People kept doing the same customer research every Monday. I spent ten months with our PM turning that habit into Automations: describe the work once, and Enterpret's agent does it on a schedule or when something happens.
- Role
- Lead product designer. Specs, interaction design, prototypes, the first coded POC and about 20 frontend PRs
- Team
- A PM, two frontend engineers and a designer on templates
- Timeline
- Dec 2025 to Sep 2026, three releases
- Outcome
- About 60% of active customers run at least one automation. Weekly runs grew about 14× from July to October. Setup completion went from 38% to 61% after the last release
TL;DR
- It started with a Monday report. People were running the same research every week, then cleaning it up by hand. In interviews, every complaint about our reports turned out to belong to one of four tangled decisions: context, analysis, presentation and delivery.
- Four names in ten months. Custom agents, scheduled reporting, scheduled tasks, automations. Each rename was us learning what the product actually was. I wrote the specs, designed it, and coded the first working prototype.
- An argument I lost. I proposed a wizard for event triggers. The team chose an inline form, and they were right about why.
- 15 minutes and 12 questions. That's how long the agent took to set up one automation when I tried it on a customer account. So I designed the agent's questions instead of the screen: a creation skill with evals that got setup under five minutes in our own tests.
- Where it landed. About 60% of active customers now run at least one automation. Weekly runs grew about 14× from July to October, and setup completion went from 38% to 61% after the last release.
Coming soon
I'm still writing up the full story: the discovery calls, the alternatives that didn't make it, the agent behaviour I designed, and the numbers that kept me honest. Check back soon.