For the first eight months of TwoPlus, we had a beautiful dashboard. Agent activity feeds, productivity charts, org-chart diagrams, budget gauges. Investors loved it. Customers politely ignored it.
What our customers actually used was a single, ugly list of items waiting for approval. We called it the review queue. By 2025 it was 70% of all time-in-app. So we did the obvious thing: we deleted the dashboard and made the queue the whole app.
The bet
The bet is that the fundamental unit of work in a hybrid team is a human deciding whether an agent's output is good enough to ship. Everything else — hiring agents, writing specs, setting budgets — is scaffolding for that one decision, repeated thousands of times per week.
If that's true, the best product is the one that makes that decision as fast and confident as possible. Not the one with the most features. Not the one with the prettiest dashboard.
How we got here
Three moments forced the rebuild:
- The Helia observation. We watched their CEO use TwoPlus for an hour. He never opened the dashboard. He lived in the queue, triaged 40 items, closed the tab.
- The power-user hack. Our best customers had built keyboard shortcuts around our UI to accept/reject items without ever leaving the keyboard. Their hack was better than our product.
- The churn analysis.Customers who did >50 reviews in week one retained at 94%. Customers who stayed in the dashboard retained at 31%.
Three design calls
One item at a time, full-bleed
The old queue was a list. The new queue is a stage — one item at a time, taking up most of the screen, with every piece of context (the spec, the agent's reasoning, the diff, the target) visible without scrolling. Keyboard arrows move between items.
This was the most controversial call internally. Product managers love lists. Users prefer stages. Users won.
Three actions, no nesting
Approve (A), Request Changes (R), Reject (X). That's it. If you need a comment, you type it after the action — not before. This inverts the usual pattern (open modal → write comment → submit) but it's 3x faster and the comments turn out to be better because they're written afteryou've already made the decision.
Coaching is a side effect of reviewing
Every rejection or change-request becomes a coaching note automatically, scoped to that agent. You're teaching every time you review. This is the loop we bet the company on: review → coach → agent improves → review gets faster → do more of them.
What we'd do differently
Two things we got wrong:
- We killed the dashboard too fast.Some customers (mostly execs) wanted the high-level roll-up. We brought a read-only version back as a page, not the home screen. Should've done that day one.
- We overestimated keyboard adoption. We assumed everyone would learn A/R/X. Half our users never do. We added a toolbar and mouse-friendly buttons in v2. Ship both — let users find their own level.
The meta-lesson: the thing your best customers do 100x more than average is the product. Not the thing investors ask about. Not the thing you put in your keynote. Watch the usage data, and be willing to throw out the rest.
