Concept projectAgentic AIB2B SaaS2026

Ariadesigning an AI copilot people can actually trust

A self-initiated exploration of agentic AI for product teams. What would it take for an AI to do real work on a team's behalf, while every person stays in control of what goes out the door?

Role
Solo designer, end to end
Timeline
~4 weeks
Scope
7 high-fidelity screens
Methods
Practitioner conversations, secondary research, scenarios
Product
Lumen, a fictional B2B product platform
The problem

AI features in SaaS tools look great in demos, then get quietly ignored by week two. Usually it's not because they're wrong. It's because people can't tell what the AI did or why.

What I did

I made up a product called Lumen and designed Aria inside it: a copilot that drafts, flags and connects research, but never sends or publishes anything without a person signing off.

The result

A seven-screen system built on one idea: show the reasoning, ask before acting, and make every action reversible.

Lumen · Orion — Q3 Launch · Overview
Live prototype
01 · Context

Most AI features feel like a black box

This project started with a question I kept running into in my own work: why do AI features in SaaS tools feel so unsatisfying to use? They impress in a demo, then people stop opening them. I wanted to design an answer from scratch, with no constraints except being honest about the problem.

To do that, I created a fictional B2B product platform called Lumen. The scenario is a product team in the middle of a launch, spending a big part of their week on coordination work: synthesizing research, writing PRDs and tracking sprint health. The AI tools they have feel too generic to trust with anything important. Lumen doesn't exist. The problem it stands for does.

The core tension: AI that's powerful enough to be useful is also powerful enough to cause real damage. So I treated this as a trust problem first and an AI problem second.
How might we

…let an AI take real work off a product team's plate, without anyone losing track of what it did, why it did it, or how to undo it?

02 · Research

Grounding a concept in real patterns

Since this was a concept, I didn't have a user base to study. So I pulled from three places and turned what I found into design scenarios, realistic situations that every screen had to hold up against.

Practitioner conversations
Informal talks with designers, PMs and engineers who use AI tools every day, about what has and hasn't changed in their work.
Secondary research
Published work on AI adoption in knowledge work, plus practitioner writing on workflow automation.
Failure patterns
Common ways enterprise AI tools lose people's trust, and the point where they tend to get abandoned.

Four patterns kept coming up, and each one shaped a part of the design.

1
People want a head start, not a finished answer
People push back on AI that makes their decisions, but welcome AI that handles the prep work: transcribing, organizing, first drafts.
2
Trust breaks when the AI doesn't say why
When an AI changes something and gives no reason, people delete it and don't come back. Not knowing why hurts adoption more than the occasional error.
3
Undoing is the most anxious moment
Wanting an undo button is really wanting to understand what changed. "If I reject it, will it keep trying?" was a real worry.
4
Synthesis is the obvious first job
Turning transcripts into themes is the most tedious part of research. It takes rigor but not much judgment, which made it Aria's first capability.

"I don't want AI to tell me what to build. I want it to do the stuff I have to do before I can think."

A feeling that came up again and again in conversations with product practitioners
03 · Who it's for

Designing for the cautious adopter

I combined the research into one primary persona to anchor every decision. Sofia is a composite, not a real person, but everything about her comes from patterns I heard and read.

Illustration of Sofia Reyes, the persona
Sofia Reyes
Senior UX Researcher, embedded with a product squad
Mid-size B2B SaaS · 80-person product org
"I'm not afraid of AI. I'm afraid of AI that makes me look bad in front of stakeholders without warning me first."
Attitude toward AI
SkepticEnthusiast

Cautiously optimistic. Trust has to be earned a little at a time.

Background

Seven years in UX research. About 40% of her time goes to synthesis, write-ups and stakeholder updates instead of research. She's tried three AI tools in the past year and kept none of them past week two.

Goals
  • Get research into product decisions faster
  • Spend more time talking to users, less time writing up
  • Stay credible with her PM and eng leads
  • Control how her research gets represented
Frustrations
  • AI that oversimplifies nuanced findings
  • No way to check what an AI did or why
  • Pressure to use AI she doesn't trust with final work
  • Insights scattered across Dovetail, Notion and Jira
Anything Aria makes is clearly labeled, so Sofia always knows what she's signing off on.
Drafts are easy to edit. Her rewriting them is the point, not a failure.
Aria flags what it's unsure about instead of sounding confident.
Nothing leaves the product without her approval.
04 · Principles

Four rules I set before designing a single screen

Every design decision after this traces back to one of these.

