Why this became a problem now
Every wave of technology added a participant to the work. First came people, paperwork and phone calls. Then came systems, which made parts of the process faster and more consistent. Next, automation connected those systems through APIs, events and queues. Now AI agents analyze, decide and act, too.
Each layer solved part of the problem and created another: the more participants, the more handoffs, and handoffs are exactly where work gets lost. Orchestration exists to coordinate those handoffs.
Orchestration isn't automation, integration or durable execution
These terms are often used interchangeably, but they solve different problems.
Task automation
Automates a single action: calling an API, sending a message, transforming data, triggering a job. It works well until the process has to wait for a person, handle a decision or survive a failure halfway through.
Integration
Connects systems so they can exchange data. It answers "how does one system talk to another?", but not "where is this case right now, and who needs to act?"
Durable execution
Makes sure long-running code completes despite crashes and restarts. The flow lives in code and is designed for developers.
Work orchestration
Coordinates the entire business process, with every participant. It uses automation, integrations and resilient execution as building blocks, and adds what they lack: a process that business and technical people read together, per-case state, clear ownership of every step, and intervention when things go off script.
The five elements of orchestration
- State. Every case knows which step it's on and what has already happened. A failure doesn't wipe out progress.
- Context. Data and earlier decisions travel with the case, so whoever takes the next step doesn't start from zero.
- Ownership. Every step has an owner (a person, a system or an AI agent) and a deadline.
- Exceptions. The process defines what to do when something fails: wait, retry, escalate or reroute.
- Intervention. When the plan isn't enough, your team acts on that specific case and resumes from there.
Example: a loan application
- A sales rep submits the application (person).
- The onboarding system validates the customer's data (system).
- An AI agent reviews the uploaded documents and summarizes the risks (AI).
- An approval-limit rule decides: below the threshold, approve automatically; above it, route to an analyst (process rule).
- The analyst approves with the full context of the case (person).
- An integration records the contract in the core system (system).
If a document is missing, a timer reminds the customer and escalates after the deadline. If the final integration goes down, that step is retried without redoing the review or the approval. At any moment, the operations team can see where every application stands.
Signs you need orchestration
- No one can quickly answer "where is this case?"
- The process relies on spreadsheets and email to know who should act next.
- Integration failures force people to redo work by hand.
- You want to use AI for real decisions, but you need limits and human review.
- Your process documentation and what actually runs no longer match.
How Kikwiflow implements orchestration
In Kikwiflow, the process is modeled visually in Craft, run by Engine with persistent state, and tracked in Monitor, where your team can intervene in any case. See the Kikwiflow overview and the use cases.
Frequently asked questions
What is the difference between orchestration and automation?
Automation runs a single action, such as calling an API or sending a message. Orchestration coordinates the whole process: it keeps the state of every case, defines who owns each step and what to do when something fails, including when the process has to wait for a person.
Is work orchestration the same as BPM?
They're related, but not the same. BPM is the discipline of business process management, usually tied to the BPMN standard. Work orchestration focuses on running and coordinating the process in real time across people, systems and AI agents, with per-case state and operator intervention. Kikwiflow uses its own JSON model, not BPMN.
Does orchestration replace system integrations?
No. Integrations stay in place and become steps in the process. Orchestration adds what they lack: retries, error handling and resuming from the exact point of failure, without redoing earlier work.
When is an orchestration layer worth adopting?
When nobody can quickly say where a case is, when the process relies on spreadsheets and email to know who should act, when integration failures force people to redo work, or when you want to use AI in real decisions with limits and human review.