An agent that knows what it doesn't know
Most agents model the world in detail and themselves not at all. Agenoia carries an explicit, inspectable model of its own confidence, authority and history — the layer that decides when to act and when to check first.
What "self-model" means here
A self-model isn't a personality or a chat persona. It's a small set of structured facts the agent keeps about its own state, separate from facts about the world: how sure it is of an estimate, how much discretion it currently has, what it already promised, and what it's optimizing for on this particular task. Without that layer, an agent can only be either overcautious — escalating everything to a human — or overconfident — acting on guesses it can't tell apart from certainties.
Five self-models Agenoia carries
Confidence-in-estimate
How sure Agenoia is about a word count, a deadline estimate or a language/subject match — not the provider's confidence, its own. This is what determines whether it can negotiate firmly or has to hedge.
Authority / discretion
A live read on how much rate or deadline concession it's actually allowed to make before it needs a human project manager. This is what stops it from over-promising, or from escalating trivial decisions that waste a manager's time.
Relationship history
Not a model of the provider — a model of what Agenoia itself has already committed to with that provider, across a negotiation and across projects, so it doesn't contradict itself between separate calls.
Goal priority
Which of speed, cost or quality it's currently optimizing for on this project, since that shifts task to task and has to be explicit for the agent to negotiate coherently instead of sending mixed signals.
Verification / escalation trigger
The point where Agenoia's own confidence drops low enough that it stops negotiating and pulls in a person, rather than quietly proceeding on a weak guess. This is stakes-gated: the threshold moves with how much is riding on the decision.
In practice: staffing a translation project
When Agenoia proposes a translator for a task, it isn't just matching a skill profile against a task spec — that part is ordinary state, not a self-model. The self-model kicks in around it: how confident is the word-count estimate it's quoting, how much rate flexibility does it actually have today, has it already made a commitment to this translator this week, is speed or cost the priority on this particular project, and — if two of those are uncertain at once — does this go to a human before it's sent.
Read more on the reasoning behind this in why enterprise agents still fail, or see the practical trade-offs in before you deploy.