Show the reasoning, not just the result
Every Aria action carries a one-line note: what it looked at, what it compared, and why it acted. You can expand it or ignore it, but it's always there.
Every screen
A person approves anything that can't be taken back
Aria never messages stakeholders, sends updates or publishes documents on its own. It drafts, and people decide. There's no auto-publish.
Synthesis · PRD · Activity log
Every action can be undone and understood
The activity log lists everything Aria did in plain language, and any action can be reversed from there.
Activity log
A collaborator, not a chatbot
Aria brings things up without being asked, but frames each one as a suggestion that needs a response, not a finished action.
Dashboard · Sprint board · Insights
05 · How much Aria does on its own

The higher the stakes, the more a person is involved

The real question in agentic AI isn't what the AI can do. It's what it should do without asking. I sorted every Aria behavior into four levels, so the line between helping and overstepping was a design decision, not an accident.

Level 1 · You read or dismiss it
Inform
Aria notices something and tells you.
Risk: stakeholder review may slipPattern: nav complaints come from new users
Level 2 · You say yes or no
Suggest
Aria proposes an action and waits.
"Want me to flag these as blocked?"Link this insight to the onboarding PRD
Level 3 · You edit and approve
Draft
Aria creates the work, clearly labeled and held as a draft.
Synthesis waiting on "Approve"PRD held in Draft status
Level 4 · Always a person
Act outside the product
Messaging stakeholders, sending updates or publishing. Aria never does this on its own.
Send for reviewStakeholder status emails
Lower stakes · Aria acts freelyHigher stakes · a person decides
06 · The system

Following one finding through the whole product

The seven screens aren't separate features. They tell one story: a team discovering that new users are overwhelmed by notifications, and Aria carrying that finding from raw interviews all the way to a PRD, with a record at every step.

Interviews
Aria transcribes sessions and tags quotes
Synthesis
18 quotes become 4 themes, waiting for approval
Insights repo
The finding is saved and linked to onboarding drop-off
PRD
Drafted from the synthesis, with 3 open questions for the PM
Sprint board
The work shows up where the team already tracks it
Activity log
Every step recorded, explained and reversible
07 · The work

Seven screens, one system

The hardest balance was making Aria visible enough to be useful and quiet enough to stay out of the way. Here's how each screen handles it. These are the live prototypes, so the panels scroll.

Screen 01

Dashboard

Aria is always there, but never in the way.

Lumen · Orion — Q3 Launch · Overview
Live
  1. 01Familiar on purposeThe main layout stays familiar. Aria lives in a panel on the right and never covers the actual work.
  2. 02Reasoning on every cardFour insight types (risk, synthesis ready, draft ready and pattern found) each end with a one-line reasoning note.
  3. 03Teal chips show provenance"Aria" chips on tasks show at a glance what the AI touched and what a person did.
Screen 02

Research synthesis

A first draft the researcher signs off on.

Lumen · Research · Synthesis
Live
  1. 01The researcher owns the conclusionA banner says exactly what Aria did and when. The main action is "Approve synthesis."
  2. 02Themes tie back to peopleEvery theme links to real quotes from named participants in the side panel.
  3. 03Honest about the dataOpen questions call out what the data can't answer, like the fact that only new users were interviewed.
Screen 03

Sprint board

Aria shows up where the work already is.

Lumen · Orion · Sprint 6
Live
  1. 01Edges, not a new tabCards Aria worked on get a teal edge and a short note. Risks, like a review with no meeting booked, turn coral.
  2. 02It asks before actingWhen four tasks stall for five days, Aria asks before flagging them and drafting a status update.
Screen 04

PRD editor

An AI that's willing to say "I don't know."

