Straight answer: a simple no-code AI agent costs somewhere between £0 and £300 a month to run, a custom-built one costs £3,000 to £80,000 to build plus £200 to £3,000 a month to keep running, and almost every quote you’ve seen online leaves out the part that blows the budget, which is the ongoing babysitting. The build is rarely the expensive bit. The running, fixing, and supervising is.
Why nobody can give you a straight number
Every article that promises “here’s exactly what an AI agent costs” and then gives you one figure is guessing, or selling something. The honest answer depends on three things: what the agent does, how many steps it takes to do it, and how much you’re willing to let it run unsupervised. A chatbot that answers FAQs from a knowledge base is a different animal entirely from an agent that reads emails, checks your CRM, drafts a quote, and books a follow-up call without asking you first. People use the term “AI agent” for both, which is part of the problem. If you want the actual distinction between a simple assistant and something that plans and acts on its own, I’ve written about the real difference between AI agents and agentic AI because vendors blur it on purpose to make everything sound more impressive than it is.
Build cost: three routes, three very different bills
There are three ways to get an AI agent into your business, and I’ve now done all three, either for my own consultancy or for clients.
Route one: no-code, DIY
Tools like Zapier’s AI agent builder, Make.com, n8n, and Voiceflow let you assemble an agent from templates without writing code. Zapier’s paid plans run £19.99 to £799 a month depending on how many tasks you run through it. Make.com starts around £9 a month and scales to £299+ for higher operation volumes. n8n is free if you self-host it on your own server, or around £16 to £48 a month on their cloud plan. If you’ve never tried any of these, I’ve mapped out which AI agent tools are free to try and where the free tier runs out, because most of them let you build something impressive for nothing and then charge you the moment it starts working.
Build cost here: mostly your own time. Budget 8 to 20 hours if you’re doing it yourself for the first time. If you pay a freelancer to set it up for you, expect £300 to £1,500 for a single-purpose agent like a lead qualifier or a booking assistant. I’ve written a full walkthrough on where to build your own AI agent without coding if this is your starting point, because for most small businesses it should be.
Route two: custom build with a developer
This is where you’re paying someone to build on top of the OpenAI, Anthropic, or Google APIs directly, with custom logic, memory, and integrations into your own systems. UK freelance AI developers charge £400 to £900 a day. A single-purpose custom agent (say, one that triages support tickets and drafts responses) typically takes 15 to 40 days of work, which puts you at £6,000 to £36,000. A multi-agent system that coordinates several tasks across departments can run £40,000 to £150,000+. These aren’t scare numbers, they’re what I’ve seen quoted to clients in the last year.
Route three: an agency
Agencies bundle strategy, build, and support. Expect £5,000 to £50,000 for the build phase and then a monthly retainer of £500 to £5,000 for maintenance and updates. Some agencies will quote you a low build fee and make their real money on the retainer, which is worth knowing before you sign, not after.
Running cost: the number that surprises people
Here’s where most articles stop being useful. Everyone talks about the API price per token as if that settles it. It doesn’t, because agents don’t make one call, they make many.
OpenAI’s GPT-4o API costs roughly $2.50 per million input tokens and $10 per million output tokens. A single customer query might only need 500 tokens in and 300 out, which sounds like fractions of a penny. But an agent isn’t answering in one shot. It reads context, calls a tool to check your CRM, reasons about the result, maybe calls another tool, drafts a response, and sometimes retries a step if the first attempt fails a check. That one “simple” customer query can trigger 6 to 15 model calls behind the scenes. Multiply your assumed cost by 10 and you’re closer to reality.
What that looks like with real numbers
Last year I built a lead-qualification agent for my own consultancy. The idea was simple: someone fills in a form on my site, the agent reads their answers, checks them against a scoring rubric, drafts a personalised follow-up, and flags the hot leads for me to call that day. I priced it out beforehand at maybe £30 to £50 a month based on the “per query” cost the platform advertised.
The first month’s bill was £340.
What happened was a loop. The agent would score a lead, decide the scoring wasn’t confident enough, re-check the data, score it again, and occasionally do that three or four times on a single form submission before settling on an answer. Nobody designs an agent to waste money on purpose, but poorly bounded reasoning loops are the single most common reason small businesses get a shock bill in month one. I fixed it by capping the number of reasoning steps and adding a hard rule that says “score once, flag for human review if confidence is below 70%, don’t re-check.” The bill dropped to £38 a month and stayed there. That single change, one line of logic, cut running cost by nearly 90%. Nobody puts that in a sales deck, but it’s the difference between an agent that’s cheap and one that quietly isn’t.
The costs vendors don’t mention
This is the part that gets left out of nearly every “cost of AI agents” piece I’ve read, and it’s the part that matters most if you’re deciding whether to build one.
