Ideas

A Run Produces Data. A Loop Produces Decisions.

The signature thesis: the data you scrape should plan the next scrape. Loops beat runs.

Most automation people build is a run. You point a scraper at a source, it collects, it writes to a table, it stops. Tomorrow it runs again against the same list, because the list is an input somebody typed once.

A loop is different in one specific way: the output of the run changes what the next run does.

The difference is where the decision lives

In a run, the decision lives with a person. Somebody looks at the results, forms a view, and edits the config. The system produces data; the human produces decisions. That works until the volume of data outruns the person, which happens almost immediately.

In a loop, the decision is part of the system. What was collected is scored, the score reprioritises the queue, and the next pass goes somewhere the last pass earned. Nobody edits a config. The list is an output, not an input.

Why this is the whole thesis

Every durable advantage I have built came from closing that gap. A prospecting system that reads which messages got replies and reweights the segment. A content pipeline where the pieces that performed shape the next brief. An enrichment stage that decides which records deserve the expensive API call based on what the cheap one returned.

None of those are clever models. They are ordinary automations with the feedback edge wired back in.

What it costs you

Loops are harder to build than runs, and it is worth being honest about why.

• You need somewhere for state to live. A run is stateless and therefore easy. A loop has memory, which means a schema, which means migrations.
• You need a scoring function you can defend. The moment the system reprioritises on its own, the priority is an opinion in code. It has to be one you can explain to whoever is affected by it.
• You need to be able to watch it. A run that goes wrong produces bad data. A loop that goes wrong compounds — it will happily spend a week going further in the wrong direction.

The test

If you want to know whether you have a run or a loop, ask what happens if nobody logs in for a month. A run produces thirty days of identical output. A loop produces something you did not specify — which is either the point, or the reason to go and look at it.

That is the bet: build the second kind, and accept that watching it is now part of the job.

Keep reading