Lumen · PRDs · Notification preferences
Live
  1. 01Plain on purposeThe editor feels like a normal doc. A banner shows who drafted it and what it was based on.
  2. 02Open questions, not guessesInstead of filling gaps with confident guesses, Aria lists open questions and routes each one to a named owner.
  3. 03Every requirement is traceableThe sources panel traces requirements back to the research they came from.
Screen 05

Insights repo

Research that doesn't die in a slide deck.

Lumen · Research · Insights repo
Live
  1. 01Evidence and follow-throughEach insight shows how many sessions and quotes back it up, and whether it's linked to a PRD yet.
  2. 02Aria connects the dotsIt spotted that notification fatigue and onboarding drop-off overlap, and suggested fixing both in the same onboarding step.
Screen 06

Aria setup

Teams turn Aria on one capability at a time.

Lumen · Settings · Set up Aria
Live
  1. 01Plain-language togglesEach toggle says what Aria will actually do, not just "enable AI features."
  2. 02A preview you can pictureA preview column shows the cards you'll start seeing, because people trust what they can picture.
  3. 03A promise up frontAria stays in the workspace, and everything is logged and reversible.
Screen 07

Activity log

Proof that Aria has nothing to hide.

Lumen · Aria · Activity log
Live
  1. 01Status and undo on every actionEvery action shows whether it's waiting on you, done or undone, and can be reversed right there.
  2. 02Misses are shown, not hiddenOne entry shows a teammate undoing a summary because she wanted to write it herself.
  3. 03A full reasoning traceIt ends with Aria keeping the PRD in draft and notifying no one until a person reviews it.
08 · Key decisions

Three calls that defined the product

1
A reasoning note on everything
Every insight, draft and risk flag comes with a one-line explanation of what Aria used and why it acted. What I learned: people don't read it every time. They just need to know it's there.
Trade-offMore text on every card. I kept it to one quiet line that expands on demand, so it builds trust without adding clutter.
2
"Aria" chips on tasks
Anything Aria drafted or completed carries a visible chip instead of a vague "auto" label. The team can see right away where AI was involved and where a person still needs to review.
Trade-offLabels can feel like noise. Keeping them small and consistent made them a habit to scan for, not a distraction.
3
Nothing goes out without a person
Aria drafts and people send. One surprise email to a stakeholder on someone's behalf would wreck trust faster than any amount of friction, so this rule was never up for debate.
Trade-offAn extra click before anything goes out. That's friction I chose on purpose.
09 · Outcomes

What the design delivers, and how I'd measure it

Concept project. These are design outcomes, not results from a live product.

100%
of Aria actions can be reversed from the activity log
0
stakeholder messages, updates or publishes without a person's approval
4
insight types Aria brings up on its own: risk, synthesis, draft and pattern

Design goals and where they live

GoalHow it's handledWhere
AI actions are clearReasoning note on every insight and draftEvery screen
People stay in controlNothing goes out without approvalSystem-wide
Mistakes are recoverableFull log with undo on any actionScreen 7
Research reaches decisionsInterviews → synthesis → repo → PRDScreens 2–5
Helpful, not pushyAria suggests and flags. People decide.Screens 1, 3, 5

If it shipped, I'd track

MetricWhat it tells me
Use after week 4Whether Aria survives the point where most AI tools get dropped
Draft approval rateWhether drafts are a real head start or get rewritten from scratch
Undo rate by capabilityWhich capabilities people don't trust yet
Interview to approved synthesisThe actual time saved on the most tedious step
10 · Reflection

What I'd do differently

Push onboarding further
The setup screen shows static previews. I'd try interactive ones, because trust gets built before someone ever uses a feature.
Test team dynamics earlier
Aria works in a shared space. What happens when it flags something a teammate made? Team scenarios earlier would have sharpened the social side of the design.
Defaults matter more than options
Most people never open settings, so whatever is on by default is the product they get. I'd test the default setup much harder than the configured one.
What this project shows

AI design is a systems problem, not a feature problem

The hardest questions here weren't visual. They were about trust, transparency and control, and where helpful automation turns into unwanted interference.

It also shows how I take something as open-ended as "AI for product teams" and give it structure: a clear scenario, grounded research, a set of principles, and a story that holds together across every screen.