What Everyone Is Building — and What's Missing
Every vendor hands AI its own records. Nobody hands it the operating model.
What Everyone Is Building — and What's Missing
Every major platform is building toward the headless enterprise. It helps to be precise about what each one actually hands to AI.
| Platform type | What it opens to AI | What its AI understands well | What it can't see |
|---|---|---|---|
| CRM and customer platforms | Accounts, opportunities, cases, conversations | Customers and the work around them | Whether the work is funded or strategic |
| Workflow and IT service platforms | Tickets, requests, workflows, approvals | How work moves through a process | What the process is for, and what it costs |
| ERP and finance platforms | Ledgers, orders, close tasks | Transactions and their rules | Which strategy and portfolio the money serves |
| HR and workforce platforms | People, roles, pay, performance | The workforce record | Which work the people are on and why |
| Work and collaboration tools | Projects, pages, issues, messages | Activity and conversation | The plan, the money and the controls |
| Enterprise search | Everything it can index | Finding documents | Structure, relationships and authority |
Each of these is a legitimate system of record. Each now exposes itself to AI, usually through MCP, usually with the user's own permissions. That's real progress.
But notice what's in the right-hand column. None of them hands AI the operating model: the connected picture of strategy, money, portfolio, architecture, controls and decisions. Each one gives AI a deep view of one slice and no view of how the slices relate.
Pricing is changing shape
The commercial model is moving with the architecture. Seat-based licenses are giving way to pricing per conversation, per action or per unit of consumption, often in parallel. IDC has predicted that pure seat-based pricing will be largely obsolete by 2028. For buyers this changes the budget envelope:
| Per-seat software | Agent economics |
|---|---|
| Predictable annual budget | Variable spend driven by usage |
| Tied to headcount | Tied to work done |
| The vendor wants more seats | The vendor wants more actions |
| Renewal debates price | Renewal debates scope and outcomes |
| Procurement can manage it | Finance needs FinOps discipline for AI |
The control-plane race
The large platforms are also each competing to be the control plane — the place where every agent is registered, observed and governed. That's useful. It's also the beginning of a new kind of lock-in: if one vendor owns the model, the agent runtime, the tools, the identity and the governance, leaving gets harder than leaving a cloud ever was.
What's missing
Put the pieces together and one layer is absent from almost every architecture: a vendor-neutral operating model that every AI — whichever vendor built it — can read, that carries the organization's own permissions and rules, and that records the decisions people make. That's the layer this playbook is built around.