How much does an AI agent cost in 2026? Build and running costs explained

The short answer: three ways to get an AI agent

Price guides from development agencies usually describe one route only: hiring a team to write an agent from scratch. Their ranges are real for that route, with simple custom agents starting around $20,000 and multi-agent systems running into six figures. But most small and mid-sized businesses have two cheaper routes, and the right choice depends on what the agent needs to do, not on how advanced you want it to sound.

If you are still deciding whether you need an agent at all or a simpler automation, read our guide to agentic AI vs AI automation first. A fixed workflow is often cheaper to build and run than an agent that decides its own steps.

• Platform agent: switch on the AI built into your CRM or helpdesk. Monthly fee, little or no build cost, limited to what the platform supports
• Low-code agent: build on n8n or a similar tool, connected to a model such as GPT or Claude and to your own systems. Moderate build cost, low running cost, much more flexible
• Custom-coded agent: a development team writes and hosts it. Highest build and maintenance cost, full control over data, logic and interface

Option 1: Use the agent built into your platform

If your sales and support already run through a platform with built-in AI, this is the cheapest way to start. GoHighLevel's AI Employee is a good example: the Growth plan costs $50 a month per sub-account and includes 1,000 Conversation AI responses and 100 Voice AI minutes, and the Unlimited plan costs $97 a month per sub-account with fair-use limits. Phone, SMS and email charges are billed separately on top. Our GoHighLevel AI Employee guide covers every plan and what each one includes.

The build cost here is mostly configuration: writing the agent's instructions, loading your FAQs and services, connecting the calendar, and testing conversations before going live. That is days of work, not weeks. The trade-off is that you can only do what the platform allows, so an agent that needs to check stock in your own system or follow unusual business rules may outgrow it.

Option 2: Build a low-code agent on n8n

For most businesses that need an agent to work across several tools, a low-code build is the best value. n8n gives you an AI agent node that connects a language model to tools such as your CRM, calendar, email and database, with memory and fallbacks you control. Our tutorial on how to build an AI agent with n8n walks through a real lead qualification agent step by step.

n8n's cloud plans start at €20 a month billed annually for 2,500 workflow executions, and the Pro plan is €50 a month for 10,000 executions. The self-hosted Community Edition is free to use, so you only pay for the server it runs on. Build time for one focused agent, such as qualifying inbound leads or answering order questions, is typically days to a few weeks depending on how many systems it touches and how much testing it needs. If you are choosing between tools, our comparison of n8n vs Zapier vs Make explains why we usually pick n8n for agents.

Option 3: Commission a custom-coded agent

A custom agent written in code makes sense when you need things platforms cannot give you: strict data residency, deep integration with an in-house system, a branded interface inside your own product, or several agents working together. Development agencies typically quote around $20,000 to $80,000 for a simple custom agent and $100,000 or more for advanced and multi-agent systems, with build times from two to five months.

Before going down this route, check whether a low-code build would cover 80% of the need at a fraction of the cost. Our guide to custom AI solutions: build vs buy gives a simple test for when custom work is worth it and when it is not.

What an AI agent costs to run each month

The biggest running cost for most agents is the language model, which is billed per token, roughly three quarters of a word. Every time the agent replies, it sends its instructions, the relevant knowledge and the conversation so far as input tokens, and gets its reply back as output tokens. That is why long conversations cost more per message than short ones.

Here is a worked example. Say a support conversation has 8 back-and-forth turns, each sending about 4,000 input tokens and getting about 200 output tokens back. That is 32,000 input and 1,600 output tokens per conversation. On a small model such as GPT-5 mini, which OpenAI lists at $0.25 per million input tokens and $2 per million output tokens, that comes to about one cent per conversation, or roughly $11 for 1,000 conversations a month. A flagship model can cost ten times as much or more for the same traffic, so match the model to the job and test whether a smaller one handles it well.

• Model usage: per token, often the smallest line item for text agents on small models
• Platform or hosting: the plan fee for your CRM's AI, n8n cloud, or a server
• Voice minutes: for phone agents, about $0.06 to $0.31 per minute depending on the provider, voice and model; see our guide to AI voice agents for missed calls
• Phone numbers, SMS and email: charged by your telephony or messaging provider, usually on top of any AI plan

The running costs people forget

The usage bill is easy to estimate. The costs that catch people out are the ones that do not appear on an invoice, and they matter more than the token price for most small deployments.

• Review time: someone should read a sample of conversations every week, especially in the first month
• Knowledge upkeep: prices, policies and services change, and the agent only knows what you give it
• Model changes: providers retire and replace models, so prompts need retesting when you switch
• Monitoring and fallbacks: alerts when the agent fails, and a clear handover to a person when it is unsure
• Storage for long documents: a vector database or knowledge base if the agent searches large amounts of content

What pushes the build price up

Two agents that look the same in a demo can differ in cost by ten times. The difference is almost always in how many systems the agent touches and how much can go wrong when it acts.

• Number of integrations: each CRM, calendar, payment or database connection adds build and testing time
• Actions vs answers: an agent that books, refunds or updates records needs more guardrails than one that only answers questions
• Channels: web chat is cheapest; voice, SMS and WhatsApp add setup, compliance and per-message costs
• Knowledge size: a page of FAQs is simple, while thousands of documents need search and retrieval
• Compliance: regulated data, consent rules for calls and audit logging all add work

How to keep AI agent costs down

Start with one narrow job that happens often and costs you money when it goes wrong, such as answering after-hours enquiries or qualifying new leads. A focused agent is cheaper to build, easier to test and quicker to prove its value than a general assistant that tries to do everything.

Use the smallest model that does the job well, keep instructions short, and send only the knowledge the agent needs for each question rather than everything you have. Cap conversation length and hand long or complex cases to a person. Our guide to AI chatbots for business websites shows how we scope a first agent so it pays for itself before it grows.

Is an AI agent worth the cost?

Judge it against the cost of the work it replaces or the revenue it saves, not against zero. If an agent answers enquiries that would otherwise wait until morning, the question is how many of those leads you currently lose and what each one is worth. A platform agent at $97 a month only needs to save one or two jobs a month to pay for itself in most service businesses.

If you would rather have it scoped, built and monitored for you, our AI chatbots and agents service covers platform setups, n8n builds and custom agents, and we will tell you on the first call which route fits your budget.

Frequently asked questions