
The Income Tax e-filing portal reportedly cost $20 M to build.
But every filing season it’s the same story: random timeouts, vanished forms, and pop-ups straight out of 2003.
What’s your honest take on the Income Tax portal?
So I gave Google Stitch, Figma Make, and UX Pilot the same challenge: redesign this disaster.
why this test actually matters
Most AI design comparisons test the same boring stuff: landing pages, dashboards, e-commerce.
I wanted something harder. Something that would actually reveal these tools' capabilities and blind spots.
Government websites are perfect because they're:

If an AI tool can handle THIS, it can handle anything.
Plus, we've all used these portals. You'll know instantly if the redesigns are actually better. But first, here’s what happened this week:
AI Tools That Made Me Question Everything This Week
An asynchronous, agentic coding assistant that clones your repo into a secure Google Cloud VM, understands full project context, and autonomously executes tasks like bug fixes, feature builds, tests, and dependency bumps. Ships plans and diffs, integrates deeply with GitHub, supports parallel tasks, and even offers audio changelogs.
Currently in public beta with usage limits; CLI “Jules Tools” adds terminal-first control and scripting.
2. Mantle
Modern cap table management with free and flat‑rate pricing and unlimited stakeholders. The dashboard shows investor activity, SAFEs signed, and scenario modeling (e.g., projected Series A dilution).
Provides gold‑standard cap tables, share classes, vesting, reports, and AI‑assisted workflows for option grants and reviews; designed for instant setup and founder‑friendly pricing.
Low‑code automation for technical teams; describe an automation or agent in natural language, and n8n generates a draft workflow to refine in the editor. Cloud availability (Trial, Starter, Pro) with monthly AI credits, rolling out to Enterprise and exploring self‑hosted.
Ranked #1 of the day on Product Hunt; bridges code with a visual interface and a large ecosystem of nodes.
The challenge: redesign India's most-hated portal
I picked the Income Tax e-filing dashboard. Here's why:
Used by 1 Bn Indians annually
Consistent top complaints: confusing navigation, session timeouts, unclear error messages
Complex: mix of forms, document uploads, status tracking, downloads
Mobile matters: 60%+ users on phones
Can't be too 'creative', needs to feel trustworthy/official
Here's what's broken with the current portal:
Here's what it needs to do as the bare minimum:
Check refund status without hunting through menus
Download Form 16 in one click
File ITR-1 without crying
View past returns without detective work
The exact brief (given to all 3 tools)
I gave each tool this exact prompt:
Let's see which tool actually gets it.
1. Google Stitch: The Overachiever
Time to first output: ~1 minute
What it produced: Full dashboard PLUS refund status page, ITR filing flow, and past returns page. Stitch didn't just redesign the dashboard, it redesigned the entire filing experience.
First impression:
Stitch went full product designer mode. It saw 'redesign the dashboard' and said 'actually, let me redesign your entire tax filing system.”
What it got right:
Systems thinking: Generated complete user flows, not just one page
Rapid iteration: Font and color alternatives in 5 seconds
Mobile + Desktop parity": Both versions feel native, not afterthoughts
Unsolicited completeness: Built login flow and filing steps without being asked
What it missed:
Aesthetic over trust: Feels like a fintech startup, not a government portal
Too startup-y: The modern, minimal design might confuse less tech-savvy users
Over-designed: Some interactions feel clever rather than clear
Lack of visual weight: Government portals need to feel "official" and Stitch's lightness works against it
The telling detail:
Stitch generated a complete login and filing flow even though I only asked for a dashboard. That's either incredibly smart product thinking or AI going rogue. Probably both.
Grade:
A for ambition and systems thinking, B- for understanding the actual user base
2. Figma Make: The Comprehensive One
Time to first output: ~2 minutes (slowest)
What it produced: Single comprehensive dashboard with both mobile and desktop versions in one shot. No separate chat needed, but the mobile version needs polish.
First impression:
Figma Make played it comprehensively. One page that tries to do everything, with built-in interactivity that actually works.
What it got right:
Interactive by default: Quick Actions are clickable, modals work, notifications open properly
Mobile-first principles: Generated responsive versions without being asked twice
Information density: Balances showing a lot without overwhelming
Native Figma integration: Easiest to hand off to dev team
What it missed:
Mobile version feels rushed: Desktop clearly got more love
Too much on one screen: Trying to be comprehensive makes it feel cluttered
Generic fintech aesthetic: Could be Paytm, could be Zerodha, definitely doesn't feel like government
Warning banner gets lost: Filing deadline should be more prominent
The telling detail:
Make is the only tool that made the Quick Actions actually work. You can click them and they take you where they should. That's the difference between a design and a prototype.
