From digital systems to automated workflows

For years we built software that made paper digital: a form instead of a letter, a grid instead of a ledger. That is still necessary. The question I now put on top of every discovery is different: once it is digital, which human moves should stay, and which should become an automated workflow? If you skip that, you get a pretty panel that repeats the same WhatsApp traffic with nicer type.

That is where my interest in AI sits — not in the fashion. I want a system that gathers information, pushes repetitive work, and calls a human with a prepared file for the hard call. Software that is only digital will feel old by 2028. Software that is automatic with no trail is dangerous.

Three layers, not one magic switch

Layer one: a visible path — stage, owner, due time. Layer two: simple rules — if this, then that. Layer three: an intelligent suggestion. Many projects want to start at layer three because it sounds better in a meeting. Then they discover there is no stage for intelligence to plug into.

Process automation is useful when exceptions have a home. A path with no exceptions looks clean in a pitch and lies in a real company. An undo and a reason for rejection are part of the design, not an admission of failure.

Where agents actually help

An agent can read a case, flag a missing document, draft a note. It cannot change company policy on its own. If you do not draw that in the product, you will return to the human-in-the-loop story — and you should. I put that in human-in-the-loop automation and in product shape in agent-native web systems.

For teams still on no-code or a template, this layer is often the ceiling. When your flow spills off the generic tool, take the custom decision seriously, or you will plant intelligence on shaky ground.

A drill before the next meeting

Draw one real job from request to done. Mark the boxes that repeat every week. Those are automation candidates. Mark the boxes that are expensive when wrong. Those are human-confirm candidates. That map beats any language-model slide.

If you have such a map and want it to become a product, send it in an order. Digital to automated is not one leap. It is a few slices that have to stay alive — the subject of after launch.