Orchestration

What question does this solve?

At any moment a demand may be ready for execution, waiting for an input, holding an output that needs evaluating, or blocked on a decision. Someone or something has to recognize what can happen next and route the work or the decision, and has to recognize just as clearly when nothing can happen yet, and why.

Definition and role

Orchestration is the coordination of the flow: it relates the Delivery State to the conditions of the Stage Contracts, to dependencies and to the commitments in force, in order to recognize the next permitted step and route the work or decision it needs. The next step can be to execute, evaluate, clarify, wait or close. There is no obligation to keep productive execution running when essential conditions are missing.

Orchestrator

The Orchestrator is the function that carries out this coordination: it coordinates states, resources, evaluations and transitions, and it may plan and replan within its authority. It need not be a single component. A person, automated mechanisms or a distributed composition can perform it, as long as the coherence of decisions is preserved.

The function does not gather every authority. Routing an authorization is not granting it. Coordinating an evaluation does not mean deciding its acceptance without the competence and delegation required. Being the coordinator does not, on its own, confer every power.

Flow coordination

Flow coordination is the articulation of execution, evaluations, dependencies, returns and decisions. The model assigns this function to the Orchestrator. It must preserve the meaning of each change in the flow, rather than record only a linear passage from one stage to the next.

Why it matters for the hypothesis

By relating outputs, recognizing pending items and preparing the next decision, AI can reduce coordination work that would otherwise be repeated at every hand-over. This matters most when the buyer chooses small advances: if every choice requires reconstructing context and gathering specialists again, the cost of coordinating can make the offer impossible. Usable orchestration links the economics of production to the freedom to continue, adjust or stop. Maintenance and error recovery are part of the cost of that gain.

How it works

Information needed to coordinate

Coordination needs to recognize the demand, the conditions in force and the relevant events. Having every document is not enough if it is unclear which information still applies. The Delivery State gives the current view; contracts give inputs and criteria; policies bound execution; decision records give commitments. Orchestration relates these references without replacing them.

Information can be incomplete or out of date. If an authorization is in doubt, the coordinator must seek confirmation before carrying out work that depends on it. The absence of confirmation is not implicit permission. In a small demand one clearly recorded decision may be enough; in parallel flows it is necessary to know which parts depend on which conditions.

A coordination cycle

The cycle organizes the questions the function asks; it is not a rigid sequence.

Moment	Operational question
Recognize the situation	What happened, and which conditions are in force?
Check continuity	Is there work or a decision possible within the authorized scope?
Route	Who or what can take the next step?
Follow up	Was the routing received, and what result or obstacle arose?
Relate the result	Which evaluation, decision or state update does the event support?

New information can interrupt the cycle. A change to a business rule can alter the conditions before routed work finishes, and coordination must deal with that effect rather than assume the old plan still holds. An automated sequence of actions does not show that every necessary decision was taken.

Routing and responsibility

Routing selects eligible resources and sends them work under concrete conditions. Orchestration identifies when routing is needed and uses its result to continue. The boundary keeps two things apart: a resource being free, and the demand being ready for execution. A resource can be available while authorization is still pending. The routing must make clear what was assigned, for what purpose and under which conditions; a vague task hands the executor decisions that may be beyond its authority. Responsibility does not end when work is sent: an unconfirmed receipt, a constraint or an output all need handling.

Dependencies and parallelism

Demands and stages can share dependencies: a business decision can affect several implementations; an environment can be needed for several evaluations. Parallel work is possible when conditions allow, but it must not create an appearance of independence between outputs that have to fit together. Two resources changing the same artifact can produce incompatible results; the model does not prescribe a technique, but it requires that responsibility for reconciling outputs be identifiable. A blocked dependency may stop only part of the route.

Confirmation, interruption and resuming

Coordination can lose sight of an execution: a system stops responding, or a person finishes without the result being recorded. Before repeating an action, what actually happened must be established. This matters most when execution has effects beyond a draft, such as deploying a change, modifying an environment or consuming extra resources. Asking an activity to stop does not show that it stopped; the situation must distinguish the intention to cancel, the confirmation, and effects that have already occurred. To resume, the Orchestrator checks inputs, authorizations and available results, without assuming that all context stayed valid during the pause. Retry and Escalation take these decisions further.

Evaluation and the decision to continue

An available output must be related to the right evaluation. The Quality Gate decides technical acceptance within its scope, and coordination works out the consequences for the flow. An inconclusive evaluation may need more evidence; an inadequate output may need correction; a mistaken input may need a return to another stage. These differences should not all be routed as a new attempt at the same execution. The Commercial Gate can condition the next step even after technical acceptance.

Planning and replanning within the mandate

Coordination can form a route, compare alternatives and replan from results. An agent can do this, not only suggest actions to a person. Its authority determines which choices it can put into effect: routing, resuming, technical decisions and economic decisions do not necessarily have the same scope. An investigation might show that prototype and integration analysis should proceed in parallel, or that more evidence is needed before the offer to build; the coordinator can organize that within the discovery commitment instead of preserving an initial plan that no longer serves the purpose. The decision to continue can itself be automated when there is a corresponding mandate and an admissible offer; see Mandates. When information or authority is missing, coordination can seek context, route the question, suspend or close. The destination can be another system or a person. The possibility of intervention and of reconstructing decisions must be preserved, but human intervention is not a mandatory step in every demand.

In the running example

The barbershop authorizes discovery. Coordination routes the clarification of the rules and organizes prototype and technical investigation when their dependencies allow. An incompatibility with the calendar sends the route back to designing an alternative, within the scope already assumed. The accepted outputs feed the offer to build. The customer can choose booking only, keeping reminders as an uncontracted alternative; with valid delegation, the economic mechanism can make that decision automatically and coordination starts using the new commitment.

During implementation, an evaluation finds a conflict between times. The flow returns to correction with the observed behaviour and the criterion in force. After re-evaluation, coordination checks operational authorization and deployment conditions. If an execution loses communication, resuming consults the state, the outputs and the confirmed effects before planning another attempt. The customer follows progress, relevant pending items and the conditions of the commitment, without reconstructing these operational decisions.

Boundaries and failure modes

Activity is not progress. Performance should not be judged by the number of routings or transitions. What matters is whether coordination helps authorized demands reach accepted outputs, without hiding waiting, rework or pending decisions.

Coordination can become the bottleneck. If it needs constant intervention for routine conditions, its implementation can limit capacity. How best to distribute authority and keep coherence in complex flows remains open, and neither centralization nor distribution is assumed to be better in general. The model requires no central agent, workflow product or specific technology.

Connected concepts

• Engine overview
• Delivery State, the situation coordination works from
• Routing, which turns a routing need into an assignment
• Quality Gate and Governance
• Glossary