AI agent development costs about $8,000 to $22,000 for a single-purpose agent, $22,000 to $70,000 for an agent that uses several tools and takes several steps, and $70,000 to $180,000 or more for a platform of agents working together, according to one development firm's published tiers. Agents built for ecommerce work, such as answering order-status questions, handling returns and recommending products, are published at $30,000 to $200,000 to build. On top of the build, a brand pays every month to run the agent: model fees, hosting, monitoring and upkeep. This page sets out published build prices, what an agent costs to run, three agents an ecommerce brand might build first with a first-year cost for each, and when to buy a ready-made tool instead. Brand Media Works builds custom AI tools for ecommerce brands; the prices below are what other firms publish about their own work.
How much does it cost to build an AI agent?
The table sets out published build prices from firms in the Brand Media Works competitor register. The firm that works mainly with ecommerce brands is listed first, followed by firms that publish separate figures for retail and ecommerce agents. Each firm is described by the kind of work it does.
| Firm type | Published build price | Published running cost |
|---|---|---|
| AI development studio with a retail and ecommerce practice (US and India) | $30,000 fixed price for a scoped build, six weeks to production. A $5,000 discovery phase is credited against the build. | Not published |
| AI development and managed AI services firm (US) | Retail and ecommerce agents for product recommendation, customer service and inventory: $50,000 to $200,000. A fixed-scope pilot: $25,000 to $50,000 over four to six weeks. | Not published |
| AI and software development firm (India, with offices in North America and Europe) | Customer support and ecommerce agents for ticket deflection, order status, returns and product discovery: $40,000 to $150,000 or more. | $3,200 to $13,000 a month for a production agent serving real users |
| Offshore software and AI development firm (India) | Retail and ecommerce agents: $30,000 to $100,000. An MVP agent: $25,000 to $60,000 over six to ten weeks. | Not published for retail |
| Software development firm (Central and Eastern Europe, with US and UK offices) | Customer support agent: $40,000 to $120,000. A prototype: $10,000 to $30,000 over four to six weeks. | $2,500 to $5,000 a month for a customer support agent |
| Offshore software development firm (India, with US, UK and Canada offices) | Single-purpose agent: $8,000 to $22,000. Multi-tool, multi-step agent: $22,000 to $70,000. Multi-agent platform: $70,000 to $180,000 or more. Most production projects: $30,000 to $60,000. | $500 to $2,000, $2,000 to $8,000 and $8,000 to $40,000 a month for the three tiers |
| Software development firm (India, serving US, UK and EU clients) | First production agent on one workflow: $15,000 to $50,000 over two to three months. An agent that acts across several systems: $50,000 to $100,000. | $5,000 to $10,000 a month for ongoing engineering |
| Software and AI development firm (India, with a US company) | Proof of concept on one workflow: from about $3,000. Production agent: $12,000 to $40,000 over six to twelve weeks. | From $500 a month, plus model fees |
| B2B marketing agency with an AI agent practice | $15,000 to $45,000 to implement | $3,000 to $8,000 a month for optimization and monitoring |
| AI product development studio (Ukraine and Portugal) | Support agent: $20,000 to $50,000 for a basic system, $50,000 to $120,000 or more with CRM integrations and analytics. | $0.01 to $0.05 a conversation in AI usage |
The range is wide for three reasons. The first is how many systems the agent has to read from and act in: an agent that answers questions from a help center is cheaper than one that looks up an order in Shopify, issues a refund and updates the helpdesk. Integration work, not the AI model, is usually the largest part of the build. The second is how much the agent may do on its own before a person checks it. The third is where the team sits: one firm puts US senior AI engineers at $150 to $250 an hour, nearshore teams at $60 to $100 and offshore teams at $30 to $70.
How much does an AI agent cost to run each month?
Once an agent is live, a brand pays for four things every month:
- Model fees. What the AI model provider charges for the text the agent reads and writes, billed per token.
- Hosting and data. The servers the agent runs on and the store that holds the documents and product data it searches.
- Monitoring and evaluation. Checking answers, catching errors and testing the agent again after every change to prompts or models.
- Maintenance. Updating prompts, switching models and fixing integrations when Shopify, the helpdesk or another system changes. Two firms put this at 15 to 25% of the build cost a year.
Model fees are usually the smallest of these. One firm plans an agent task as five model calls, each reading about 6,000 tokens of instructions, context and tool results and writing about 400 tokens. That is 30 million input tokens and 2 million output tokens per 1,000 tasks. At OpenAI's published standard rates, read 20 October 2026, the model fees come to:
| OpenAI model | Price per 1M input / output tokens | Per 1,000 tasks | At 10,000 tasks a month |
|---|---|---|---|
| gpt-6-luna | $0.10 / $0.50 | $4 | $40 |
| gpt-6.1-sol | $2.00 / $10.00 | $80 | $800 |
| gpt-6-astra | $10.00 / $50.00 | $400 | $4,000 |
The model is the biggest lever on this line: the same workload costs 100 times more on the most expensive model than on the cheapest. Sending routine questions to a small model and only hard ones to a large model keeps fees down. Published all-in running budgets are higher than model fees alone because they include hosting, monitoring and people: from $500 a month for a simple agent at one firm to $3,200 to $13,000 a month for a production agent serving real users at another.
