AI Adds Helper Layer to Enterprise Software
Mike Knoop, co-founder of Zapier, on Zapier's LLM-powered future
This is why AI will add a helper layer to enterprise software, not replace the underlying screens. In products like Salesforce, users spend all day moving through records, forms, filters, and dashboards where a click is faster than typing a full request. Chat is strongest at turning a vague goal into the first draft of a workflow, but mature systems still need structured screens so thousands of reps can do repeat work quickly and without retraining.
-
Zapier is building for chat as an entry point, not as the whole product. Its Natural Language Actions API lets a model turn plain English into app parameters, but Zapier still adds previews, overrides, and human approval so users can control risky actions like sending email or posting to Slack.
-
Salesforce style software changes slowly because enterprises optimize for change management. Large customers often want new features exposed behind admin controls instead of pushing a new default interface to every seller at once, because moving buttons or flows can disrupt daily work across huge teams.
-
The deeper pattern is that complex systems need simpler surfaces for non experts, but those surfaces are usually purpose built, not just one text box. Airtable ran into the same issue, where raw database power was flexible for builders but harder for everyone else to use without clearer forms, views, and app specific entry points.
The next step is software that starts with natural language, then quickly turns that intent into generated forms, widgets, and task specific screens. The winners will keep the speed of point and click for expert users while using AI to remove setup friction, build custom interfaces faster, and make complex systems easier to approach.