the underlying systems in a way that respects how the work flows.
But the screen was only one of the locks. There are at least three more, and they’re the reason composable architecture has stayed mostly aspirational so long after the term entered the vocabulary. The systems still keep their data on a short leash. Stitching anything coherent together at runtime is its own discipline that very few organizations have built yet.
The composable promise assumes that data flows out of these systems as readily as instructions flow in. That’s a generous assumption. SaaS vendors have spent two decades building proprietary data models, and going headless doesn’t undo that overnight. The API surface might be open. The data semantics underneath it are still the vendor’s. Workday’s notion of an employee, Salesforce’s notion of an account, ServiceNow’s notion of an incident… these are all subtly different shapes of what looks like the same thing. Composing across them means resolving those differences in your own integration layer, every time, for every workflow.
There’s also the question of how much of the data (your data!) the vendor will let you see, and how fast. Bulk extract limits, rate caps, premium tiers for real-time access. The headline says headless, the fine print says you can pull what you want as long as you don’t pull too much of it too quickly. The vendors are not unaware. They know that data gravity is what kept customers in for the last twenty years, and they will not give it up gracefully. Expect a period of negotiation, ugly contract addenda, and selective cooperation as customers test what headless actually means in practice.
The Commercial Model Question
The other quiet hurdle is pricing. SaaS sold by the seat made sense in a world where humans logged in. When the primary consumer of the system is an agent, what’s a seat? Vendors are working that out, and most of the early answers point toward consumption pricing that mirrors how the AI vendors charge for tokens. Per API call, per workflow execution, per record modified.
That shift sounds technical, but it’s a real procurement problem. Seat licenses were predictable… you knew how many people were on your team and you could budget against it. Consumption pricing is volatile by nature, and agents in particular are inclined to retry, loop, and fan out queries in ways humans never would. A poorly governed agent can run up a SaaS bill the same way it can run up an AI vendor bill, and most enterprise procurement functions are not set up to monitor thousands of metered SKUs across multiple vendors in real time. The ole’ pay by the drink with an open bar tab model.
Orchestration Is the Whole Game Now
On top of the data and the contracts, there’s the runtime problem. Who decides which system answers which question? How does identity and permissions travel as an agent moves between backends? How do you preserve audit trails when a single human sentence touches four systems? How do you keep business logic from being silently rewritten in a prompt? Oh, ye of many questions. These are not solved problems, and most organizations do not yet have the muscle for them.
What This Means If You’re a SaaS Buyer
There’s been a lot of talk about a SaaSpocalypse, the idea that AI is going to hollow out the SaaS industry. I think that’s both true and overstated. The big platforms are not dying. But the basis of their moat is shifting but is just as deep. When the interface monopoly is gone, the line of defense becomes the data model, the embedded workflow logic, the regulatory wrapper, and the integrations that took years to mature.
It may be worth rethinking funding proprietary UI work in your big platforms; that spend is depreciating quietly while you’re not looking. Instead, it’s a great time to experiment with alternate models. Pick one casual-user, high-friction workflow and pilot agent-mediated access against it. The user response will tell you more than another roadmap deck would. While you’re at it, check your SaaS contracts before your renewals come due for data egress terms and where the vendor is heading on consumption-based pricing. Those will start to matter and your leverage is much higher before the meter starts than after. Then start building the orchestration competency now, because identity federation, audit, and governance across agentic flows is a capability you can’t buy off the shelf yet, and the companies that wait for a vendor to solve this will inherit the vendor’s choices.
And watch the performance reviews roll in.
