Hetchr was a bi-directional dashboard for developers and their managers — read GitHub, your project tools, and your DevOps stack on one screen, and let the dashboard act on them too. Merge a PR, ship a deploy, close a ticket, ack an alert. We learned the hardest way that the action layer is the easy part. Trusting the tool to act is what users actually struggle with.
Role
Founder & Product Designer
Products
B2B SaaS - Developer Tools
Dates
2020 - 2022
Building a developer agent in 2020 — and discovering the trust gap before "agent" was a category.
Hetchr was a bi-directional dashboard for developers and their managers — read GitHub, your project tools, and your DevOps stack on one screen, and let the dashboard act on them too. Merge a PR, ship a deploy, close a ticket, ack an alert. We learned the hardest way that the action layer is the easy part. Trusting the tool to act is what users actually struggle with.
One dashboard to see your day, one dashboard to do it.
Developers in 2019-2020 lived inside 8 to 12 tools just to ship code. GitHub for the source, Jira or Linear for the tickets, Datadog or Sentry for the alerts, the deploy pipeline somewhere else, Slack to argue about all of it. The cost wasn't just tab-switching — it was the mental reload that came every time you bounced between worlds, plus the time-on-task spent acting in five different UIs each day.
Hetchr's bet: aggregate the read layer, and add an action layer on top. We pitched it to three audiences — solo developers managing their own stack, small teams (2-10) needing a shared command center, and engineering managers wanting a 360° view of in-flight work.
Read everywhere, act anywhere — in four clicks.
Hetchr integrated GitHub for source control, Jira / Linear / Trello for project tracking, and the DevOps surfaces — deploy systems, alerting, monitoring — that punctuated a developer's day. The dashboard read all of them in one view, and it could act on them: merge an approved PR, trigger a deploy, close a Jira ticket, acknowledge a Datadog alert. The same actions you'd do in 4 different tabs, collapsed into 4 clicks in one tab.
We shipped to a public beta with hundreds of users. The reading layer was loved. The acting layer was avoided.
Convenience didn't overcome caution.
The blindspot wasn't in the engineering. The blindspot was in the assumption that convenience would overcome caution.
Users who happily told us in interviews "I'd love to merge PRs from one place" did not, in practice, merge PRs from Hetchr. They opened Hetchr, saw the PR, and switched to GitHub to merge it there. The same pattern repeated across every actionable surface — read in Hetchr, act in the source tool. We had inadvertently increased time-on-task instead of decreasing it.
The reason was simple, but we missed it: when an action carries real consequences — production deploys, irreversible merges, alerts that page someone — users want to perform it inside the tool they trust to own that action. A dashboard layered on top of GitHub is, by definition, not the tool that owns the merge. It's a translator. And users, when stakes are real, refuse to ship through a translator.
The pivot, in one sentence.
We pivoted out of Hetchr — and that pivot directly seeded NineFive, where we tried the inverse hypothesis: build a workspace that owned the customer relationship, and put the agent inside it.
The lesson that generalizes to every agent shipped since.
"Trust isn't a slider you turn up. It's a sequence: low-stakes action, high-stakes action, irreversible action."
What Hetchr exposed in 2020 is what every product team building agents in 2026 is now living through. You can ship an agent that's technically capable of doing X. That doesn't mean your users will let it do X. The action layer needs more than reliability — it needs to be designed to earn trust, progressively, on actions whose stakes match the level of trust already established. Most agentic products in 2026 still skip that sequence — and pay the same cost we paid in 2020.