Just recorded a webinar on Zapier's new AI Workflow Index. A finding that got a lot of reaction: among our leading adopters, only 18% of the steps in an AI workflow actually run on AI. The other 82% is pedestrian. Rules, logic, app connections, moving data from one place to the next. This is old fashioned automation. When we modeled it, this approach cost 71% less to run than routing every step through a model. Hundreds of companies across our panel are basically converging on the same idea: your AI architecture is your strategy. In the webinar, I go deep with Rebecca Hinds, PhD and Ryan Anderson on what this approach actually looks like inside workflows. And how controls have to change once AI stops drafting messages and starts making decisions. Where's this wrong? If you've built AI into real workflows, I'd love to hear where selective beats all-in and where it doesn't. Recording here: https://lnkd.in/gbyRUYGa
The four jobs AI actually does at work: A first look inside Zapier's AI Workflow Index
https://www.youtube.com/
The 82% deterministic path is cheap until it hits an edge case nobody coded for. Then the incident and the model bill land the same day
Great work here, Andre!
imagine paying a premium for pedestrian work... good insights!
"this approach cost 71% less" I'm in 🤝
Loved the conversation Andre Vanier and learning more about your impactful work
82% rules and logic tracks with what I've seen building agents - the AI part is usually a thin layer on top of a lot of plumbing.