What does the myth say?
That the point of a capable model is generality, so a business should have one agent that does whatever is asked: reconcile this morning, answer tickets this afternoon, draft a campaign tonight. Separate roles look like an artefact of selling seats. The belief is attractive because it promises one thing to set up and one thing to manage, and it is wrong for the same reason a company does not hire one person to be the whole company.
The examination is about what is genuinely shared and what has to be separate.
Where does the belief come from?
From the models themselves, which are general, and from assistants that answer any question. From frameworks that present an agent as a loop with tools, where adding a tool looks like adding a job. From vendors whose product is one agent with a growing list of skills. And from the honest observation that much of what a Digital FTE knows about a company, its voice, facts, and policies, is the same regardless of function, so why not one agent that knows it all.
The last observation is correct and is exactly what the platform shares.
What is true in it?
That the underlying model is general and the same across roles. That the company knowledge is shared: the Brand Brain is organization-wide, pre-filled once and never re-asked, and every role reads it. That the wallet is one, the workspace is one, and the manager's morning starts in one combined brief across roles. That a role can be composed for a function no catalogue role covers. Generality is real at the model, the memory, the money, and the management layer, and the platform is built to share all four.
What is not true is that the role, the remit, the scopes, the budget, the record, and the targets should be shared too.
What does the mechanism actually do?
Each Digital FTE on ftes.ai is a role: a remit stating what it owns and never does, connections scoped to that remit's tools and one property, a per-role budget inside the organization wallet, its own approval record and autonomy ladder per class, its own targets with evidence, and a named manager. The workforce view combines every role's brief into one list ordered by what blocks the most, and the operations lead manages from there. Usage and cost per outcome are reported per role, so a Finance Controller FTE's tokens and a Support FTE's tokens are never blended, and a promotion in one role never touches another.
The whitepaper on hybrid team design in the library describes how roles are added and managed together.
- Shared: model, memory, wallet, workforce view
- Per role: remit, scopes, budget, record, targets, manager
- Per class within a role: autonomy level
Why would one general agent be worse?
Because its scope would be the union of every function's scope, a single credential set that can read the ledger, the helpdesk, the CRM, and the site, which is the blast radius every security lead refuses. Because its approval record would blend outcomes that mean different things, so a clean record on order-status replies would appear to justify autonomy on payout matching. Because its cost could not be read per function, so no return could be computed. And because nobody would manage it: the head of support and the finance lead would share one direct report with one send-back log, and neither would read it.
Accountability needs a role. That is not a pricing artefact; it is how a manager knows what to review.
How can you check for yourself?
Take a week of send-back reasons from two functions, or imagine them: reconciliation send-backs about fee schedules and timing, support send-backs about tone and knowledge gaps. Put them in one list and try to decide what to tune. Then compute cost per outcome from one blended usage figure across a reconciled transaction and a resolved ticket. Then decide, from one blended approval rate, whether to promote a class. Each is impossible, which is the argument. Then open the workforce view with two roles and see that the combined brief gives the one-screen management the myth wanted, without blending anything underneath.
The piece on running multiple Digital FTEs describes what the second role adds and what it does not.