Assign specialist agents separately. Hand off and review their files deliberately.
Assign research and writing separately; have a person check versions, pass the shared brief, and accept the result.
Buda gives each Agent a persistent workspace and provides Space Drive for shared files. A person assigns the next task and checks the handoff; Buda does not automatically orchestrate every Agent.
More agents do not automatically mean one coordinated result
The cost of a multi-agent process is the space between assignments: missing source links, incompatible versions, and decisions nobody owns.
A research Agent finds new information, but the drafting Agent works from an older prompt.
The final copy arrives without the source trail, assumptions, or outstanding questions.
The team mistakes an Agent completing its task for a person approving publication.
A workspace for responsibilities, not an invisible swarm
Use Agent sessions and files to inspect work, and deliberately place shared material in Space Drive. A person checks the version and access scope before assigning another Agent; this is not built-in automatic orchestration.
Specialists with boundaries
Separate research, writing, and checking roles. Each Agent has its own sessions and working files; verify access before sharing any Space resource.
Evidence beside the draft
Request a brief, source links, draft files, and unresolved questions in the task. Open the files to verify them; Buda does not automatically assemble a handoff packet.
Tools with an owner
Use Drive to keep artifacts, and Browser, Terminal, or Git where the task warrants inspection. A human decides whether the result is ready for an external action.
Illustrative workflow and file list—not Buda UI, a built-in queue, or automatic approval states.
Six handoffs in a customer-update workflow
These are separately assigned Agent tasks with human-led file handoffs, not an automatic orchestration or approval chain built into Buda.
Set the assignment
Name the audience, confirmed facts, deadline, claims to avoid, and publication owner.
Collect verifiable evidence
Gather primary links, dates, quotations, and open questions; flag any claim you could not check.
Transfer only approved context
Place an approved brief in Space Drive. Have a person share the path and version with the next Agent; private Agent Drives are not automatically shared.
Draft for the actual channel
Produce a reviewable customer update with citations or links back to the brief, plus a list of deliberate omissions.
Check claims before sending
A person tests the draft against the verified brief, returns corrections, and owns the publication decision.
Keep the next run cheaper
Store the accepted brief and checklist. Note where the handoff broke so the next run improves rather than merely repeats.
Pilot one handoff, then measure the rework
Do not start by creating a fleet. Use one customer update or internal briefing with two Agents and a person who owns acceptance.
1. Set the boundary
Pick a low-risk deliverable and an owner. List the sources Agents may read and the actions they may not take.
2. Run and inspect
Research Agent prepares evidence; writing Agent uses the approved brief. Inspect output files, source links, and unresolved claims.
3. Judge the result
Compare correction time, missing evidence, and the number of back-and-forth handoffs to the previous process.
4. Expand carefully
Add another specialist only if it removes a real bottleneck without obscuring accountability.
Human orchestration still matters
Buda makes sessions and artifacts visible, but default autopilot does not pause every tool call for approval. If work must be reviewed before sending, restrict tool access and use a human sending step or a verified external approval flow. Record-level review and merge belong to a separate system such as Busabase ChangeRequests.
External action → explicit authorization
Formal record write → audited review path
Multi-agent orchestration: common questions
Does Buda automatically coordinate all agents?
No. Assignments, context sharing, permissions, and acceptance criteria must be designed for the workflow.
Are Agent Drives automatically shared?
No. Distinguish individual Agent Drive from Space-level resources and confirm the access scope before a handoff.
How many agents should a team start with?
Use the smallest number that makes roles clearer: often two specialists plus one named human reviewer.
What is the output of the pilot?
A verified source brief, a reviewable draft, an unresolved-question list, and a recorded human decision.
Can Buda enforce approval before every tool call?
Do not assume so. Tool behavior varies; put consequential actions behind explicit permissions and process checks.
Is this different from an AI Agent Workspace page?
Yes. This page is about assigning and reviewing work across specialist Agents; the Workspace page explains the general product surface.
Start with one human-led file handoff
Assign research and writing separately; have a person check versions, pass the shared brief, and accept the result.
Start with Buda →