Grade:
A- for interactivity and completeness, B for visual design and mobile execution
3. UX Pilot: The Surprise Winner
Time to first output: ~1 minutes (fastest)
What it produced: Clean, focused dashboard with smart prioritization. Less is more approach.
First impression:
Wait, that's it? My first reaction was disappointment. Then I actually looked at it.
What it got right:
Clear information hierarchy: Eye knows exactly where to go first
Strategic color use: Green = good, Orange = action needed, Gray = neutral
Recent Activity timeline: Smart addition that adds context
Export flexibility: Can get code, export to Figma, generate variations easily
What it missed:
Too minimal: Hides complexity rather than organizing it
Lacks depth: Doesn't show as much information as Make
Quick Links feel tacked on: Right sidebar doesn't integrate well
Financial numbers could pop more: The actual refund amount gets lost
The breakthrough moment:
UX Pilot did something neither other tool did: it understood that govt portal users aren't just 'less tech-savvy', they're risk-averse and trust-sensitive. Look at how it handled visual hierarchy and trust signals.
It's the only design that didn't try to be cooler than it should be.
The telling detail:
UX Pilot was the fastest to generate, easiest to export, and had the lowest wow factor. But it's the only one I'd trust during the actual tax season when I'm stressed and don't want surprises.
Grade:
A for understanding context and user psychology, B+ for ambition and completeness
what actually happened (and what it means)
These tools produced different design ideas and in the process, revealed completely different philosophies:

The insight: AI tools can't escape their training bias. Stitch was trained on innovative products. Make on enterprise solutions. UX Pilot (being designer-made) on actual user constraints. That DNA shapes everything they produce.
the problem none of them solved
Not one tool asked:
Who's actually using this?
What devices are they on?
What are they afraid of?
What are the regulatory requirements?
They just started designing.
All three optimized for the design problem I gave them and not the user problem behind it. So, you still need to be a good product thinker.
what they're actually good at
All three beat the current $20 Mn portal at fundamentals:
Clear information hierarchy
Status visibility (progress bars, badges)
Mobile responsiveness
Reduced cognitive load
Modern interaction patterns
The gap: Context, constraints, edge cases.
These tools are exceptional at producing clean, usable designs. What they still need to improve on is the human psychology and the ability to understand confusion, hesitation or any emotion on the other side of the screen.
None of them thought like a 50-year-old in Jaipur filing taxes on their phone for the first time, afraid of triggering an audit. They thought like what they were trained on: PMs, enterprise designers, and consultants.
so which tool won?
For redesigning a complex government portal with real constraints:
Change the brief, change the winner.
Example: Designing a crypto startup landing page? Stitch crushes it.
Making an enterprise dashboard for the client? Make nails it.
Want to build a trustworthy interface that won't scare users? UX Pilot.
Pick based on your actual problem, not who 'won' a comparison test.
what you should actually do
Test with YOUR constraints: I tested government portals because they have unique constraints. What are your edge cases? Test with those.
Treat them like junior designers: Great at execution, need guidance on strategy. They won't ask "who's using this?" or "what are they afraid of?" You need to tell them.
Spend time on your prompt: My prompt specified device constraints, user literacy, and trust requirements. That shaped everything. 10 minutes on a detailed brief beats hours of vague iteration.
Combine tools: Stitch for exploration. UX Pilot for refinement. Make for production. They're tools, not religions.
You still need judgment: If I didn't understand government UX psychology, I wouldn't have known UX Pilot nailed it while Stitch missed. These tools are incredible, but they're not magic.
try this yourself
Pick the worst website you use regularly. Give it to these tools. See what happens.
Then reply: which tool surprised you? Which disappointed you? What did you learn about how these tools actually think?
Until next time,
Vaibhav 🤝
P.S. - If you redesign a government website and it's genuinely better, tag the department. Maybe someone will listen. Probably not. But maybe.
If you read till here, you might find this interesting
#AD 1
How to pick the right global payroll mode
Find your fit: Deel’s free guide breaks down 3 global payroll models with key benefits and tradeoffs for HR and finance teams.
#AD 2
8 Weeks. Actionable AI Skills. MBA-Style Networking.
The 8-week AI for Business & Finance Certificate Program helps you:
Build AI confidence with role-specific use cases
Learn directly from leaders at top financial firms
Build a lifetime network that continues to open doors long after the program ends
Earn your certificate from Columbia Business School Executive Education—program starts November 10.