Ahmad Eldeeb · Deeblab (@deeblab) header
Ahmad Eldeeb · Deeblab (@deeblab) profile picture

Lead product designer, 10 yrs shipping apps and games. Sharing how I work and where design and AI are going. Deeblab studio for product teams, DMs open.

Cairo · MENA & remote
Joined January 2026
Photo by Ahmad Eldeeb · Deeblab (@deeblab)Photo by Ahmad Eldeeb · Deeblab (@deeblab)Photo by Ahmad Eldeeb · Deeblab (@deeblab)Photo by Ahmad Eldeeb · Deeblab (@deeblab)Photo by Ahmad Eldeeb · Deeblab (@deeblab)Photo by Ahmad Eldeeb · Deeblab (@deeblab)Photo by Ahmad Eldeeb · Deeblab (@deeblab)Photo by Ahmad Eldeeb · Deeblab (@deeblab)Photo by Ahmad Eldeeb · Deeblab (@deeblab)
One rule from a food ordering flow I designed. Edit comes before cancel. If cancel is the only way to fix a wrong address, a quick fix turns into a lost order, a refund and often a support ticket.
5
A one prompt check will catch the low contrast and the missing alt text, and that's worth running. It won't tell you a screen reader user got stuck in the menu on step two. I'd run the check and still sit with one real screen reader user every release. teapaw.org/framer/status/21224450…
Make your site more accessible with one prompt. Run /accessibility-check and let your Framer Agent find and fix accessibility issues.
2
Before and after of one screen. Shop owners opened a B2B grocery app to a discount banner, category shortcuts and brand rails, the same for every store. They weren't browsing, they were restocking. So the first screen became their usual basket with one Add all button.
4
I'd keep both loops. Agents are a new kind of user, so evals make sense for them. But a person still reads what the agent did and decides whether to trust it next time. You only learn that part by watching people. teapaw.org/rauchg/status/21224155…
The old way: users are human, you run user interviews, get inspired, mock up user journeys. The new way: users are agents, you run evals, evals inform the software engineering loop. Good read on how we do evals at Vercel.
3
Audi says touchscreens have gone as far as they'll go and buttons are coming back. BMW says it could follow if drivers ask. Cars learned the hard way that whatever you do without looking needs a fixed home. Apps too. The thing people do every day shouldn't live in a menu.
7
Which app from outside tech is better designed than most tech apps? An airline, a bank, a supermarket, your car. Drop the screenshot.
3
I'd call it a compliment too, but the tell isn't the era. It's the empty state, the error, the slow network, the long name that breaks the row. Fit and finish lives in the states nobody demos, and AI makes it very easy to skip them. teapaw.org/jsngr/status/212165202…
Rex by @superlogical is phenomenal software. tremendous fit and finish, feels like it was built and designed pre-AI (compliment)
1
ARC Raiders has no enemy health bars. You read the damage on the machine itself, smoke, sparks, armor falling off, and the sound changes. The studio says a number on top would be redundant. More apps should show state on the thing itself, not in a badge next to it.
2
Gates and metrics will catch the broken build and the drop in signups. They won't catch the screen that works and still confuses people. That only shows up when someone watches a real person use it, so I'd keep one human gate in the loop. teapaw.org/rauchg/status/21217113…
This is how agents will ship all code. → Agent writes feature-gated code → Verification gates on push (proof checking, security audits) → Agent monitors metrics You can live the future today with Vercel CLI: 𝚟𝚎𝚛𝚌𝚎𝚕 𝚏𝚕𝚊𝚐𝚜, 𝚟𝚎𝚛𝚌𝚎𝚕 𝚖𝚎𝚝𝚛𝚒𝚌𝚜, 𝚟𝚎𝚛𝚌𝚎𝚕 𝚍𝚎𝚙𝚕𝚘𝚢.
2
Apple launched a health AI study app this weekend. You ask a health question, the answer comes from sources like NEJM and JAMA, and you rate every answer you get. In health the model is the easy part. Getting people to believe the answer is the design job.
6
What's the best empty state you've seen in an app this year? The kind that tells you what to do next instead of just saying nothing here. Screenshots welcome.
