# What is orchestration, and what happens when a step in it fails?

Orchestration is the part of a system that decides what runs, in what order, and what happens when a step fails. It owns retries, timeouts, ordering, and the record of what already ran. Every system has one, whether it was designed or whether it emerged as a cron job calling a script that calls three more.

Also called: workflow engine, pipeline
Checked: 2026-09-27

In one line

What runs, in what order, and what happens when a step fails.

## Why it matters to you

The question worth asking about orchestration is where the state lives. If a run half completes, what does the system know? A system that records each step as it finishes can resume. One that does not has to start again, which is how a retry turns into a duplicate.

The craft retailer's outreach system splits this deliberately into two planes that never talk directly and share only a database. A reasoning session on one machine finds, scores, drafts and schedules. A deterministic service on a small server polls for jobs whose time has come, marks each job as it goes, and posts a confirmation per send. The intelligence and the irreversible action are in different processes on purpose.

That split is also what makes the failure behaviour describable. When the sending half has no judgement in it, there is nothing to debug about why it decided to send.

## Where it turns up in something we run

- Two planes sharing one database and nothing else: a reasoning session that drafts and schedules, and a small service with no browser or model that sends (recorded 2026-09-27)
- Failed profiles go to a reprocess queue instead of stopping the crawl, because rate limits and blocked requests are expected, not exceptional (recorded 2026-09-14, [WhoFits](https://octyn.co/casestudies/whofits))

## Next to this

- [Webhook](https://octyn.co/glossary/webhook)
- [AI agent](https://octyn.co/glossary/ai-agent)
- [Rate limiting](https://octyn.co/glossary/rate-limiting)

## Argued out at length

- [A glued workflow versus a built system at 10,000 runs a month](https://octyn.co/compare/glued-workflow-vs-built-system)
- [What it costs to change an AI workflow six months later](https://octyn.co/compare/cost-to-change-ai-workflow-six-months)

## Related

- [A glued workflow versus a built system at 10,000 runs a month](https://octyn.co/compare/glued-workflow-vs-built-system)
- [Webhook](https://octyn.co/glossary/webhook)
- [The Consult](https://octyn.co/consult)

Canonical: https://octyn.co/glossary/orchestration