The simplest possible surface over the most powerful possible engine.
Our agents can run an almost unlimited diversity of workflows. That's the strength — and the trap. If every workflow earns its own buttons, panels and screens, the product collapses under its own optionality. We can't win by adding surface area. We win by removing it.
Not a menu of things you could do. The single thing that needs your judgment, right now. Everything else runs underneath and stays out of the way until it's ready for you.
A named, repeatable play. It runs on a trigger, pauses at a gate when it needs you, and leaves a receipt of what it did. The same shape wraps a metrics review, a competitor watch, or a feature ship — so the UI never has to grow to fit them.
The gate is where judgment lives. Everything to the left of it is automatic; everything to the right is trust.
A Skill is an invisible capability pack; a Routine is the named, gated play it powers; a Watch is a routine left running; Today is the one thing it surfaces.
A wall of everything the agent could do. Optionality the user has to triage.
When nothing needs you, it says so. Your agent is watching.
Today is where you live. The other four exist so Today can stay a single decision. Playbook is yours to shape — Watches run quietly; Explore is the conversation where your agent suggests and re-ranks, and you correct it. Activity is the proof — the canonical log of what each routine did and why it earned your attention.
Next: pressure-test the home model, wire the gate behind the existing approval flow, and ship the first three routines — metrics review, competitive watch, feature ship — to a small internal cohort. The prototype is ready to click through today.