Key Takeaways

  • The Grok Bot team built and shipped the product in seven weeks, spending the final two weeks actively removing experimental UI and feature panels.
  • Stopping at 90% task completion fails to reduce mental load, because the user still has to monitor and intervene during execution.
  • Software teams default to announcing what a product "has" (buttons, dropdowns, integrations) instead of expanding what an agent "can do."
  • Visual automation builders create unnecessary friction; Grok Bot replaced UI controls with plain natural language instructions.

The 90% Trap and the No-Look Pass

When Roman Ugarte and his team built Grok Bot in seven weeks at SpaceXAI, they faced a standard product temptation. They had built experimental features, settings panels, chain-of-thought streams, and automation controls to test internally.

Two weeks before launch, they did the opposite of what most teams do. They unshipped the interface.

As Ugarte explained: “We unshipped a lot. I wish I could have shown you what things looked like maybe two weeks out from launch where I think we'd realized that the core team, we had a lot of experimental features that we wanted to get internal feedback on.”

Instead of shipping complex scaffolding, they stripped user-facing controls so the product felt like a competent coworker rather than a complicated software suite.

Founders often celebrate reaching 90% task completion with AI. But as host Lenny Rachitsky noted during the conversation: “An AI that does 100% of the job feels categorically different from one that gets you 90% there.”

A 90% completion rate creates a psychological cliff rather than a minor shortfall. When an AI gets 90% of a task done, the user retains all the mental overhead. You must watch the output, anticipate failure modes, and prepare to step in.

As Ugarte put it: “When you delegate something to someone and you're like, 'I know I'm going to have to be thinking about this while you're doing it and I know it probably is not going to be quite there and I'm going to have to intervene'... That's not 90% task completion. You're still doing the thing and it feels that way.”

True delegation feels like throwing a no-look pass to a teammate. You hand off the work, turn around, and trust that it gets executed. Every extra toggle, preview pane, and confirmation button reminds the user that the system cannot be trusted alone.

Replacing "Has" with "Can Do"

Traditional software products win by accumulating visible surface area. Product managers ship new drop-down menus, configuration pages, and integration directories. Ugarte argues this habit ruins AI agents.

“There is this old school software tendency to say things like 'Grok Bot now has'—and when you think of completing that sentence it would be like a new button to press or it'd be a new drop down or it'd be a new integration that you can press plus and add,” Ugarte said. “Instead, reframe it as 'Grok Bot can now,' which is a much more human way of describing these capabilities.”

Take automations. Most enterprise tools force users through visual logic builders with triggers, conditional branches, and manual schedule selectors. Grok Bot eliminated that entire UI layer.

“You should just define automations in natural language,” Ugarte explained. “You should tell your bot, 'remind me that at 8 a.m. every day, please.' And then it should just do it. And you should never ever have to see that interface of creating an automation.”

What to Do With This

Audit your product roadmap for the upcoming sprint. Identify every UI element designed as scaffolding, such as manual confirmation modals, trigger selectors, or setting sliders, and ask whether the model can handle that step in the background. Pick one multi-step configuration flow in your app and replace the entire visual builder with a single plain-text instruction field.