- Human oversight time. Somebody has to check the agent’s outputs, especially in the first few months. Budget at least 2 to 5 hours a week of a real person’s time reviewing what it did, even for a “fully automated” agent. That time has a cost even if you don’t invoice yourself for it.
- Drift and maintenance. Your CRM changes a field name, your pricing updates, your tone of voice shifts. The agent doesn’t know unless someone tells it. Plan for a few hours of maintenance monthly, or a maintenance retainer if you paid someone else to build it.
- The rogue agent problem. Agents that are given too much autonomy without guardrails can take actions you didn’t intend, send the wrong email to the wrong list, or make a decision based on bad data with total confidence. I’ve written in detail about why AI agents go rogue and how to stop it happening to your business, and the short version is that the fix always costs less than the cleanup, but people skip the fix because it’s not exciting.
- Failed experiments. Not every agent you build will earn its keep. I’ve scrapped two out of the five I’ve built for my own business because the maintenance overhead wasn’t worth what they saved. That’s a real cost too, and it’s the one nobody puts in a case study.
Here’s the uncomfortable bit that most people selling AI agents won’t say out loud: for a lot of small businesses, a cheap agent that needs constant supervision costs more in your time than a slightly more expensive one that’s been built with guardrails from the start. Cheap and unsupervised is not the same as cheap.
What it costs by agent type
To make this concrete, here’s roughly what I’ve seen across different use cases, combining build and the first year of running cost.
- FAQ or support chatbot answering from a knowledge base: £0 to £1,000 to build on a no-code platform, £30 to £150 a month to run. This is the cheapest category by far because it doesn’t take independent action, it just answers.
- Customer service agent that resolves tickets end to end, including issuing refunds or updating orders: £5,000 to £25,000 to build with guardrails, £150 to £800 a month to run. If you want to understand exactly what these can and can’t do without a human stepping in, I’ve covered how AI agents handle customer service without human input and where the limits are.
- Lead qualification or sales development agent: £300 to £15,000 to build depending on complexity, £40 to £400 a month to run, based on my own experience above.
- Internal ops agent (scheduling, reporting, data cleanup across systems): £3,000 to £40,000 to build, £100 to £1,000 a month to run, because it usually needs to talk to several tools at once.
- Multi-agent systems coordinating several of the above together: £40,000+ to build, £500 to £5,000+ a month to run, and only worth it once you’ve proven the simpler versions first.
Build vs buy vs hire someone: how to decide
If your business has never used an AI agent before, don’t start with a custom build. Start with a no-code platform, prove the use case with one narrow job, watch the real running cost for two months, then decide whether it’s worth investing more. I’ve seen too many businesses skip straight to a £15,000 custom build for a job a £20 a month Zapier flow could have handled.
If you’ve already tried the no-code route and it’s not enough, or you don’t have the internal time to build and maintain it yourself, that’s when bringing in outside help earns its cost. A good AI consultant will tell you honestly whether your use case even needs a custom agent, or whether you’re overcomplicating a task that a simpler tool already solves. If you’re weighing that decision, it’s worth understanding what an AI consultant costs before you commit to a build, because the right five hours of advice upfront regularly saves five figures on a build you didn’t need in that form.
My rule of thumb after building agents for my own business and watching clients build them for theirs: budget three times whatever the initial build quote says for the first year, because that’s roughly what the running cost, the fixes, and your own oversight time will add up to. If the number still makes sense at three times the quote, build it. If it only makes sense at the headline price, it probably won’t survive contact with real use.
Frequently asked questions
How much does a simple AI agent cost per month to run?
A simple, well-built agent running on a no-code platform typically costs £30 to £150 a month in API and subscription fees. Poorly built agents with unchecked reasoning loops can run into hundreds of pounds a month on the same task, so the build quality matters more than the platform you pick.
Is it cheaper to build an AI agent myself or hire someone?
Building it yourself on a no-code tool costs mostly your time and £0 to £300 a month, and is the right starting point for most small businesses. Hiring a freelancer runs £300 to £1,500 for a simple agent or £6,000 upwards for a custom one, and is worth it once your needs go beyond templates or you don’t have the hours to spare.
Why did my AI agent cost more than I expected?
Almost always because it made more model calls than you assumed. Agents that check their own work, retry failed steps, or call multiple tools per task can use 10 to 15 times more API calls than a single simple query, which is the single biggest reason first-month bills come in higher than the advertised per-query price.
What’s the biggest hidden cost of running an AI agent?
Human oversight time. Even a fully automated agent needs someone checking its decisions regularly, especially in the first few months, and that time has a real cost whether or not you’re invoicing yourself for it.
