Onboarding engineers with live agent sessions
Published July 28, 2026 · by the AQ team
The fastest way to learn how a team really works with AI agents is to watch it happen. In AQ, a new engineer opens the team's workspace list on day one and sees the actual work in flight: which agents are running, what they were asked, how seniors phrase prompts, and what the code looks like while it is being changed. Onboarding becomes observation of real work instead of a stale wiki tour.
Two ramps happen at once for a new hire in 2026: learning the codebase, and learning the team's agent practice. Both are tacit knowledge that documentation chronically trails. AQ makes both visible because the work itself is visible.
The workspace list is the map
Instead of asking "what is everyone working on," the new engineer reads it: each workspace is a named unit of work with its branch, its terminal, and its state. Opening any team-visible workspace is allowed by construction, so exploring is encouraged rather than an interruption. The connected Linear board stays in sync, so the issue context is one click away.
Prompting is learned by reading scrollback
Effective agent prompting is the skill gap between a new hire and a senior, and it is normally invisible: prompts live in private terminals. In AQ the terminal scrollback of every open workspace is a worked example: how the senior scoped the task, what context they gave, how they corrected the agent when it drifted. Reading three real sessions teaches more than any prompting guide.
First contribution without a local setup
A new engineer's first task can start in a workspace before their laptop is fully configured: the agent CLI, the worktree, and the dev server all run server-side, and the editor and preview are in the browser. Signing into the agent CLIs with their own accounts is a one-time step in Settings. The classic week-one wall (getting the app running locally) stops gating their first PR.
Help is ambient
When the new engineer's own agent session goes sideways, they do not have to compose a perfect question in Slack. They share the workspace, a teammate opens the same live session, sees exactly what happened, and can be granted the keyboard with the new engineer's one-click approval. The mechanics are covered in debugging an agent session together; as an onboarding property, it means being stuck is cheap and asking for help is low-friction.
Frequently asked questions
Can a new engineer watch a session without affecting it?
Yes. Opening a workspace attaches them to the live session as a viewer of the same terminal; the agent continues unaffected unless someone deliberately prompts or interrupts it.
Do new hires need a full local dev setup first?
No. Workspaces run server-side with an editor and live preview in the browser, so a first contribution can happen before the laptop is fully configured. They connect the agent CLIs with their own accounts once, in Settings.
How do juniors learn effective agent prompting?
By reading real sessions. Terminal scrollback in team-visible workspaces shows exactly how experienced engineers scope tasks, provide context, and correct agents mid-run, which is the material prompting guides can only approximate.
Can we limit what a new person sees during their first weeks?
Yes. Sensitive work can live in private workspaces shared with specific people, while the rest of the team's workspaces stay visible. Visibility is explicit and owner-managed.