In July 2025, I wrote about the pressure agents put on seat-based pricing. I still expect more software to be priced around usage and completed work. The product question is just as interesting: what are customers paying for when they use less of the interface?
Salesforce’s Headless 360 announcement on 15 April 2026 makes that question concrete. The company described access to its platform through APIs, MCP tools and CLI commands. It announced more than 60 new MCP tools and over 30 coding skills, intended for use through tools including Claude Code, Cursor, Codex and Windsurf.
The same announcement included capabilities for building interfaces and controlling agents in production. Salesforce is making more of the platform available through other tools while retaining the data, business logic and permissions underneath.
For a SaaS company, that is a reason to examine where its value really sits. Some products depend heavily on the way people interact with them. Others have become difficult to replace because they hold the records, integrations and business rules an organization depends on. Many have both.
What the interface has been doing
A good UI carries a surprising amount of work. It makes the next step obvious, sets defaults, asks for required information and guides people through approvals. It also helps them learn the product and understand what matters in a workflow.
When an agent calls the underlying service directly, some of that guidance is missing. The product has to expose the relevant rules and actions in a form the agent can use. Access to a database is not sufficient if the meaning of its fields and the conditions for changing them remain implicit.
This could weaken the position of a product whose main advantage is a convenient way to perform routine actions. It does not follow that every product with a good interface is vulnerable in the same way. In a design or collaboration tool, the interaction can be part of the work itself. An API does not automatically replace that experience.
Established enterprise platforms have other sources of value, including accumulated data, customization and integrations. Those are also sources of switching cost. Giving agents a new way into the platform may strengthen that position as much as it challenges it.
Selling access to useful actions
An agent helping with a lead needs to retrieve the account history, apply the company’s qualification criteria and record what changed. An agent handling a support request needs the relevant policy, the customer’s context and permission to take the next step.
The system of record remains necessary in both cases. What changes is how the work reaches it. A well-defined action is easier to use safely than a collection of low-level operations that leaves the agent to infer the business process.
That is what interests me about the headless approach. The platform can expose capabilities such as updating a case or starting an approved workflow while preserving the rules underneath. Whether it does so well is a product and engineering question. An MCP tool by itself does not make a workflow complete or trustworthy.
A different kind of dependency
I do not expect this change to automatically weaken the large platforms. A provider that holds the data, enforces permissions and offers useful actions has several reasons to remain in the workflow, even if its own screens are opened less often.
For customers, it is worth asking where that leaves their options. Can another agent use the same capabilities? Can they move their records and rules elsewhere? Which parts depend on a proprietary data model? An open interface can make a service easier to integrate while leaving substantial work involved in replacing it.
Pricing is part of that assessment. Metering work at an API boundary is technically convenient, but calls are not necessarily a useful measure of value. A poorly designed workflow can make more calls while accomplishing less. Providers still need a unit that customers can understand and a way to make costs predictable.
I expect product architecture and pricing to influence each other. I would be cautious about treating either as settled while the way customers delegate work is still changing.
A working example from SaaStr
In his account of SaaStr’s setup, Jason Lemkin describes a team of three people using more than 20 AI agents while continuing to rely on established B2B software, including Salesforce.
That is a useful example because the software underneath has not disappeared. The agents still need reliable records and services. It is also one company’s account of its own operation, rather than evidence that the same staffing or workflow model will suit every business.
For me, the question it raises is practical: which parts of a workflow benefit from someone using the application directly, and which can be delegated through well-defined capabilities? That is something a team can investigate without committing to a prediction about the whole software market.
Interfaces for review and intervention
As more routine execution happens through agents, I expect interfaces for supervision to need more attention. A person reviewing an automated process has different questions from a person entering every field by hand.
What has happened? What is about to happen? Which decision needs approval? Where did the information come from? How can I correct or stop the process?
Those questions deserve deliberate interface design. A dashboard full of activity can still leave someone unable to understand a decision or intervene in time. The useful view depends on the work and on who is responsible for it.
Execution and supervision will also overlap. People may need to take over a task, investigate an unfamiliar case or explore an option the agent has not considered. I would keep those transitions visible in the design rather than assume that every workflow can be separated neatly into automated execution and human review.
Making intent explicit
Some organizational rules have been expressed mainly through interfaces: a required field, an approval threshold or the sequence of a form. People learn these conventions by using the system. Agents need access to the underlying meaning and constraints.
At SQUER, we use the term Intent Engineering for work that makes the business objective, success criteria and limits explicit enough to guide implementation. My colleague Manuel Klein describes the approach in Why we created the Intent Engineer.
For a headless workflow, I would want to know which actions an agent may take, what evidence it needs, when it should ask for help and who can change those rules. The organization has to agree on these points before a supervision screen can make them useful.
That is where I would start with an existing SaaS product: choose a workflow, identify what its screens currently teach or enforce, and check whether the same meaning survives when an agent uses it. The answer will say more about the product’s readiness than the number of tools exposed through its API.
Sources and further reading
- Salesforce, Introducing Salesforce Headless 360. No browser required. (April 2026).
- Jason Lemkin, SaaStr’s account of using AI agents alongside established B2B software.
- Manuel Klein, Why we created the Intent Engineer.
- VentureBeat, coverage of the Headless 360 launch.
