Product designer and builder.
Nashville, TN
Joined April 2007
- Tweets952
- Following1.1K
- Followers1.3K
- Likes10.2K
Color me impressed. AI one-shots can be really good with the right context.
This week, I wrapped the design for a new greenfield project. We’re building it with AI-DLC, a development process from AWS that gives structure to how AI fits into the development lifecycle.
Next up: requirements for Inception, the first phase of AI-DLC. I started with a colleague’s template and filled in about half from my own knowledge of the project. Then, in Cursor, I pointed Opus 5.5 (High) at it with this prompt:
“Would like to create a copy of this doc [redacted].md and populate with relevant information from this vision doc [redacted].pdf. I also have the designs for this new product in Figma on this page: [redacted]”
Cue the mechanical whirring sounds I like to imagine. Minutes later, in one shot, the requirements doc was populated accurately.
But it wasn’t magic. I’d spent real time organizing the Figma file, naming every screen by its place in the information architecture, its purpose and state, and its intended user. When every prompt has a price tag, good context isn’t a nice-to-have. It’s the cheapest optimization you’ll ever make.
Anyway, the robots are ready when you are. Your “Frame 427 copy 3” might not be.
Discernment in tool choice, is becoming even more important these days.
Spent the past few weeks doing design work for a greenfield project almost exclusively in Figma. Leading up to that, agents helped me synthesize research and brainstorm feature sets. I also used them to help me create a design system (both in Figma and in code).
It's so easy though to get caught up in all the hype. It's exciting and I think most of us love learning new tools and techniques. Remember though, agents are effectively very smart hammers. It's ok to pick an old trusty "dumber" one when it's a better fit for the job. It might just get you there faster too.
Threw together a very rudimentary .art site so I have a place to share random drawings and stuff. I'm a bit rusty but trying to draw something most days.
Everything there is drawn with a Wacom MovinkPad Pro 14 and Clip Studio Paint (thanks years of Amazon reward points).
When I finish a drawing, I upload it to a Google Drive folder and a CRON job on Vercel picks up new files and adds them to the site automatically.
Anywho, it's here if curious: alday.art/
Who has tackled the complicated Venn diagram of DESIGN.md, a shared code-based UI component library, and Storybook?
Feel like this is generally greenfield territory so looking for advice and landmines to avoid.
Thanks!
So, so rusty. Picked up a new Wacom tablet to get back into the drawing habit. Love how their stylus and screen combination feels compared to the Apple Pencil.
Geoff Alday retweeted
you may not know this, but you don’t have to have 100 agents simultaneously looping for you to ship something people want.
Geoff Alday retweeted
I was just in Rome, Italy this month doing a collaboration with 40 teens from around the world. I tried "imagination portraits" where I used AI to create their imagination and then projected it onto their face. Can't wait to show you the results. Would you want to try this?
If you’re a designer and aren’t yet paying attention to adaptive UI, now is the time.
It’s an emerging area and you’ve probably seen it in use in your favorite chat app (E.g., Claude’s interactive apps).
A few protocols and initiatives to start:
A2UI
a2ui.org/
JSON-RENDER
json-render.dev/
MCP-UI
mcpui.dev/
An in-progress project I'm proud of, warts and all.
Lirio's content team writes health messages for 7+ channels and couldn't see how they'd render while writing. That meant truncated text, blown SMS limits, and rework.
So I built them a live preview tool covering email, SMS, RCS, WhatsApp, MyChart, push, and voice, plus segment-aware SMS optimization, LLM translation, readability evaluation, and branded mockup export.
They've used it since day one. It's not polished yet and that's ok. It was more important to get something functional to them to help unblock their work. Sharing as an encouragement to show what one person can do with AI as a collaborator.
Walkthrough:
alday.io/#lirio-preview-tool…
Geoff Alday retweeted
A lot of designers follow every rule during the interview loop.
Keep 90% of that advice, but find the 10% that makes you memorable.
Differentiate yourself from the other candidates. Put on a show. The people reviewing your portfolio are taking 45 minutes out of their busy day to meet with you, don’t bore them. Keep them engaged.
My go-to move is raising my standing desk and presenting on my feet, talking through my work with energy, moving my hands around, and being my own hype man.
Skills get you into the room. Belief in yourself, personal brand, and conviction is what people remember.
Geoff Alday retweeted
One of the key advantages of wireframes over Figma or vibe-coded prototypes is how they let you express an actual idea. Not a fake finished product that everyone has to x-ray to understand what it implies.
In this week's adventures in AI...
We're about to launch a lightweight reporting tool for our users. It covers historical message performance, health outcomes, and delivery across all programs and clients. There's also a daily operational view.
Our Data Science team built it with Streamlit. They picked it because they're more at home in Python than in the JavaScript-based tool we'd been using, and it ships with a lot of useful capabilities out of the box.
Once they had it wired up to our Analytics Data Model (ADM), they asked me for a UX review. Functionally, everything was fine. But it didn't align with our brand or UI standards, and those aren't just aesthetic. They cover usability too: how readable the metrics are, how findable the controls are, how predictable things feel as you move around. My first instinct was the standard move: redline it and hand it back. Then I reminded myself it's 2026, and we have robot friends now.
One catch. I didn't have access to the protected environments the ADM lives on. There's an approval process (I know, we all love those), and my manager was on vacation, so I was looking at a week or more before I could even get in.
So I fired up Cursor and pointed it at the repo that has the schema for the backend server Streamlit uses. I had it build me a mock data server with accurate, representative values. No dev data access meant this was the fastest way to get moving and actually help the team. With that done, I could run the front end locally against the mock server.
Then I got to see how CSS works in Streamlit. Oh boy. I've been writing CSS a long time, and let's just say: this is weird. I fought with its gnarled teeth, and came out mostly unscathed. I stayed primarily in the CSS, but did modify some of the HTML generated by a few Python scripts: reorienting one section into rows because compressed columns were truncating metrics, stripping out redundant phrases like "You set values to..." that prefaced every section (the controls are always right there), capitalizing for consistency between settings and display. Small things.
I submitted the changes as a PR, only having to tweak a few Python project organization standards along the way. It's been a minute since I've worked in Python.
A day's worth of work, and the Data Science team got back a front end that actually looks like ours, plus the mock server to keep building against. My manager is back next week. I still don't have access to the ADM. Guess I don't need it.
Super happy with how this turned out. Wanted to add something interactive to my portfolio site that wasn't entirely frivolous. Came up with an interactive hero section over the weekend and put it together.
Let me know if you try it out!
alday.io/
