AI agents have limits
In the model an AI agent is an organisation member just like a human: it fulfils roles, uses competences and appears in the same candidate list. And it has limits — just different ones.
This is the statement this view has to carry, because it is not anchored in most people’s minds: an agent is not an unlimited colleague.
Two limits, and either can bind first
Section titled “Two limits, and either can bind first”| Limit | What it means |
|---|---|
| Throughput (h/week) | It needs time to work. Even with unlimited budget it manages only a certain amount per period. |
| Budget (amount/period) | Every run costs. A different currency. |
Both are shown side by side, in the same visual vocabulary as for a human, marking which one binds first.
The budget is deliberately not converted into hours: a cost-per-hour rate for an agent would be an invented figure. The price is that an agent has no single “utilisation in %” but two — which is the more honest presentation.
Never “unlimited”
Section titled “Never “unlimited””If a limit is missing it says “no limit on record” — no infinity sign and no filled bar. That is exactly the difference between “has no limit” and “we have not entered one”.
The tenant norm for weekly hours does not apply to agents either: using it on an agent would be an invented figure.
roleALPHA does not run agents
Section titled “roleALPHA does not run agents”The module describes an agent and provides it with context over MCP — it is executed in the customer’s own runtime. Its tokens and costs arise there and are not measurable here.
What is planned and limited here is therefore a commitment (“this much work and this much budget is granted to this agent”), not measured consumption. Just like with a human: their capacity too is in the contract, not on a stopwatch.
Not to be confused with rALPH. The assistant runs inside the application, and its consumption is indeed logged. That is a different matter — those logs say nothing about an agent.
Reporting consumption — optional
Section titled “Reporting consumption — optional”If you want to see actual consumption alongside, report it from the agent’s runtime (per agent and period: tokens, cost, runs). Without that report the budget reads “consumption not reported” — not “0”. A zero would mean: the agent consumed nothing. roleALPHA does not know that.
While the report is missing the budget bar is incomplete, and it says so. A half-filled budget that looks like a full one would be worse than none.
Related
Section titled “Related”- Reading the candidate list — Agents in the candidate list
- Capacity — baseline, deductions, available — Capacity for humans
- AI Agents — The “AI agent” entity type
- Demand & Resource Planning — The overview