Personalized AI infrastructure
Superintelligence Company Org Chart: Roles Beyond the Technical Lead
Who should own an AI agent's daily operation, memory, permissions and workflow? Five named roles, a fill-in plan and clear boundaries for each.

A superintelligence company org chart needs more than an engineer who can connect the agent to a model. Around every agent, a company should name five roles: an executive sponsor who owns the outcome, an operator who runs the agent day to day, a memory owner who decides what it may know, a permission reviewer who decides what it may do, and a process owner who keeps the work it touches accurate. Each role can be held by someone already on staff, but none of them should be left blank.
Key takeaways
- The technical lead builds and maintains the agent, but operations, data, risk and process roles decide whether it stays useful and safe.
- Name one operator who owns daily behavior, with a backup and a clear rule for when the agent must be paused.
- Keep memory ownership and permission review in different hands so no single person both fills the agent's knowledge and approves its access.
- Give the process owner authority over the workflow the agent serves, not only over the software that runs it.
- Review every role on a fixed calendar and write down who made each decision.
Why a technical lead alone cannot run an agent
The technical lead answers one question: can the agent perform the task? That work is necessary. It covers the integration with the company's tools, the prompts and evaluation scripts, the logging and the service uptime. It does not answer the questions that appear once the agent is in use. Who decides which price sheet is current? Who approves a refund the agent proposes above a certain amount? Who notices that the supplier process the agent depends on changed last month?
Those decisions belong to people closer to the business. The NIST AI Risk Management Framework treats AI governance as a cross-functional responsibility that runs through design, deployment and operation, not as a task for one technical team. A company that gives the engineer every decision creates a single point of failure, and it also makes the agent's behavior hard for managers to question. The org chart is the place to fix that before the first incident, not after it.
The operator: the person who runs the agent every day
The operator is the person who knows what the agent is doing this week. They read the summary of its activity, handle the cases it escalates, decide when to narrow its scope and tell the technical lead when behavior has changed. In most small and mid-sized companies, the operator is an operations leader or a senior coordinator, not an engineer. What matters is judgment about exceptions in the business, because most problems with agents appear as edge cases that the prompt did not anticipate.
Give the operator three written tools: a one-page description of the agent's job, a list of situations where it must stop and ask a human, and a contact for the technical lead when something behaves oddly. A company that skips the written stop rules usually discovers them during an incident. The operator should also have a named backup, so that the agent is never left without someone accountable during vacations or busy weeks.
The memory owner: who decides what the agent knows
Most agents answer questions using documents, notes and records that the company supplies, either through retrieval from a document store or through stored context about customers and cases. The memory owner decides which sources are allowed, which versions are current, how long notes are kept and which information must never enter the store. Retrieval-based designs are only as reliable as the documents they search, so an outdated policy PDF can produce a confident wrong answer that looks well sourced.
In practice, the memory owner approves each new source before it is connected, retires documents that have been replaced, and reviews what the agent has written into its own notes. This role also handles personal data. Anything about customers, employees or patients needs a retention rule and a lawful reason for being stored, and in Brazil that means checking the requirements that the national data protection authority enforces. The memory owner is often the person who already manages the company's records, such as an office manager, a finance controller or a practice manager.
The permission reviewer: who checks what the agent can do
The permission reviewer keeps the agent's authority narrow and visible. This person looks at every tool the agent can call, the scope of each connection, the thresholds at which a human must approve an action, and who granted each permission and when. The principle is least privilege: the agent should hold only what the current job requires, and expanding access should be a deliberate decision with a record.
Keep this role separate from the memory owner. The reason is practical. If one person decides what the agent reads and also decides what it may change, nobody else sees the full picture. Security guidance on authorization makes a similar point: access decisions should be enforced by the systems the agent calls, not only described in instructions to the model. An agent that reads untrusted text, such as customer emails or web pages, can be steered by instructions hidden in that text, so the reviewer should assume that some inputs will try to misuse the agent's tools. For a step-by-step method, see how to run a permission review for your AI agents.
The process owner and the executive sponsor
The process owner owns the workflow the agent participates in: order handling, quote follow-up, appointment reminders or incident triage. They define what a good output looks like in that workflow, decide when the process itself should change, and approve changes to the steps the agent performs. Many agent failures are process failures in disguise, where a step nobody wrote down was assumed by the agent and then skipped. The process owner is usually the head of the department that feels the work most directly.
The executive sponsor owns the outcome and resolves conflicts between the other roles. When the operator wants a higher approval threshold and the permission reviewer wants a lower one, the sponsor decides and records the reason. The sponsor should be a single person with budget authority and the time to read a monthly summary. For a charter that defines how these decisions are made, see a committee charter for small companies.
How to fill the chart in four steps
The following plan works for a first agent with one job. Complete it before any further integration work begins.
- Write one sentence describing the agent's job and one measurable success criterion that the operator can check each week.
- Assign each of the five roles to a named person, and name a backup for the operator and the permission reviewer.
- For each role, write what it may approve or change alone and what it must escalate to the sponsor or to the technical lead.
- Schedule the first memory review and the first permission review, and record who signed off on each.
Illustrative scenario: a small wholesale distributor with one operations manager wants an agent to draft reorder suggestions from sales history and supplier price sheets. The operations manager becomes the operator and the purchasing lead becomes the process owner, since the purchasing lead decides which products are reordered. The owner's office becomes the memory owner, because it controls which price sheets are current. The finance lead becomes the permission reviewer and notices that the agent can submit purchase orders above a limit that nobody intended, so the limit is lowered and the change is logged. The owner acts as sponsor and approves the decision. No one in this example needed to write code to make the chart useful.
Be Human's managed approach puts these roles into the operating agreement from the first engagement. The first engagement defines one useful role, the memory and tools it needs, its approval boundaries and an observable success criterion. Managed operations can then include monitoring, recurring evaluations, permission review, incident response and controlled improvements, so each role has a defined task on a known schedule instead of an informal expectation. Changes to the agent's scope go through a written request, so the people in these roles see each decision before it takes effect.
If you are not sure which of these roles your company already has, the free Be Human AI diagnostic can help you find the first task worth handing to an agent.
Frequently asked questions
Do we need to hire new people for these roles? Usually not. Most companies already have someone who fits each role, such as an operations leader for the operator, a controller or office manager for the memory owner, and a department head for the process owner. The real work is making the assignment explicit and writing down the boundaries. A new hire becomes necessary only when the agent's workload grows beyond what current staff can review.
Can one person hold more than one role? Yes, in a very small company one person may hold the operator and process owner roles, or the sponsor and permission reviewer roles. Be cautious about combining the memory owner and the permission reviewer, because that combination removes the independent check on what the agent can read and change. If you must combine roles, schedule an outside review of the permissions at least once a quarter.
Who should be the operator if we have no technical staff? The operator does not need to write code. They need to understand the business process well enough to recognize when the agent's answer is wrong, and to follow written stop rules. Pair the operator with an external implementation partner that handles integration, monitoring and incident response, and keep the operator as the person who decides when the agent is paused or changed.
How often should these roles be reviewed? Review the memory and permissions on a fixed schedule, typically every quarter for a first agent, and review the operator's stop rules and the sponsor's success criterion at the same time. Review sooner after any incident, any change to the underlying tools, or any new data source. Record the outcome of each review in one page so that the next person in the role can see what was decided and why.
Your next step
See where an AI agent fits in your company first.
Take the free Be Human AI diagnostic. A few questions about your work show which task to hand to an agent first, what it needs to know and where a person approves.
Be Human editorial guidance. Examples are illustrative unless explicitly attributed. Sources provide background; implementation choices depend on your workflow.


