The short answer
A custom AI agent connected to two or three of your systems costs $18,000 to $40,000 to build at Zeebrix and is live in four to eight weeks. A single automation workflow with an AI step is $12,000 to $30,000. Several agents running a whole process, such as quote to invoice, is $50,000 to $120,000. After launch, a small or mid-sized business usually spends $200 to $1,500 a month running an agent, including model usage and hosting, and the model usage is the smallest part of that. If you want us to look after it, support retainers start at $2,500 a month.
Those are the numbers. The rest of this guide explains what you are paying for, why other published estimates run far higher, and how to keep your own build at the lower end of the range.
Based in Australia? See our AI development prices in AUD.
Why published estimates range from $5,000 to $800,000
We read the cost guides ranking for this search in October 2026. The floor is around $5,000, which buys a template agent on a no-code platform configured by a freelancer. The ceiling passes $800,000 for an enterprise system with many agents, custom models and a compliance programme. Most mid-market guides cluster a “production” agent between $40,000 and $150,000, and the authors are not wrong. They are describing a bigger company than most of our clients.
Three things explain the gap. What counts as an agent (a chatbot with one tool, or a system that plans and acts across six). Who builds it, with US agency rates running $80 to $250 an hour. And what gets bundled in: data cleanup, security review, compliance work and change management can double a quote before any agent code is written.
Our figures assume an owner-led or mid-sized business, systems that already have APIs (your CRM, help desk, accounting and calendar almost certainly do), and one process at a time. That describes most of the agents we build.
AI agent cost breakdown: what you are paying for
The AI model is the cheapest ingredient. The expensive part is connecting the agent to your business safely and proving it works before it touches a customer. Here is how a $30,000 agent build breaks down.
- Discovery and process mapping, one week: $2,500 to $4,000. We watch how the job is done today and write down every exception
- Integrations, the biggest line: $3,000 to $8,000 per system. Each one needs authentication, error handling and a test account
- Agent logic: instructions, tools, memory and guardrails: $6,000 to $10,000
- Evaluation set and testing: 10% to 15% of the build. Thirty to fifty real examples the agent must handle correctly before launch
- Approval flow: the screen or message where a person says yes before anything is sent or paid: $2,000 to $4,000
- Deployment, logging and monitoring: $2,000 to $3,500
- Handover and training for your team: included at any serious agency
What affects the cost of an AI agent?
When two agents that sound identical are quoted $20,000 apart, one of these is usually the reason.
- Number of systems. One CRM is a $20,000 agent. CRM plus help desk plus accounting plus a phone system is a $40,000 agent
- Autonomy. An agent that drafts and a person approves is cheaper than one that acts alone, because full autonomy needs more guardrails, more testing and rollback paths
- Data quality. If your CRM has three spellings of every customer, someone has to fix that before the agent can use it, and it is usually us
- Compliance. Health, finance and legal work add security reviews, audit logs, data residency and sometimes a legal opinion
- Voice. A phone agent costs 30% to 60% more than the same agent in chat or email, because real-time speech is harder to get right and to test
How much does an AI agent cost to run each month?
Model usage is billed per token, and for the volumes most businesses see, it is small. On OpenAI’s API price list, checked in October 2026, GPT-6.1 Terra is $1 per million input tokens and $5 per million output tokens; the smaller GPT-6 Luna is $0.10 and $0.50; the flagship GPT-6 Astra is $10 and $50. Claude and Gemini models sit in the same bands.
Take an agent that handles 2,000 enquiries a month. Each one uses about 6,000 input tokens (your instructions, the retrieved policy documents and tool results) and 800 output tokens. That is 12 million input and 1.6 million output tokens a month. On Terra it costs $12 plus $8, so $20 a month. On the flagship model it is $200. A well-built agent routes easy steps to the cheap model and reserves the expensive one for judgment calls, which is how we cut one client’s model bill from $18,000 to about $6,500 a month.
What actually costs money after launch is everything around the model.
- Higher API tiers on your own software: some CRMs and phone systems charge more for API access
- A person reviewing the logs weekly for the first month, then monthly. Budget two to four hours a month
| Cost line | Typical monthly cost |
|---|---|
| Hosting for the agent and its database | $50 to $400 |
| Search or vector database, if the agent answers from your own documents | $0 to $300 |
| Logging and evaluation tooling | $0 to $200 |
| Model usage, 2,000 enquiries on a mid-tier model | About $20 |
| Model usage, 50,000 support tickets, depending on routing | $800 to $3,000 |
| Typical total for a small or mid-sized business | $200 to $1,500 |
Should you buy, use a platform or build a custom AI agent?
