AI Agents

An AI agent is a system that answers or acts without a person driving each step: a support agent that resolves the routine half of the inbox, an internal knowledge agent that answers staff questions out of your own documents, a sales agent that qualifies enquiries and books time. The distance between a demo and a production agent is everything wrapped around the model, retrieval over real data, a defined and limited set of actions, permissions, escalation to a human, and an evaluation suite that tells you whether it is still working.

An AI agent build is $12,000 to $30,000 over six to ten weeks and starts with a discovery conversation, because scope varies more here than anywhere else in the catalogue. Typical engagements: a customer-facing support agent grounded in product documentation and order data; an internal agent over policies, contracts and past projects; an agent that drafts quotes or replies for a person to approve; an agent that reads inbound documents and writes the result into Xero, HubSpot or Shopify under controlled permissions.

Delivery follows a fixed order. Scope the actions the agent is allowed to take, then retrieval over your documents and data, then tool use against your systems with read and write kept separate, then guardrails, permissions and a clean escalation path to a person. The evaluation suite and monitoring go in before launch rather than after, so the failure rate is a number you can look at instead of a feeling. A privacy and data-handling review is part of every build and documents which data the agent sees and where it goes.

This suits businesses with genuine volume in a repetitive conversation or judgement task, and with the documents or data to ground it in. It is the wrong purchase if the underlying content does not exist yet, if the process changes weekly, or if the honest fix is a better form and a clearer FAQ page. Where an agent is not the answer, the audit says so, and a $6,500 to $15,000 workflow automation build is usually the cheaper result.