6
Agree, and AI makes it harder. Every screen can now come from a different prompt on a different day. The products that still feel like one person made them will be the ones with a short list of rules someone actually enforces. teapaw.org/jsngr/status/212173574…
aim to make your software feel like it came from one person
11
I came into design from engineering. Built my first interface in 2015, became a product designer in 2017, and by 2019 I was owning whole systems, not just screens. Ten years later it's the same curiosity, the systems just got bigger.
10
I'd push back a little. Same behaviors matter, but people also learn apps by watching each other. If my date picker looks nothing like yours, nobody can send a screenshot and say tap here. A shared interface is how most help actually happens. teapaw.org/shadcn/status/21217786…
You and I don’t need the same interfaces. We need the same interface behaviors. Your agent might generate a calendar to pick a date. Mine might generate a simple input. Someone else’s might come up with something completely different. That’s fine. The interfaces can be different. The underlying guarantees should be the same. That’s what we’re building UI SDK for.
14
ChatGPT Wallet lets your agent do the shopping, but every purchase still stops on one screen with the items, the total and the card, then Face ID. The agent does all the work and the whole design lives in that last screen. It's the only part you'll actually look at.
7
Found a photo of the sticker in a bus cab that explains its warning lights. Over a hundred of them. It still reads fast because they're sorted into four priority groups, red first and green last. Most apps I use send every alert at the same volume.
8
My bet is both, and it makes sense from the design side. Agents need surfaces built for them, and most tools are still bolting MCP onto screens made for people. Building your own sheet or whiteboard is the fastest way to learn what that surface should be. teapaw.org/humphreybc/status/2121…
Anthropic has been quite open about the fact their strategy is for Claude to be an employee that you hire, and that Claude will use the software that companies already have to do work, like a person would. After all, salary expenses are greater than software. This launch feels inline with that strategy, in the sense that Claude joins meetings as a person would, but what I can’t reconcile is why they’re also building all of these SaaS apps like docs, whiteboards, sheets, etc. That doesn’t feel inline with the strategy of Claude being an employee that uses your existing tool stack and systems of record. Maybe they decided that a bunch of the productivity and collaboration tools like Miro, Excel, Google Drive, etc haven’t been building the right interfaces for agents quick enough so they felt the need to build their own. Yet Salesforce, Atlassian, Figma, Notion, us (Dovetail) are all working towards essentially MCP equivalence and headless or embedded versions of the products. So which is it? Is Anthropic trying to steal budget from Microsoft, Google, Atlassian, Miro, Notion, and so on to be the next productivity and collaboration Super SaaS? Or are they focused on building an AGI generalist ‘employee’ that companies hire to replace humans (and divert budget from salary), which uses existing tools? Maybe they just have so much product velocity that they are like, fuck it, we’ll have a go at both.
5
Step 6 is the one that should worry people. A design partner that calls every assumption excellent is just a mirror. I'd take one engineer asking what happens when the list is empty over all seven steps. teapaw.org/hobdaydesign/status/21…
I haven’t opened Figma in 6 months. And honestly? I’ve never been a better designer. AI has completely transformed my workflow. What used to take hours now takes seconds of carefully engineered prompts. Here’s my process: 1. I ask ChatGPT to design an interface. 2. It generates something incredible. 3. I ask it to make the interface more intuitive, delightful, and user-centric. 4. It agrees enthusiastically. 5. I ask it to challenge my assumptions. 6. It tells me my assumptions are excellent. 7. I ship. No components. No auto layout. No tedious conversations with engineers. Last week, I designed an entire enterprise SaaS platform in 11 minutes. I don’t know what it does, but AI assures me it’s a paradigm shift. Some designers worry that AI will replace them. I think they’re asking the wrong question. Why are we still designing things at all? My productivity has increased by 4,700%, my design debt has fallen to zero, and I haven’t had an original thought since March. Would you like me to make this post sound more authentic and personal, or keep the confident, thought-leadership tone?
8
Adobe will let outside AI agents work inside Photoshop, Illustrator, Premiere and After Effects, private beta early next year. The app stops being where the work happens and becomes something your agent opens for you. Curious what happens to all the menus nobody ever learned.
7
What's the best design feedback you ever got? The kind you still hear in your head every time you open a new file.
1
1
42