Designing for Trust Without an Existing Precedent for AI Features

AI Task Helper saved 5 minutes per task setup, measured post-launch. AI Timeline Wizard adoption rose from 5% to 25% after a fix to the launch pattern, and time to create a project timeline dropped 60%.

The problem

Project managers were writing descriptions for every task in a project from scratch, wasting hours on manual, tedious work that could be automated. User feedback and support-ticket signals about the task panel, routed through the product manager, surfaced grievances with manual task-description entry. Those indirect signals formed the basis of Pronto's first AI-powered feature.

Separately, project managers were building project timelines from scratch, or from a saved template, then manually adjusting timings, assigned users, and linked tasks. The blank canvas of a timeline grid was daunting: 72% of users abandoned their first timeline, citing it as too complicated with no clear starting point. The product manager pushed for an AI-powered assistant here too.

Discovery

Those indirect signals showed where the friction was. To confirm the cause, I ran user interviews and week-long diary studies tracking how project managers actually set up tasks and timelines. AI-generated synthetic users were used only to shape early hypotheses ahead of those sessions, as a supplement to real-user research, never a replacement for it.

On a 40-task project that is hours of writing before work even starts. My team spends more time writing tasks than doing them.

Project manager, task-setup diary study

Tried to start a new project timeline and gave up. Blank grid, no idea where to begin with phases, dependencies, or who to assign. Went back to a spreadsheet in the end.

Project manager, timeline research session

Research synthesis surfaced why AI-generated content carries a different risk profile than a typed field: a user filling in a field knows what they typed, but a user accepting AI output has to trust the system got it right, every time, across hundreds of projects and timelines. Two findings mattered before either feature could ship: as Pronto's first AI feature, no trust pattern existed anywhere in the product for AI-generated content, and users wanted the time saved but weren't willing to accept auto-committed output, since errors could compound across a project. Stakes also varied by feature. A task description is easy to fix after the fact; a timeline structures dependencies across an entire project, which makes mistakes there far more expensive to unwind.

The decision

Without a precedent for what AI should look or feel like in the interface, the identifying visual cue also needed deciding: a red-purple gradient pattern, matching the intensity of the primary CTA, was used consistently across both features as a reusable signal for "this is AI-generated."

Image placeholder

AI Task Helper entry point, multi-option button in the task description header

Drop a 2x PNG export here

Image placeholder

AI Timeline Wizard entry point, single-click button in the control panel

Drop a 2x PNG export here

What shipped

AI Task Helper turns a short brief into a task description and stages it for review before anything is written, a 2-step flow. AI Timeline Wizard takes a plain-language brief plus optional supporting materials and produces a full project timeline the user previews, refines, then commits, a 3-step flow.

Image placeholder

AI Task Helper 2-step generate and review flow

Drop a 2x PNG export here

Image placeholder

AI Timeline Wizard 3-step generate, review, refine flow

Drop a 2x PNG export here

Failure and fix

Under deadline pressure, the Timeline Wizard launched with the preview and refine step cut, and AI-generated output auto-committed straight into live projects. Adoption came in under 5%.

I should have pushed harder for user testing before launch, and I didn't. That's on me.

Outcome

The review-refine structure became the reference pattern for every AI feature that followed. When it was skipped under speed pressure for Timeline Wizard, the adoption numbers made the case for why it existed in the first place, and the fix validated it a second time.

Learnings

• Trust with AI features isn't earned by accuracy alone. It's earned by making the user's ability to catch and fix a mistake visible before they commit to it.
• A pattern working in one AI feature doesn't mean it transfers to a higher-stakes one without testing. Speed pressure after one successful launch is exactly when testing discipline is most likely to slip, and exactly when it matters most.

What's next

Extending the review-refine pattern to batch task generation, and building adoption-rate monitoring into the review process for any new AI feature at launch, so a Timeline-Wizard-style drop gets caught in week one, not after several sprints.