Why recurring workflows need a different kind of automation
Most automation tools are built around triggers: a form is submitted, a row is added, a message arrives. That model works well for reacting to events. A lot of valuable work does not start with an event though. It starts with a calendar. Someone checks competitor sites on Monday, refreshes a prospect list every morning, reconciles invoices on the first of the month and writes a customer health summary on Friday.
Recurring AI workflows in Crons start from the schedule. Each workflow knows what it found on the previous run, so it can report differences instead of repeating everything. It keeps a full run log, so you can compare this week with last week. And it can pause for a human before doing anything that leaves your company, which matters when nobody is watching the run happen.
The building blocks of an AI workflow
When you write a sentence like "Every weekday at 7 AM find new companies hiring sales reps in Texas, enrich them and add them to our HubSpot list after I approve", Crons builds a workflow from these step types. You can reorder, edit or add steps before saving.
| Step type | What it does | Plans |
|---|---|---|
| AI reasoning | Filters, classifies, summarizes, extracts or drafts text | All plans |
| Web fetch | Reads a public page or feed | All plans |
| Browser action | Clicks, scrolls and reads pages in a headless browser | All plans, counted in browser minutes |
| HTTP or API request | Calls an API with stored credentials | All plans |
| Integration action | Reads or writes Slack, Gmail, HubSpot, Google Sheets, Notion | All plans |
| Memory read and write | Stores and compares state between runs | All plans |
| Condition | Continues only when a rule is met | All plans |
| Approval gate | Pauses until a person approves | All plans |
| Code step | Runs sandboxed JavaScript or Python | Pro and Teams |
| Webhook out | Posts the result to any URL | All plans |
Browser steps and integrations have their own pages. See browser and web actions and integrations for details.
A worked example from a marketing team
Imagine a marketing lead at a 40 person SaaS company who wants a weekly content brief. Today she opens five competitor blogs, three industry newsletters and the company's own analytics sheet every Monday, then writes a summary for the team. Here is the same work as a recurring AI workflow.
- 1Schedule fires every Monday at 9 AM New York time.
- 2Web fetch steps read the five competitor blog feeds.
- 3Memory read loads the list of posts seen last week.
- 4AI reasoning keeps only new posts and groups them by topic.
- 5Integration action reads last week's top pages from a Google Sheet.
- 6AI reasoning drafts a one page brief with three suggested topics.
- 7Integration action creates the brief in Notion and posts a link to Slack.
- 8Memory write stores this week's post list for the next run.
Nothing in this workflow touches customers, so it needs no approval gate. If she later adds a step that emails the brief to an external agency, an approval gate goes in front of it. A ready version lives in the marketing use cases.
Reliability you can check
- Retries: each failed step gets three attempts with backoff before the run is marked failed.
- Alerts: failed runs notify you by email or Slack, so a broken source never goes unnoticed for weeks.
- Run log: every step, input, output, retry and timing is stored for 30 days on Starter, 90 days on Pro with CSV export, and one year on Teams.
- Dry run: test the full workflow without sending messages or writing to tools.
- Hard limits: runs and AI calls stop before a plan limit, with an alert at 80 percent.
Choosing an AI workflow builder for scheduled work
If you are comparing AI workflow builders, ask four questions. Can it start from a schedule rather than only from an event? Does it remember previous runs without you building a database? Can it stop for a human before sensitive actions? And can you read exactly what happened in a run three weeks ago? Crons is built around yes to all four. Event driven tools are a better fit when you mostly react to incoming messages; the Zapier Agents comparison covers that trade off in detail.
Frequently asked questions
What is the difference between AI workflow automation and a classic automation tool?
Classic automation moves data between apps with fixed rules. AI workflow automation adds steps that read unstructured content, make judgments and write text, so it can handle tasks like summarizing changes or qualifying leads that rules alone cannot.
Can a workflow run on an event instead of a schedule?
Yes. Every workflow can be started with Run now, and on Pro and Teams you can trigger it from an inbound webhook or the REST API in addition to its schedule.
How many steps can one workflow have?
There is no fixed step limit per workflow. What counts toward your plan is the number of runs, AI calls and browser minutes each run uses.
Can I write custom code inside a workflow?
Yes, on Pro and Teams. Code steps run sandboxed JavaScript or Python, which is useful for custom calculations or reshaping data between steps.
What happens if a source I read from changes its layout?
AI reasoning steps read content the way a person would, so small layout changes usually do not break the workflow. If a step does fail, it is retried three times and you get an alert with the error in the run log.
Who can edit workflows on a team?
Pro includes 5 seats with owner, editor and viewer roles. Teams includes 25 seats and approval rules per team, so different groups can approve different actions.