Roles: business, IT, vendor, AI director
An agent project has four roles, and each has its own question.
1. The business, that is, the owner and the department head: which process loses money and how much. They also accept the agent’s work.
2. The IT department: access, security, connection to the CRM and the accounting system (in Russia often 1C). It makes sure the agent works and does not break the systems.
3. The vendor: builds the agent to the brief and is responsible for the metric if the contract states it.
4. The AI director: chooses which processes to give to agents, calculates the effect, sets the vendor’s task and checks the result.
A simple test: write four names next to each agent in the plan. Who chose the process, who gave access, who built it and who accepts the work by the number. An empty cell in the last column means the agent will launch, but nobody will say whether it paid off.
The first three roles exist in almost any company, but the fourth usually does not. Then the choice of task goes to whoever came first with a proposal, and nobody is responsible for the result in money.
McKinsey’s figures for November 2025 show where companies get stuck: 62% of organizations experiment with agents, but only 23% scale them. Moving from a trial to work in several departments needs a decision that neither IT nor the vendor makes.
Why choosing the task is not IT’s job
The IT department knows the systems, but it does not have to know where the company loses money. The question “which agent do we need” sounds technical, but the answer is financial: which process is expensive, how many leads are lost, who will accept the agent’s work.
When IT chooses the task, the agent works fine but is not needed: it runs, and the business numbers do not change. According to the MIT NANDA report of August 2025, 95% of generative AI pilots give no measurable effect on profit.
In June 2025 Gartner predicted that more than 40% of agentic AI projects will be canceled by the end of 2027 because of costs, unclear value and weak risk control. Unclear value is a question of choosing the task, not of code quality.
IT still has its own work that must not be taken away: give the agent access with minimal rights, check where client data goes, and make sure the agent does not break the CRM during an update. This work should be separated from choosing the task, not handed to someone else.
The vendor is not the one to choose either. It is responsible for building and offers what it can do, so it gets its task before any talk about the estimate.
Why you should not hand AI entirely to the IT department is covered in detail on the page about the IT department and AI.
Outsourced AI director
A mid-sized business usually does not need a full-time AI director: there are not enough processes for agents to fill a full working week. So the role is outsourced.
Our AI director costs from $999 a month and works 5 hours a week: a meeting with the CEO, a quarterly plan, a review of agent reports, estimates, and acceptance of vendors’ work.
The company still needs a responsible person inside, someone with the authority to change processes. The director works through this person and brings options with price and effect, and the decision stays with the owner.
An outsourced director is not tied to a vendor. For each agent, they compare a ready-made product, an assembly from open components and custom development, and choose by calculation. The rule that saves the most: the vendor’s estimate is checked by someone who earns nothing from it.
In a small business the roles are split even more simply: one responsible person and a director for a few hours. How this works is covered on the page about who is responsible for AI in a small business. What the director’s work includes and how they report is collected on the AI director page.
AI department in a large company
In a large company, one director is not enough: agents work in several departments, and each has its own rules. Then the company needs an AI department, the people and rules that answer where AI works and how much it brings.
It includes the AI director, department heads, IT, a lawyer and the security team. The department keeps the overall plan, the metrics of each agent and the rules on which data may go to models.
The department reports to the CEO, not to the IT director. This way decisions about money stay with the person responsible for them, and IT gets a clear task on connection and protection.
Do not confuse it with an AI team. By an AI team we mean a team of AI agents that does the work, while the department is the people who decide which agents are needed.
A department starts with one responsible person and one process, not with a new unit. The first task is to find processes where agents will return money. An AI roadmap for the whole company costs from $93,000 and takes 7 weeks.
The gap from the McKinsey study sits right here: 88% of organizations use AI, but only 39% see an effect on profit. Between these numbers stands the question of who in the company is responsible for the result.