Skip to content
// Glossary · Agent oversight

Agent fleet

Also called: agent fleet management, concurrent agent sessions

What is an agent fleet?

An agent fleet is all the coding agent sessions an organization has running at once, across people and machines, treated as one thing to watch and manage. Once several agents run concurrently, the interesting questions stop being about a session and start being about the set.

// In more depth

What changes when there is more than one.

The problems are not the single-session problems repeated. They are new.

01

Attention becomes the scarce resource

One blocked session is easy to notice. Six sessions where two are blocked, three are working and one has quietly failed is a triage problem, and triage needs a view of the set rather than six views of one.

02

Machines become a shared resource

Sessions land on specific devices. Which machines are connected, which are busy and who has access to them turn into questions worth answering at the fleet level.

03

Policy stops being per-session

Deciding permission modes session by session does not scale past a handful. What the organization allows becomes something to set once and apply.

// Why it matters

Why the fleet view is the useful one.

Because the thing you most need to know is which run needs you next.

Blocked work is invisible per-session

You only see a stalled run if you go looking at it. A fleet view brings the blocked ones to you.

Machines and sessions are related

Fleet questions are usually machine questions: what is connected, what is busy, who can reach it.

Patterns only show in aggregate

One agent asking the same permission repeatedly looks like nothing. Across a fleet it is a policy you should change.

// Commonly confused with

Commonly confused with.

Three things it is not.

Subagents in one sessionA single run can spawn several subagents. That is internal structure in one session, not a fleet of sessions.
A clusterA fleet is not necessarily shared infrastructure. It is usually sessions on individual developer machines, coordinated rather than co-located.
A job queueQueued work waits to start. A fleet is what is already running, and its problem is supervision rather than scheduling.
// FAQ

Agent fleet questions

What is an agent fleet?

All the coding agent sessions an organization has running at once, across people and machines, treated as one thing to observe and manage rather than as unrelated processes.

How is that different from subagents?

Subagents are structure inside one session. A fleet is many separate sessions, often on different machines belonging to different people.

What is the hard part about a fleet?

Attention. Knowing which of several concurrent runs is blocked, which is working, and which has quietly failed needs a view of the set, not a view of each one.

Does a fleet mean shared infrastructure?

Usually not. With Virex Remote the sessions run on individual machines you own; what is shared is the oversight, not the hardware.

Can an agent supervise a fleet?

That is the AI-driving-AI pattern: an orchestrating agent clearing routine decisions across runs and escalating the consequential ones to a person.

See every run in one place.

Virex Remote shows the sessions your organization has running and which of them need you.