What an AI Agent Build Actually Includes
A custom AI agent build is one job, wired into your real systems, that a person can still stop. Ours runs $25,000 to $60,000 fixed over 30 to 90 days and starts with a $7,500 audit credited toward the build. The box is the scoped workflow, the success metric, the kill threshold, shadow mode on your data, a supervised go-live, and ownership of the prompts and code. It is not a chatbot, a platform licence, or a promise that nobody will need to approve an irreversible action.
The question people type is “Which agencies build custom AI agents for marketing and operations?” The question they mean is “what do I actually get, and what does it cost?” Most vendor pages answer with a capability list. Here is the box.
What is in the first build
One bounded workflow. Not a fleet. The first one is the one that already eats senior hours and touches a system you can name.
You leave with these, in writing:
- The workflow, named. Intake, reporting, documents, research, or an internal knowledge job. We pick one. The ones you should not automate yet stay on a list so they do not sneak into scope.
- A success metric and the person who measures it. Hours given back, leads qualified to a rule, tickets resolved with a citation. If nobody owns the number, the project cannot fail, which means it also cannot succeed.
- A kill threshold. The point at which we tell you to switch it off. Agreed before any build. Gartner expects over 40% of agentic projects to be canceled by the end of 2027 for cost, unclear value and weak controls. Those are first-two-weeks problems.
- Shadow mode on your data. The agent proposes. It does not send, charge, or delete. You watch the disagreements. Accuracy is measured before time saved.
- A supervised go-live. The agent acts. A person approves anything irreversible. The approval rate tells you where to widen autonomy. Runs are logged and readable by you.
- Ownership. You own the prompts, the code, the configuration and the evaluation set, and you can export them. We settle that in the contract.
- The security answers in week zero. Data flow, subprocessors, a data processing agreement on request. We do not hold SOC 2 and we say so. Your reviewer should not discover that in week six.
- A decision at the end of the pilot. Expand, extend once with a specific fix, or stop. Workflows two and three are priced after the first one proves out.
The catalogue of the eleven systems we build most often, each with its own page, is at AI systems. The hire for the build line is AI agent development.
What it costs, in the same sentence as the box
Our pilot is $25,000 to $60,000 fixed, 30 to 90 days. It starts with a $7,500 AI Opportunity Audit over two weeks, credited toward the build if you proceed. Running the thing afterwards is a production retainer from $8,000 a month. Budget 15 to 25% of build cost per year for model use, monitoring and the day a connected tool changes its API.
A simple single-workflow automation at a small shop, no shadow mode, no kill rule, handed over, typically runs $3,000 to $8,000. If your first automation is internal, reversible, and nobody outside the company sees its output, that cheaper shop is the right buy and we will say so. Come back for a governed build when the workflow touches revenue, a customer, or a regulated record.
The full band table, including DIY and enterprise, is in how much AI agents cost in 2026.
What is not in the box
A chatbot on your homepage. A licence to a platform we do not operate. Unattended sends. An agent that pauses live ads. We will not build payment, refund, or bulk-export jobs that a stranger’s assistant can start. Should your business be agent-callable is the filter for that last line.
If you already know you want the operations layer (monitor, maintain, add one workflow a month) rather than a one-off build, that is AI workflow automation. If you have not decided what to build, start at AI consulting. The audit stands on its own.
A number only we can put on this page
We run the thing we sell. The internal fleet reclaims 50 or more hours a week across our own team, with a person approving anything that matters. That figure is our estimate, and the case study says so. The public agent surface at mcp.strataigize.com, the server that lets another company’s assistant query our business, listed eight tools when we fetched it on 13 September 2026. External clients on that channel in the week to 6 September 2026: zero. The build is real. The adoption number is honest.
We asked ChatGPT, Gemini and Perplexity “What do AI automation agencies charge for building custom agents?” and “Which agencies build custom AI agents for marketing and operations?” 44 times each in the 30 days to 12 September 2026. Zero of those answers named us. The Canada-qualified version of the same subject hits 34%. The pages were written as if the buyer already knew we were in Vancouver. This one is written for the buyer who does not.
Before you take a call, use 24 questions to ask an AI agent vendor. Apply them to us. Then book the audit if the box above is the box you want.
Talk to us