· 6 min read · Bytes Coder
What does a custom AI agent cost?
The short answer
The cost of a custom AI agent is set mainly by how many systems it connects to, how much risk its actions carry, and how varied its inputs are. There are two parts: a one-time build, and running costs for model usage, hosting and support. A narrow first agent with one or two integrations costs a fraction of a multi-agent system, which is why most teams should start there.
What you are paying for
Very little of the cost of an agent is the AI model itself. Most of it is the engineering around the model: connecting it to your systems, defining what it may and may not do, testing it on real cases and making sure a person is involved where it matters.
A demo that answers questions can be built in days. A system your team trusts with customer conversations or financial entries takes longer, and that difference is what a quote reflects.
The five things that decide the build cost
When we scope a project, these are the questions that move the estimate most.
- Number of integrations. Each system the agent reads from or writes to (CRM, order system, accounting, WhatsApp) adds work. Systems with a good API are quick; older software without one takes longer.
- Risk of the actions. An agent that drafts replies needs less safeguarding than one that issues refunds or posts accounting entries. Higher-risk actions need approval steps, limits and more testing.
- Variety of inputs. Ten document layouts are easier than a thousand. Clean typed text is easier than voice notes in mixed languages.
- Channels. A web chat is the simplest. WhatsApp adds setup and template approval. Voice adds telephony, speech recognition and latency work.
- Evaluation. Building a test set from your own past cases takes time, and it is the only way to know the agent works before customers meet it.
Running costs after launch
Budget for these monthly costs as well as the build.
- Model usage, charged by the provider according to how much text or audio is processed. This scales with volume.
- Channel fees, such as WhatsApp business message charges or telephony minutes.
- Hosting and monitoring for the agent and its integrations.
- Support and improvement: fixing issues, updating the agent when your policies or systems change, and reviewing its performance.
How to keep a first project small
Pick one process that is frequent, well understood and costly in time. Limit the first version to one channel and one or two systems. Keep a person approving anything sensitive, which also reduces the testing needed before launch.
A narrow agent that works earns the trust and the budget for the next one. A broad agent that half-works does the opposite.
How we quote
We start with a discovery conversation to understand the process, the systems and the volumes. You then receive a written scope with a fixed price for the build and an estimate of monthly running costs for your expected volume, before anything is built.