AI Initiative logoAI Initiative

Work architected at Fifth9

Maestro: coordinating specialist intelligence

Maestro approached coordination through specialist roles, workflow structure and continuing session context. The later Maestro Platform broadened the engineering foundation; Conduir carried forward implementation and architectural work from that history.

Earlier platform work This explains an earlier coordination architecture and its lineage. It does not identify Maestro, Conduir and AI Engineering Workforce as interchangeable products.

The design question

How can specialist intelligence carry an idea through coordinated work?

Organise the different kinds of work needed to turn an idea into a reviewed result.

Specialisation helps organise a task, but each role still needs a clear responsibility and usable context. A handover should carry the decision, its reason and the next question; passing only the latest output can leave the next contributor to recreate the intent.

Coordination with a continuing context

A work brief reaches specialist roles through coordination. Their contributions return with handover context, and a reviewer relates the combined result to the original request.

Earlier Maestro context: requester, coordination design, specialist participants, session context and result reviewer exchange briefs, contributions, handovers and review decisions.
Retained context diagram of the earlier Maestro coordination design at Fifth9. It explains this lineage rather than attributing every later Conduir capability to the earlier system. Open the full-size diagram ↗
A view of the responsibilities

Preserve intent through specialist handovers

  1. 01Frame the requestAn understood request
  2. 02Assign responsibilitiesUseful roles
  3. 03Coordinate contributionsDependencies and handovers
  4. 04Carry context forwardResumable context
  5. 05Examine the resultAn examined result
Across the flow

Keep context, responsibility and the boundary between a proposal and an accepted result visible.

This explains an earlier coordination architecture and its lineage. It does not identify Maestro, Conduir and AI Engineering Workforce as interchangeable products. Select a step to read its walkthrough.

What each part is responsible for

The building blocks

Task brief

Describe the work and its intended result.

Receives
Requirements and constraints.
Produces
A shared starting point for the workflow.

Specialist roles

Give different activities distinct responsibilities.

Receives
The brief and role-relevant context.
Produces
Analysis, designs, implementation or review contributions.

Workflow coordination

Organise dependencies and parallel contributions.

Receives
Tasks, dependencies and available contributors.
Produces
An ordered or parallel work plan and handovers.

Session context

Keep useful task state available when work continues.

Receives
Intermediate work, decisions and session progress.
Produces
Context for the next contributor or resumed session.

Review and result

Bring outputs back to the question the work was meant to answer.

Receives
Deliverables and review findings.
Produces
A reviewed result or a clear request for more work.

Following one piece of work

How the flow fits together

01Frame the request

Make the request understandable as work with an intended outcome. Requirements that remain uncertain should be carried explicitly into later stages rather than silently replaced by an assistant’s assumptions.

02Assign responsibilities

Select specialist responsibilities appropriate to the task. Role names organise contributions, but the resulting work still needs to be assessed for the specific question it is answering.

03Coordinate contributions

Arrange compatible contributions in the needed sequence or parallel structure. Dependencies matter: a later stage must be able to recognise when required context or decisions are still missing.

04Carry context forward

Preserve decisions, unresolved questions and the latest work for handovers or resumption. Resuming a session should restore the meaning of the activity, not merely a list of files or earlier messages.

05Examine the result

Bring testing and review of the combined output back to the original request. The output may need revision, further information or a narrower claim; a chain of completed role steps does not itself establish a useful result.

Illustrative example · not a benchmark result

Create a first application version

A person wants a first version of a small application.

  • Requirements inform design; design informs implementation; testing and review examine the result. A resumed session needs the earlier decisions as well as the latest files.
  • If a requirement is unclear, later roles need that uncertainty carried forward. A polished design should not silently substitute its own assumption for the person’s intention.

The useful result: A contribution that can be reviewed and resumed with its purpose still intact.

A generic explanation of the workflow architecture, not a promise of unattended application delivery.

Why these responsibilities are separate

Resumable work also needs to distinguish a previous plan from completed action. The earlier engine and broader platform contributed foundations to later work; their history does not make every modern Conduir capability part of one unchanged product.

Reading the architecture accurately

Scope and status

What the work covers

  • Specialist personas, parallel and hierarchical workflow structures, and resumable sessions.
  • A wider platform phase supplying implementation later carried into Conduir.

What still needs to be established

  • Does coordination improve the combined result?
  • What context is essential for another contributor to continue?
  • Where should human judgement interrupt the workflow?

Where the design must hold up

Boundaries and evaluation

BoundaryWhy it mattersHow to examine it
Role competenceA persona name does not establish expertise on a task.Give a role a bounded task and examine the usefulness of its contribution.
Usable handoversA successful handover requires usable context, not just a completion message.Resume an interrupted workflow and check whether its decisions and unresolved questions survive.
Historical scopeEarlier platform work and present integration state are different claims.Identify which capability belongs to the earlier engine, wider platform or subsequent work before making a combined claim.