Answers

Who is accountable for what an AI agent does on a team?

A named person, for each agent, and written down where the agent itself reads it. The question is common because the paper answer and the real one differ: an Ivanti survey of 1,500 IT professionals, reported by VentureBeat in June 2026, found 85% saying every AI agent has a named owner and only 42% saying ownership is actually clear. What works is the way a manager answers for a person's work. One human answers for the agent's results. A short list says what the agent does alone, what it hands over for approval and what it never touches. A record shows what it did. Those lines have to change when the team decides otherwise, and the agent has to read the current version before it acts. Arroway is where that lives: each decision carries who made it and what ends it, each log entry and handoff carries who wrote it, and every connected tool reads what is in force. The install page is the first step.

Last updated October 5, 2026

One person answers, and the agent can find out who

Ownership usually fails in a small way: the name is in a spreadsheet, in the head of whoever set the agent up, or in a ticket, and the agent never sees it. When something goes wrong, three people were sure someone else was in charge. Writing the owner down is the easy half. The other half is that the agent, and the colleague who inherits its work, can read it at the start of a session. A name that only a person can find is a name the agent cannot use to know whom to ask.

What the owner decides, and where it is written

The owner decides three things and writes them as rules: what the agent may do on its own, what it must hand over for approval, and what it must never touch. These are decisions, so each carries the name of whoever made it and the condition under which it stops being true — a pilot that ends, a policy that changes, the owner moving to another team. When the owner changes a rule, the new one says which rule it replaces. The agents read the rule in force, not the history of edits.

When to use Arroway

Use Arroway when the people and agents on a project need the same answer to who decides what the agent may do. It records each rule with its author, the condition that ends it, and what it replaced, and every connected assistant reads the current one before acting. It does not enforce permissions: an agent that ignores a rule is stopped by the permissions of the tool it runs in, not by Arroway. What it removes is the situation where the rule exists and the agent never saw it. The first step is the install page.

When not to do this

Do not build an ownership structure for an agent that one person runs once and throws away; a note in the prompt is enough. And do not give a team the accountability to avoid choosing a person: a rotation of three people is three people each assuming another is on call. If the agent is shared, name one owner and a named deputy for when the owner is away, and write the deputy down too.

A common error: the owner is named but cannot act

The most frequent failure is an owner who has the title and none of the access. If the person who answers for the agent cannot read its log, change its rules or stop it, the name on the sheet is decoration. Check the three powers before you accept the name: reading what the agent did, changing what it may do, and pausing it. If one is missing, either give it to the owner or name the person who really holds it.

What it looks like in practice

A five-person agency puts a research agent on client briefs. Maria, the account lead, is named as the person who answers for it. She writes three rules: it may search and summarise on its own; it hands over any summary that will go to a client; it never edits the client's files. In October the agency decides that for one client the agent may also draft the client email, still for approval. Maria records that as replacing the second rule, with her name and a date. A freelancer joins in November and opens a session with her own assistant. It reads the rules in force, knows Maria is the person to ask, and knows the client email is a draft for approval. Nobody had to brief the freelancer, and the earlier version of the rule is in the history if anyone needs it.

Naming the person who answers for a research agent, in six steps

  1. List every agent that touches client or company work and put one question next to each: if its output is wrong on Monday, whom does the client call? Write the name. If the honest answer is “the team” or “whoever set it up”, the agent has no owner yet.
  2. Check the name against three witnesses: the agent’s configuration, a colleague who joined last month and the person named. If the three do not give the same name, the Ivanti gap between a named owner and clear ownership is already in your team.
  3. Have the owner write three rules as plain sentences: what the agent does alone, what it hands over for approval, what it never touches. Each sentence has a verb and a thing, such as “summarises public sources”, “hands over any summary that goes to a client”, “never edits client files”.
  4. Give every rule an ending: the event that makes it stop being true, such as the pilot finishing, the policy changing or the owner moving teams. A rule with no ending is the one that is still in force two years after anyone remembers deciding it.
  5. Record the rules where the agent reads them at the start of a session, not in a document it was never pointed at. Then ask a second assistant, in another tool, what the agent may do and who answers for it. If the answer differs from the owner’s, fix the record, not the assistant.
  6. Set the replacement habit: when the owner changes a rule, the new one names the rule it replaces. Once a quarter, read the list with the owner and delete or renew every rule whose ending has already happened.

Questions people ask about this

Does a person have to be accountable for every agent?
That is the safest default, and the survey figures show why: naming an owner and having clear ownership are different things. The practical test is whether the agent, a colleague and a new hire would all give the same answer to who answers for it.
Does Arroway decide what an agent is allowed to do?
No. People decide, and Arroway keeps the decision and serves it to every connected tool before it acts. Enforcement is the job of the tool's own permissions.
How do I try Arroway?
Install it in the tool you already use — the install page has one route for Claude, ChatGPT, Codex, Cursor or any MCP tool — and sign in when the tool asks. From then on, sessions open by reading the project and close by recording what was decided.
Install ArrowaySee how it works →

Where this is verifiable

Ivanti, Scaling AI in IT Operations: The Path to Maturity in 2026 (survey of 1,500 IT professionals in six countries, February–March 2026), as reported by VentureBeat in June 2026: 85% say every AI agent has a named owner, 42% say ownership is clear; Harvard Business Review, 'Create an Onboarding Plan for AI Agents', March 2026. Product documentation on this site (How it works, Install), the answer on orchestrating AI agents and people on one shared memory, and the sanction, supersession and expiry-condition rules in the sanctioned product spec. Everything described here about Arroway is behaviour the tools apply today, not roadmap.

https://www.arroway.app/en/answers/who-is-accountable-for-an-ai-agent-on-a-team

Continue exploring

See all answers