← Notes

What an agentic pipeline charges to cover the absence

Agentic systems trade latency and cost for task performance, and the decision is about when that trade makes sense; in a service with nobody on call, the comparison is not the ideal, it is silence.

Anthropic's engineering guidance on building agents frames the choice in the right terms: "Agentic systems often trade latency and cost for better task performance, and you should consider when this tradeoff makes sense".

In a service with a team, that trade competes against a competent human who replies fast, and it frequently loses. In a service with nobody on call, the comparison changes: the real alternative is not the immediate human reply, it is no reply for hours. Against that starting point, a few extra seconds of latency and a cost per conversation stop being the axis of the decision.

What changed in the feasibility of building this alone is the standardisation of connecting to systems. The Model Context Protocol describes itself as "an open-source standard for connecting AI applications to external systems" — the same interface for the agent to reach a database, a calendar, a tool and a flow, instead of a hand-written integration per system.

That does not make the agentic pipeline free or guaranteed. It makes the cost of trying compatible with a small budget, which is different. And it leaves standing the demand of the guidance itself: deciding when the trade makes sense is a claim about the concrete case, and it needs a measurement — how many conversations are lost today, and how many of those the agent would have held.