Not every agent needs to be built. Agent features inside tools you already pay for, such as your CRM, help desk or phone system, cost nothing extra to a few hundred dollars a month, and if your job lives inside one tool, try that first for a month.
Agent platforms such as Microsoft Copilot Studio, Salesforce Agentforce, n8n and Relevance AI sit in the middle. A freelancer can configure one for $500 to $5,000, you pay the platform monthly, and you accept that it works the platform’s way and keeps your data where the platform keeps it.
Custom makes sense when the job crosses systems no platform connects, when customer data has to stay in your own cloud account, when the process is what makes your business different, or when per-seat platform pricing across a growing team costs more than a one-off build. That is the $18,000 to $40,000 bracket. Our guide to AI agents for small business covers the jobs worth handing over first.
How to keep the build under $40,000
Every agent we have delivered inside this range had the same shape. Every one that went over broke at least two of these rules.
- One process, one owner on your side, one month to a working version
- Only connect systems that have APIs. Screen-scraping a legacy tool can add $10,000 on its own
- Start with the agent drafting and a person sending. Add autonomy once the logs are clean
- Use the model providers’ standard models. Fine-tuning adds cost and rarely helps a business agent
- Agree what “done” means before kickoff: a list of thirty real examples the agent must pass
- Insist on a fixed price for a fixed scope so the estimate risk sits with the builder
What a fair quote should include
If any of these are missing from a quote, they are not free. They are coming later as a change request.
- The list of systems being connected, by name
- The evaluation set: how many test cases, who writes them, what pass rate counts as launch-ready
- The approval and rollback design: what happens when the agent is wrong
- Logging and monitoring, and who gets the alert at 2am
- Who owns the code, prompts and data at the end. It should be you, from day one
- The first month of post-launch fixes
- An estimate of monthly running costs at your expected volume
Our numbers, side by side
Every Zeebrix project is quoted as a fixed scope and fixed price after a free discovery call. These are the bands almost every agent project lands in. Typical running costs for a small or mid-sized business are $200 to $1,500 a month, model usage included.
| What you are building | Build cost | Time to launch |
|---|---|---|
| Automation workflow with an AI step (one or two systems) | $12,000 to $30,000 | 4 to 6 weeks |
| Custom AI agent, two or three integrations, human approval | $18,000 to $40,000 | 4 to 8 weeks |
| Several agents running a whole process | $50,000 to $120,000 | 10 to 16 weeks |
| Monthly support retainer, with monitoring, model updates and improvements | From $2,500 a month | Ongoing |
Frequently asked questions
How much does a simple AI agent cost?
A simple agent that handles one job inside one system, with a person approving its actions, costs $12,000 to $20,000 to build. If the job lives entirely inside a tool you already pay for, the agent features built into that tool may cost nothing extra and are worth trying first.
How much does a customer service AI agent cost?
A customer service agent connected to your help desk and CRM, answering from your own documents and escalating to a person when unsure, is typically $20,000 to $40,000 to build. Running costs depend on volume: a few hundred tickets a month costs tens of dollars in model usage, while 50,000 tickets a month runs $800 to $3,000.
How much does it cost to run an AI agent per month?
For most small and mid-sized businesses, $200 to $1,500 a month covering hosting, model usage, search and monitoring. Model usage is usually the smallest line. An agent handling 2,000 enquiries a month on a mid-tier model costs around $20 in tokens.
Is it cheaper to just use ChatGPT or Claude agents?
For a single person doing their own work, yes, and you should. For a business process that touches customer data, needs approvals and has to connect to your CRM or accounting system, you need the agent running under your own accounts with your own permissions and logs, which is what a custom build gives you.
How long does it take to build an AI agent?
Four to eight weeks for a custom agent with two or three integrations, including a week of discovery and two weeks of running read-only alongside your team before it acts on its own. Multi-agent systems covering a whole process take 10 to 16 weeks.
How much does an AI agent cost in Australia?
About A$27,000 to A$60,000 before GST for a custom agent connected to two or three systems, and roughly A$300 to A$2,250 a month to run. We quote in US dollars ($18,000 to $40,000), so the AUD figures are approximate at about A$1.50 per US$1. Our Perth office works with Australian clients, and every project is a fixed price after a free discovery call.
Want a fixed price for your agent?
Tell us which systems the agent needs to work with. After a free discovery call with our team in San Jose or Perth you get a fixed-price quote for your AI agent.
Book a free discovery call
