How much does it cost to build an AI agent for a business?

An AI agent is software that uses a language model to decide its next step and call tools, such as reading an inbox or writing a record. The model is rented by the token. Everything around it is built by people, and that is where the money goes.

Five lines make up the build. Each one can be estimated before work starts, once the workflow is named and the systems it touches are counted.

Take a finance team that receives supplier invoices in a shared mailbox. The agent reads each invoice, finds the matching purchase order in the finance system and drafts the posting. A person in accounts payable approves it before anything is posted. The agent reads two systems and writes a draft to one.

The person-days below are our assumption for this scope. The prices are 1AYM’s published day rates, which start at £750 to £1,250 depending on seniority and the project. The totals are our arithmetic, for illustration only. Real fixed-scope work is quoted after scoping.

Two things move this example most. If the finance system has no usable API, the integration line grows, because the agent needs another way in. If the invoices are scanned images in several languages, the test set grows and so does its line.

Data, infrastructure and access work sit outside this table. The purchase orders may be spread across systems that disagree, or the agent may have nowhere approved to run yet. Then that groundwork comes first and is priced on its own.

To sense-check a quote against the market, our guide to agents for business gives indicative build-cost bands drawn from agencies’ published price guides. Read those bands as a rough guide, because none of them has seen your data.

The person-day ranges below are our assumption for each type of agent. The prices are 1AYM’s published day rates, so the bands are our arithmetic, for illustration only.

The bottom of each band assumes your data is ready and your systems have usable APIs. The top assumes some of that work lands inside the build.

Model fees are charged per million tokens, a token being a piece of a word. On 22 October 2026 Anthropic listed Claude Sonnet 5.5 at $2 per million input tokens and $10 per million output tokens. OpenAI listed GPT-6.1 Sol at the same $2 and $10. Their small models, Claude Haiku 5.5 and GPT-6 Luna, were each $0.10 and $0.50.

For the invoice agent, assume 2,000 invoices a month. Each uses about 20,000 input tokens and 2,000 output tokens across all its model calls. On either mid-tier model that is $0.06 an invoice, or $120 a month. On either small model it is $0.003 an invoice, or $6 a month. Those are our assumptions and our arithmetic at the vendors’ prices.

Hosting on Anthropic’s Claude Managed Agents adds $0.08 for each hour of running time, on top of tokens. At two minutes of running time per invoice, 2,000 invoices is about 67 hours, or about $5 a month on our arithmetic.

Someone has to read the monitoring, retest the agent when a vendor changes a model, and fix what the business changes. At our assumed two person-days a month, that upkeep is £1,500 to £2,500 a month at 1AYM’s day rates, again our illustration. Over a year it outweighs the model bill many times over.

Before paying for a build, check whether a business plan from OpenAI or Anthropic already covers the job. Both sell workplace plans, priced per seat, that list agent features such as ChatGPT Work, workspace agents, Claude Code and Claude Cowork.

For 50 people on ChatGPT Business or Claude Team, billed annually, seats come to $1,000 a month or $12,000 a year on our arithmetic. That is often the cheaper route for work people do at their desks, such as research, drafting and analysis.

Work a person drives in a chat window suits a bought agent. Work that runs on its own inside your systems, at volume, with checks a vendor cannot know about, suits a build.

A build pays back where the volume is high, the steps are repeatable and a mistake can be caught before it lands. These are typical cases, with what sets the cost of each.

If you are pricing an AI agent for a real workflow, bring in 1AYM. Our Enterprise AI Platform & Agentic Workflows engagement is a fixed-scope architecture plus production build. We scope the workflow, build the agent, connect it to your systems, and put the tests and approval steps around it. Fixed-scope work is quoted after scoping, and it can start within a day of the scope being signed.

Many of our engineers have worked inside large Silicon Valley technology companies, and they hold Anthropic’s Claude certifications, the founder among them. They are senior engineers who build working systems, priced by the day, with no six-figure slide decks. Day rates start at £750 to £1,250, depending on seniority and the project, and our pricing page sets out how 1AYM prices its work.

We are model-selective across OpenAI, Anthropic, Gemini and open-source, so the agent runs on the model that suits the job and its budget. 1AYM is an OpenAI Select Partner. We roll out ChatGPT Enterprise and Claude Enterprise, and we have rolled out Claude Enterprise for a client. We take small fixed-scope statements of work as well as larger builds, and we work with clients across the UK, the US and the Gulf.

Book a 30-minute call about the workflow you have in mind. We will tell you roughly what it would take to build and to run. Or email enquiries@1aym.com.