What would an ecommerce brand's first agent cost?
The table costs three agents an ecommerce brand might build first. Build prices are taken from the published ranges above, model fees use the mid-priced OpenAI model and the five-call assumption, and maintenance is 15 to 25% of the build. Monthly volumes are our assumptions for a mid-sized DTC brand; hosting and monitoring tools are not included.
| Agent | What it does | Build | Model fees | Maintenance a year | First year |
|---|---|---|---|---|---|
| Support ticket agent | Reads each new ticket, tags and routes it, and drafts a reply from the help center and order data for an agent to send | $40,000, the low end of two firms' published ranges for a customer support agent | 10,000 tickets a month: $800 a month, $9,600 a year | $6,000 to $10,000 | $55,600 to $59,600 |
| Order status and returns agent | Answers where-is-my-order questions, checks return eligibility against policy and starts the return in the store and returns tools, handing exceptions to a person | $50,000, the low end of one firm's published range for an agent that acts across several systems | 5,000 conversations a month: $400 a month, $4,800 a year | $7,500 to $12,500 | $62,300 to $67,300 |
| Catalog agent | Writes and checks product titles, descriptions and attributes for new listings against brand rules, for a merchandiser to approve | $20,000, within published single-purpose agent ranges | 2,000 listings a month: $160 a month, $1,920 a year | $3,000 to $5,000 | $24,920 to $26,920 |
Two things stand out. The build is most of the first-year cost for all three, so scoping the first version tightly matters more than the model bill. And an agent that only reads and drafts, like the catalog agent, costs far less than one that takes actions in the store, because every action needs its own integration, rules and testing.
Should an ecommerce brand build an agent or buy one?
Ready-made agent products are published at $20 to $150 per user a month with little setup, according to one firm. The same firm puts custom agent logic on a managed platform at $5,000 to $20,000 and a custom production system at $12,000 to $40,000. Another firm publishes $5,000 to $20,000 for agents built on no-code platforms, $20,000 to $50,000 on low-code frameworks and $60,000 to $250,000 or more for custom development.
- Buy when the job is standard and lives inside one tool, such as answering common questions from the help center inside the helpdesk.
- Build when the agent has to act across the brand's own systems, such as the store, the returns tool, the warehouse and the ERP, or follow rules no product supports, or when customer and order data has to stay in the brand's own cloud account.
- Combine the two when a ready-made tool covers most of the job and a small custom piece connects it to the brand's data.
Gartner predicts that over 40% of agentic AI projects will be canceled by the end of 2027 because of escalating costs, unclear business value or inadequate risk controls. Gartner also warns of "agent washing", where existing chatbots and automation tools are relabeled as agents, and estimates that only about 130 of the thousands of agentic AI vendors are real. Both are reasons to ask what a product or a quote actually does before signing.
What should a brand ask before paying for an AI agent?
- Is there a paid pilot first? A fixed-scope pilot on one workflow, ideally with the discovery fee credited against the full build.
- Does the quote separate the four running costs? Model fees, hosting, monitoring and maintenance, each with an estimate at the brand's own volumes.
- Which model, and can it change? The agent should be able to switch models, and route simple tasks to cheaper ones, without a rebuild.
- When does a person take over? Clear rules for what the agent may do alone and what it hands to the team, with the conversation history attached.
- Who owns it? The code, prompts and evaluation tests should belong to the brand, so the agent can move in-house or to another partner.
- How will success be measured? One number agreed before launch, such as tickets resolved without a person, response time or hours saved, compared with a baseline.
For most ecommerce brands the best first agent handles one high-volume job that already has written rules, such as ticket routing or order-status questions. Custom AI tools sets out how Brand Media Works scopes and builds this work. For agents that run fixed workflows rather than open conversations, see how much an AI automation agency charges, and if customer data has to stay in the brand's own account, see private AI for business. Published prices for every kind of AI consulting work are in AI consultant cost in 2026.
Frequently asked questions
How much does an AI agent cost per month?
Published running costs for a single agent range from about $500 a month for a simple agent to $3,200 to $13,000 a month for a production agent serving real users. One firm publishes $8,000 to $40,000 a month for a multi-agent platform. Model fees alone can be under $50 a month at 10,000 tasks on the cheapest of OpenAI's three GPT-6 models.
How long does it take to build an AI agent?
One firm puts a prototype at four to six weeks, an MVP at six to ten weeks, a simple agent at eight to twelve weeks and a complex agent at twelve to twenty weeks. An ecommerce-focused studio quotes six weeks to production for a scoped build.
Is it cheaper to build an AI agent in-house?
Usually not for a first agent. One AI development firm puts a small in-house AI team at $400,000 or more a year in salaries, benefits and infrastructure before anything ships. An in-house team starts to make sense once a brand runs several agents and has a steady stream of new ones to build.