FOUNDATIONS
How it works
Once you have seen the shape, most no-code automations stop looking like software and start looking like a procedure. Every workflow here is built from the same four parts, in the same order. Tools differ; the shape does not.
The basic shape
Trigger
Something happens and the workflow starts: an email arrives, a form is submitted, a row is added, a file lands in a folder. Nothing runs until the trigger fires, so the first question is always what event starts this.
Rules or an AI step
The workflow decides what the event means. Fixed rules handle anything predictable, such as a subject containing "invoice". An AI step is added only for what rules cannot do, usually reading messy text or pulling fields out of a document.
Output
The result is written somewhere you can inspect: a row, a draft, a file, a task. A workflow whose only output is an email already sent has nothing to inspect, which is why these guides avoid that ending.
Human check
A person reads the output before it reaches anyone outside the business. This is not a training stage to be removed later. On anything that sends, deletes, pays, or approves, it is the step that makes the rest safe to run.
What an AI step is good at, and what it is not
Most disappointment with automation comes from asking a model to do a job it is badly suited to. The split is fairly clear.
| Good fit | Bad fit |
|---|---|
| Sorting a message into one of a few categories you define. | Deciding whether something is correct, compliant, or acceptable to send. |
| Extracting a name, date, total, or reference number from unstructured text. | Applying a policy that has exceptions and judgment calls inside it. |
| Summarizing a long thread into three lines a person can scan. | Guaranteeing that a number it extracted is the right number. |
| Drafting a reply that a person will read and edit before sending. | Sending that reply, or deciding who should receive it. |
A model is useful when its output is text a person reviews, and risky when its output is an action taken on the world. These guides keep model output on the text side of that line.
Why every guide ends with a human review step
A workflow fails in ways that stay invisible until someone looks: a label guessed wrong, a field left empty because the source changed, a duplicate row because the trigger fired twice. None of that announces itself. A review step turns a silent wrong answer into a visible one, and it costs about ten seconds per item.
- Sending: the message reaches a customer and cannot be recalled.
- Deleting: the record is gone, and the backup may be gone too.
- Paying: money leaves the account, and recovery is a support ticket.
- Approving: a decision is recorded against your name that you never read.
Automate the preparation, not the judgment. A workflow that drafts, sorts, extracts, and stages work for a person is worth more than one that acts alone and gets switched off after the first mistake.
The five categories covered here
Triage a shared inbox, draft replies, route attachments, flag what needs a person.
Spreadsheets
Keep a sheet current from incoming events, deduplicate rows, fill in skipped fields.
Forms
Move submissions into the place the work happens, and catch the incomplete ones.
Files
Rename, sort, extract from, and file documents that sit in a downloads folder.
Meetings
Turn notes and transcripts into actions, owners, and follow-ups a person confirms.
A short glossary
Trigger
The event that starts a workflow. Until it fires, nothing runs.
Action
One step a workflow performs: creating a row, sending a draft, moving a file.
Label
A category attached to an item, usually by a rule or an AI step, so later steps can branch on it. Labels are only as good as the list you define.
Structured output
A result returned in a fixed shape, such as named fields instead of a paragraph, so the next step can use each field on its own.
Rate limit
A cap on how many requests a service accepts in a period. Hitting it looks like a failure even when the workflow is correct, so guides say where a limit is likely.
OAuth and authorization
The permission screen where you let one tool act on another account. The tool receives a token instead of your password, and you can revoke it later in that account's settings.
Idempotency
The property that running the same step twice leaves the same result as running it once. It is what stops a retried trigger from creating a second row or a second charge.