what is agentic orchestration?

what the orchestrator is responsible for

Five jobs, and every real system has to answer all five whether or not it names them:

1. Decomposition. Turning "ship this feature" into tasks with an order and dependencies.
2. Assignment. Deciding which agent gets which task, based on role, tools, or availability.
3. State. Holding what has been done, what is in flight, and what everyone learned, so the fourth agent is not working from stale facts.
4. Conflict resolution. Two agents touching the same file, or disagreeing about the answer. Somebody has to decide.
5. Termination. Knowing when the goal is met, when to retry, and when to stop and ask a human.

Skipping the last one is the most common and most expensive omission in the category.

the three topologies

Hierarchical. A manager decomposes and delegates to specialists, who may delegate further. Predictable, easy to debug, and the shape almost all business deployments settle on.

Peer to peer. Agents negotiate task ownership directly, contract-net style. Flexible, and hard to reason about when something goes wrong at three in the morning.

Blackboard. Agents read and write a shared store and pick up work when the state suits them. Loosely coupled, and prone to two agents doing the same thing.

Most working systems are hierarchical over a shared store, because that combination gives you clear delegation and shared context without open negotiation.

where it actually breaks

Not in the reasoning. In the seams:

• Context loss at handoffs. Agent B gets a summary of agent A's work and misses the thing that mattered.
• Duplicate work. No shared state, so two agents solve the same subtask differently.
• Cascading failure. One agent's wrong output becomes another's input and the error compounds silently.
• Runaway cost. No global step budget, so a retry loop between two agents bills for an hour.
• No audit. Something wrong shipped, and reconstructing why takes longer than the work did.

Every one of those is a systems problem rather than a model problem, which is why better models do not fix them.

how this works in aldena

Aldena implements the hierarchical topology directly, as a workplace rather than a config file. The org chart is the orchestrator: you set who reports to whom, and a manager decomposes work and delegates down its own line without you writing routing code.

The seam failures above are addressed where they happen. Context loss is handled by shared room memory, 100 entries every agent reads plus 50 private each, so a handoff carries the facts rather than a summary. Cascading failure is contained by a reviewer role above the workers and an approval gate before anything irreversible. Blast radius is bounded by the room, which has its own server and its own credentials.

For the difference from ordinary workflow automation, see workflow versus agentic orchestration.