Incorrect password
Cisco · Duo Mobile · AI Pod

Onboarding
AI Experiment

Can a small pod move faster — and build better — using AI?

Team
Lei · John · Marshall · Angela
Duration
4 weeks
Cisco · Duo Mobile · AI Pod

The project

Duo Mobile is Cisco's two-factor authentication app. Onboarding is the first-run flow that gets a new user's account linked and push-based verification working — first impressions that shape whether people trust and stick with the app.

◆
My role — Product Designer
I led the design work on this project: the redesign direction, the screens, and the design → engineering handoff. I worked alongside Lei, John, and Marshall in a 4-person pod exploring how far AI tooling could push speed and quality across product, design, and engineering.
Product
  • AI-assisted product discovery
  • Faster alpha → beta cycles
  • Use Amplitude data to guide prioritization
Design
  • Increase design velocity with AI
  • Translate feedback into faster iterations
  • Tighten design → engineering handoff
Engineering
  • Requirements → technical plan via AI
  • AI-assisted code review
  • Ship faster without sacrificing code quality

The problem

Amplitude showed us that users were actively choosing to skip parts of the flow. Are these screens actually useful, or are they just in the way?

Current flow
1
Create & Name Account
2
Practice Push
↓ Drop-off
2a
Backup Encouragement
3
Settings Encouragement
Account Linked screen
Account Linked
~60%
choose to skip
practice flows

Where the friction adds up

Many screens are purely informational — no meaningful action, just taps to proceed.

Total taps — with practice
iOS
15
taps
Android
18
taps
Informational screens — no meaningful action

Our approach

Move fast on discovery and prototyping by putting AI directly into the design workflow, not just around it.

Leveled up on AI tooling together
Shared how each person was already using AI in their work, then got the whole pod set up on the same tools — Git, Xcode, MCPs — before any real work started.
Competitive research via Claude
Used Claude to survey how other apps handle onboarding — surfacing patterns and informing design direction faster than a manual audit.
Prototyped directly in Xcode
Used Claude Code to generate SwiftUI screens directly in Xcode — seeing designs run as real code in the simulator, not just as static mocks.
Xcode with app running in simulator

The redesign

Reordered to surface permissions earlier, made Practice Push optional, and removed screens with no meaningful action.

Before
After
1
Create & Name Account
2
Practice Push
↓ moved to end
2a
Backup Encouragement
Android only
3
Settings Encouragement
↑ moved earlier
1
Create & Name Account
2
Settings Encouragement
2a
Backup Encouragement
Android only
4
Practice Push Skippable

Screens we changed

Changes were kept as lightweight as possible. The priority was improving the flow, not a full visual redesign.

Merged
Welcome screen
Removed
Name your account
Redesigned
Almost there → You're all set
Redesigned
Perfect!

Design → Eng handoff

After finalizing the screen changes, we used the Jira MCP to have Claude generate a full epic and task breakdown directly from the design decisions — no manual ticket writing.

Claude + Jira MCP
Created epic ZTMOBILE-4444 and 10 child tasks directly from design decisions — no manual ticket writing.
iOS + Android split
Tasks automatically paired per platform, 5 each — ready for engineers to pick up directly.

From design to working code

Design outputs → implementation plan
Claude was given the Jira epic + Figma screens and generated a full implementation plan and code.
~80% of Claude's code merged as-is
"It will take some time for me to fix everything up and get it merged into master."

Results

The redesigned flow, built and running — engineering's initial implementation, with the refinement that followed.

Android

* Backup steps not shown in this recording.

w/o practice
w/ practice
Before
14
taps
18
taps
↓
−5 taps
↓
After
9
taps
13
taps
iOS

* Recording starts after the welcome screen.

w/o practice
w/ practice
Before
11
taps
15
taps
↓
−5 taps
↓
After
6
taps
10
taps

What worked

✓ A small, focused team meant faster decisions, more room to experiment, and a notably fast path from concept to working demo.
✓ Blurred discipline boundaries — everyone had more visibility into each other's work and direction across the whole project.
✓ Claude sped up design ideation — explored more directions before committing to one.
✓ The Amplitude MCP made product data universally accessible and easier to interpret across the whole team.
✓ Generating Jira tickets directly from design decisions made the handoff to engineering more streamlined and less manual.
✓ ~80% of the code Claude generated was usable by engineering as-is.

What didn't work

Friction points that emerged across each discipline during the experiment.

PM
•Problem definition still needs a human owner.
•No clear decision-maker meant progress stalled.
•Weekly alignment still required a dedicated sync.
Design
•Weak foundations limit how much AI can actually accelerate your work.
•Code as a design deliverable doesn't add value for eng.
•Sometimes it's just faster to make the change in Figma than to prompt and wait.
Eng
•Eng couldn't start until design was finalized.
•Initial ramp-up on docs and tickets was unchanged.
•PR reviews and merging took the same amount of time.

Key takeaways

↑ AI can compress a lot of the upfront work: research, ideation, and getting things moving faster.
– Bottlenecks still exist. Anything that requires judgment, decisions, or polish still takes time.
→ Better handoffs unlock more acceleration. The quality of what you pass forward matters.
PM
handoff
Design
handoff
Eng
AI accelerates
Bottleneck
One experiment is a starting point, not a playbook.
Cisco · Duo Mobile · AI Pod

Thank
you

A 4-week experiment in using AI to move faster from problem to shipped code — without losing craft.

Questions
Happy to dig into any part of this
More work
angelaro.com