Tastewise, Product Designer home assignment

Radar

Agents that notice what matters, so Maya doesn’t have to know what to ask.

Roni Brot-Kolinkó
01 · The problem

Tastewise has the answers. Maya doesn’t have time to ask.

Tastewise can answer almost any question, but Maya has to start the conversation. She’s a brand manager, not an analyst, and she rarely opens the product. The challenge is to make the product useful before she thinks to open it.

Push should mean more than sending Maya notifications. It should mean the product keeping watch for her, noticing what matters and bringing it to her while there is still time to act.
Maya

Meet Maya

Brand Manager at a mid-sized CPG company, responsible for five natural supplement SKUs.

Measured on

Sales across her existing range, and whether new products succeed.

Her day

Meetings, email, spreadsheets and decks. She isn’t an analyst and doesn’t spend her day in the product.

How she uses it today

About twice a month, usually for a quarterly review, retailer meeting or launch brief.

How her work runs

Daily: meetings, approvals and questions from the team.
Weekly: performance reviews, status updates and management reporting.
Periodic: retailer reviews, quarterly planning and launch briefs.

What she needs

Walk into retailer meetings with data she can defend. Choose the next launch and explain why. Spot important changes before they become surprises.

What’s missing

She has to know what to ask before the product can help. The most valuable signals may be the ones she hasn’t noticed yet, while her team is waiting for answers.

02 · The thinking

What I explored, and why I bet on Radar.

Six directions I looked at before choosing one.

DirectionChannelWhy not

Weekly digest

EmailA newsletter she skims. It can’t tell her what’s urgent.

Alerts with thresholds

ProductNeeds setup skill she doesn't have.

Briefing before meetings

CalendarThe night before is too late. Only fires around meetings.

Team bot

SlackThe team becomes the user. Maya watches.

"For you" home

ProductShe's rarely in the product.

Slides plugin

SlidesNeeds a plugin, and only helps when she’s already building a deck.

Each direction solved part of the problem. The final concept brings the strongest pieces together.

The bet

Radar

Radar gives Maya agents that keep watch on the market from day one, set up from her brand profile. She can adjust them at any time to match her own cadence, and get updates where she already works, like Slack or email. Each update is prioritized by importance and time sensitivity, and suggests next steps. Maya decides which ones to turn into a slide, an email or a meeting.

Assumptions

  • Maya creates her brand profile in a short onboarding when she first starts, and the agents are set up from it.
  • Updates can reach her in Slack, Teams or email.
  • Agents work from the market data Tastewise already collects. They don’t need new data sources.
  • Retailer calendars (resets, promo windows) are available, so a window can carry a date.
  • Her team can see what she shares without a license.

Design principles

The decisions that shaped the prototype.

Agents are ready from day one

Radar sets up five agents from her brand profile, so she doesn’t start with an empty screen. She can turn any of them off or add her own.

Updates follow her schedule

Each agent reports daily, weekly or before a planning moment, matching how her work already runs.

Only deadlines interrupt

If something has a closing window, she hears about it right away, with the date. Everything else waits for its regular time.

Each update starts with why it matters

The first line explains what it means for her brand. The numbers come after, each with its source.

The next step is suggested

Maya chooses a slide, an email or a meeting invite, and Radar creates it for her to review and send.

Feedback happens inside the update

She marks an update as relevant or not, and the agent learns from it and tells her what will change.

Walk through the prototype →
03 · The flow

Radar notices, Maya decides.

How an update reaches Maya, and how she shapes what comes next.

Click any step to see it in the prototype.

Flow 1 · From noticing to acting
Agents watchfive preset agentsClosing window?yesArrives immediatelywith the deadlinenoArrives as scheduleddaily, weekly or periodicThe update arrivesone or more insightsMaya picks onein Slack or emailThe insightwhat changed, why it mattersCreate slidefor the Target deckDraft emailto the teamSchedule meetingbefore the deadlineThumb up or downRadar adjusts the agent
Flow 2 · Creating or adjusting an agent
Maya writes or edits the promptin her own wordsShe reviews the agent cardsees a sample, then savesAgent runningready for its next update
04 · Working with AI

What I asked for, where AI fell short, and what I decided.

I used ChatGPT and Gemini for background research, and Claude Code to plan and build. I made the product decisions and wrote most of the final copy. Claude handled the implementation, and I reviewed every round in the browser.

What I asked for

  • A plan and open questions before any code.
  • A full copy draft to review before the build.
  • Short rounds: build, review in the browser, refine.

Where AI fell short

  • Its first concept, a pre-meeting briefing, was too narrow.
  • It proposed a metric the product couldn’t measure.
  • Its copy read as generated. I rewrote most of it.
  • It treated cadence as a setting and missed urgency entirely.

What I decided

  • Reframed the concept twice, away from the meeting.
  • Added Maya’s three cadences and the closing window.
  • Put feedback inside the update, not in settings.
  • Kept the scope to one piece: the update and what she does next.
05 · What I'd measure

Three numbers that would tell me it worked.

Habit
Weekly active use, up from roughly two sessions a month, plus update open rate by channel.
Insight to action
Share of updates that lead to a slide, email or meeting.
Signal Quality
Agent retention over time + decline in “Not relevant” rate.