AI Workflow: I'm a designer by nature. I'll always be designing, no matter what tool I'm in or where I start.
Since the dawn of generative AI discussions, I've seen new tool suggestions (usually AI based) for design at least a few times a week — in my Instagram, threads and X (read: Twitter) feeds, across the design podcasts I frequent, in my work Slack channels — anywhere they can get me. I've seen all sorts of tools — from ones that tackle different parts of the design workflow to those trying to augment the whole thing (check out designtools.fyi for the rundown). Like most designers I know, I've found navigating this new age to be complex. In so many ways, it's a rewarding experience to try new tools — develop your thinking and doing things that were previously too difficult, but it's also hard to master skills and cram every tool into a workflow that is also constantly evolving.
I've been on this grind for a while now — I've picked up and dropped many tools and I might just pick up those tools again in the future. In 2024 I had tried out Cursor, v0, Replit and Lovable and I couldn't seem to slot the tools into my workflow well (they were also not as sophisticated as they are now!), but fast-forward to 2025/6, I start all my thinking in Cursor and do most of my prototyping there too — after a serious stint with Lovable that I've since paused.
Point being, tools are living and changing mechanisms that are applied to the design workflow. Sometimes, we will use these tools to develop business value — to debias research, explore multiple concepts, ship faster. Other times, we will use these tools just to avoid doing tedious tasks (I'm looking at you, Dovetail tagging) or to create joy in our work (I'm looking at you, fun motion interactions). Regardless, I, as a designer, am here to create and build products and experiences to life — I will continue doing that no matter what tool I'm in.
Here's how my day-to-day looks right now (dated, May 2026)
Discovery
My workflows around conducting discovery are the ones that have changed the least. Research to me is something that is uniquely human — it's in the slight shift of tone in the voice, eyes darting with distraction, confusion hidden by bravado that makes research nuanced. Sometimes, I believe the human bias / feeling instinct here pushes us to the right decisions. As such, my toolkit and workflows to collect research have stayed similar, but preparing for research, accelerating to insights and converting research into artefacts have been completely turned on their head.
- Research planning: I have developed skills in Cursor to create a first iteration for interviews and surveys — they follow my general structure for interviews, and plug in some of my interviewing style. In turn, they can also audit my interview guide to de-risk against leading questions. All of this ready for me to review and inject my own creative thinking.
- Conducting research: This has stayed largely the same. I am a strong believer that we need to still test with humans, not synthetic AI created personalities (although these can be great thinking partners). I run my quant analysis wherever I can — from Qualtrics to Lyssna (then displayed interactively in Flourish) and my qual interviews with the tried and tested Zoom, Dovetail, Miro combination.
- Analysing research: The big change for me is here — turning data into tangible insights. With raw research inputs, I can plug them back into my Cursor to derive patterns across multiple pieces of research (including older research), to quantify qualitative insights, to scan across qualitative and quantitative research together and debias myself. Insights then are stored in a .md file, ready for re-use — to bring personas to life, to write a PRD, to query again.
- Creating discovery artefacts: Processes that would have taken days aligning research across multiple artefacts can now be done in hours. Again, I have a collection of skills here to develop the insights and thinking with the product details (more on that here) — skills to draw out the journey map, develop personas (or behavioural archetypes), write problem statements etc... All of this ends in a PRD (this PRD skill is quick and thorough enough to start with: awesome-copilot/prd)
Ideate & Design
Whilst my toolstack hasn't changed much here, I get to an answer a lot quicker, and my outputs are more consistent. Whilst I still tend to start on some canvas (paper or Figma), I find myself designing less screens and moving faster to code-based prototypes to get a better sense of what feels right.
- Creating multiple design directions: This bit still feels best on canvas for me. I might start with Cursor to help me develop a few directions and ideas, but seeing them visually is a lot clearer for me. As of now, I still think I'm better at getting an idea down by hand vs. prompting an agent 10 times to place the box more to the left. This will probably change the minute they let me control code like a canvas.
- Designing — audit, copy: As someone who started life in research (not UI), I've had to train my eye to better self-critique and audit my work (spacing, typography, copy, consistency...). Whilst my own judgement has gotten much sharper, I now have a few self-written skills for the products I work on which have been trained on a database of visual and language content. A pair of second eyes to check that we always place 24x24px icon buttons 8px apart from each other — not 4px!
- Applying design tokens: Agents in canvas! One of my favourite things with the Figma MCP is that I now have a personalised skill to apply design tokens across my design. So if I'm being a bit scrappy with my initial design and hard-coding tokens, I now just run my skill and have my Cursor fix everything whilst I get started on the next bit.
Prototyping in Code
This is where the fun is. This bit has doubled my work satisfaction (dare I say improved my life satisfaction). With all that time I freed up in discovery, problem solving and ideation, I sink all my time (and teeth) here. Perfecting a interaction, checking if something that I designed actually feels right, comparing multiple versions at once — these are all tasks that were hard to do well in Figma prototyping.
My current set up here, is to show my agent the design (usually 1/2 screens) in Figma, and the CSS behind it and prompt it to generate the UI. I can then go in and tweak, add hover interactions, define flows, error states and edge cases. Presenting becomes infinitely easier, and sharing via GitHub or Vercel.
Small note on Lovable. I started using Lovable seriously after watching how Atlassian prototypes in code (watch here). I sunk all my spare time into making it work — it became an obsession for 2 weeks (weekends included) where I coded the whole design system so our PMs could quickly build prototypes that also used our design system.
Whilst it was definitely a useful exercise, I find that Lovable is not as capable or advanced for my current needs. Since most of my workflows and skills live in Cursor, I've naturally moved to that as my main prototyping tool — and paused Lovable for now.
Bettering Craft
The simplified view of the design workflow above covers ~85% of my day to day, the other 15% I try to ensure I am sharpening my critical eye, developing myself into the designer I want to be and staying on top of the latest and greatest.
You're probably getting sick of hearing me talking about all the skills I've written in Cursor (14 times to be precise), but this is where a previously messy and imprecise research process has become a lot smarter. Here, I have a daily rundown skill which writes out the latest from a few defined sources I have, but also trawls the internet for anything new. It churns out a report on the latest in AI x Design, and B2B software design. Every report gets stored in my Obsidian to avoid repeating information.