Case study · Żabka Future challenge · AI+UX
Designing an internal communication tool with AI as my design partner
Żabka Future asked how product teams can communicate their internal tools without extra work. In two weeks I designed TeamMessage, a SaaS concept that takes one source document and turns it into ready-to-send communication for every channel. I used ChatGPT, Bubble AI, Figma Make and Claude across the process. The decisions about scope, flow and UI stayed with me.
~2 weeks
Timeline
From the brief to a working, clickable prototype.
4
AI tools
ChatGPT, Bubble AI, Figma Make and Claude, each for a different step.
1
Designer
Solo, from research to prototype, with feedback from Żabka Future.
Context
Supercharge: AI+UX Skills for Product Makers (2025)
Challenge by
Żabka Future
Users
Product teams that build internal tools
My role
Product designer, solo
Platform
Web / SaaS
Prototype
01 · The challenge
How can we take good care of communicating internal products?
Teams that build internal tools rarely have anyone dedicated to communicating them, so people don’t know a tool exists or what changed. Żabka Future’s Product Design Manager asked for a prototype that turns documents about an internal product into ready-to-use materials: intranet posts, emails, release notes and presentations.
Consistent communication
Every release and change gets a prepared message.
Product awareness
Employees know the tool exists and what it’s for.
Less work for the team
Less time spent preparing communication materials.
02 · Research
Understanding the work around a message
I talked with the challenge owner at Żabka Future and used AI-assisted desk research to see how similar products handle publishing, approvals and distribution.
Key insight
The problem isn’t writing the message. It’s the decisions around it: who gets it, who approves it, which channel it goes to and when.
03 · AI in the process
How I worked with AI
AI sped up every stage of this project, but it never made the call. Here is what each tool did, and what I decided.
Step by step
Seven steps from the brief to the prototype. On the left, the tool. On the right, what AI produced and the decision that stayed with me.
01 · Research
AI desk research + stakeholder talks
AI: Mapped how similar products handle publishing, approvals and distribution.
Me: Decided the real problem is the decisions around a message, not the writing.
02 · First ideas
Pen and paper
03 · Expanding the idea
ChatGPT
AI: Turned my notes into 12 feature ideas and 3 “big” features.
Me: Cut it to the core flow: no integrations, a simpler compliance check, slide decks later.
04 · App blueprint
Bubble AI
AI: Generated a blueprint of 14 key features from my prompt.
Me: Used it as a scope checklist and refined the prompt myself.
05 · Prototype, light and dark
Figma Make
AI: Built a working prototype in light and dark mode, plus design tokens and WCAG contrast guidelines.
Me: Defined the flow, information architecture and brand direction, and decided what stays.
06 · Refining by hand
Figma → Figma Make
AI: Rebuilt components from my Figma designs.
Me: Redesigned key components by hand, like the team and recipient picker.
07 · Copy and improvements
Claude
AI: Wrote the UX copy and proposed improvements in 18 areas.
Me: Picked toast notifications and the scheduler for the MVP. The rest waits.
One real prompt · Bubble AI (translated from Polish)
“A tool for employees who create messages for the whole team. It accepts documents, and you can also write a new information document for the team in it. It guides you step by step and outputs ready communication that can be sent through the channels you choose: email, intranet, release notes, presentation – exactly as in the brief. It should use green colours in line with Żabka. Documents can be created by different teams: give 5, e.g. legal, HR, general…”
Left: my hand-written notes from step 2. Right: the blueprint Bubble AI generated from the prompt above.
AI gave me speed and breadth. I gave it direction, constraints and the final say.
04 · The solution
One workflow, from a source document to a sent message
TeamMessage brings the whole lifecycle of an internal message into one tool. Four decisions shaped it.
Audience is an explicit choice
Pick a whole department or specific people in a two-column picker.
Approval lives inside the workflow
Draft → Review → Approved → Sent, as a state inside the tool, not an email outside it.
Distribution is part of creating
Channels and schedule are set while the message is being written.
AI as an assistant, not the author
The AI generator drafts versions per channel. The communication manager edits, approves and sends.
05 · Prototype
Try it yourself
Click through the full flow: log in, create a message, pick the audience and channels, then send it for approval or schedule it. Demo login details are on the sign-in screen, and the sun icon switches between light and dark mode.
All names and content in the prototype are fictional.
06 · Outcome
What came out of it
Completed and submitted in the Supercharge: AI+UX Skills for Product Makers programme. It’s a high-fidelity concept, so there are no usage metrics.
“The well-thought-out project structure and the completeness of the SaaS solution deserve recognition. I like that the process can be done by hand or with AI support at different stages. Features like tone-of-voice customisation and checking the written content are also worth noting. If I were to point to one next step: spend a bit more time refining the insight.”
Product Design Manager, Żabka Future
07 · Learnings
What I learned, including about AI’s limits
01
AI expands, designers narrow down. The real work was choosing what serves the core flow.
02
Generic by default. Brand fit and key interactions came from my own designs, not from prompts.
03
Data stays synthetic. I used only fictional names and content in AI tools.
04
Next time: validate the insight earlier and research the employee side, how people actually read internal messages.