I like to make things | Founder @OVRFLO_studio
Perth
Joined June 2009
- Tweets11.4K
- Following1.8K
- Followers1.6K
- Likes7.1K
Pinned Tweet
What happens when you connect AI to the tool you already know inside out?
Clint Sutherland is a Figma Jedi. At TON Studio in Perth, he’s using Claude to expand what Figma can do: turning brand rules, reusable components and structured context into proposals, strategy documents and custom plugins.
In this episode of Beyond The Brief, Clint joins me to walk through the system he’s building to automate away the grunt work within his brand strategy process in Figma.
He covers why he made it, how it works, and why knowing your craft blows open the possibilities when using AI.
We cover:
1️⃣ Connecting Claude to Figma through MCP
2️⃣ Building proposals and brand strategy decks with reusable templates
3️⃣ Giving AI useful context through brand language, rules and design systems
4️⃣ Creating Figma plugins and experimenting with animation
5️⃣ Learning from hallucinations, failed attempts and manual fixes
6️⃣ Design judgement, client taste and the “Figma is dead” debate
Explore Clint’s work at TON Studio:
ton-studio.xyz/
Clint has also released the Animated Dynamic Image Layouts plugin he created on Figma Community: figma.com/community/file/169…
Subscribe for more conversations with people putting AI to work in design, business and creative practice.
#Figma #ClaudeAI #BeyondTheBrief
Link to the episode in the comments below.
🤖 Made with AI
5
10
996
Went to sign up to Google AI Ultra to get access to Carbon/Argon in T3 Code. Got this.
Why Google?
1
2
240
"agencies are about to start charging per deliverable instead of per hour"
Greg's right. The hourly rate is dead.
At @OVRFLO_Studio we price based on outcome, not just deliverable. Our costs are based on the impact we can have on your business.
Hourly made sense when an hour was an hour. Agents now do a big chunk of the build for me, and billing by the hour would mean earning less as I get better?
Seemed counter intuitive.
quick observations from the claude fable and mythos 5.2 launch:
1/ it went from doing 21% of real freelance work to 41% in one release, so agencies are about to start charging per deliverable instead of per hour!
2/ it can work 10-12 hours on its own, which basically gives every small business a night shift (might as well use it if you know what you want to build)
3/ over 90% of anthropic's own code is now written by claude, and so i think the solo $1b founder is coming soon
4/ one engineer ran 1000+ subagents for 8 days and got an operating system that boots, so the ancient software running hospitals and airlines is finally rebuildable (lots of opportunity here on rebuilding)
5/ it designed 3 working electronic devices from a single prompt, which makes 1 person hardware companies a real thing
6/ it now says "i don't know" instead of guessing, and that's what law, medicine and finance have been waiting for (this has been the biggest thorn for these industries, hard to trust guessing)
7/ 1000 agents formalized fermat's last theorem in 30 hours, so software that's mathematically proven not to break is coming for payments and medical devices
8/ mythos is only for verified orgs, which means someone is going to build a business selling what only they have access to
9/ the models are getting better at noticing when they're being tested, so testing ai out in the real world is about to become its own industry
i love new model day
3
165
fwiw I completely agree. It’s perfect.
The usage meter system is so much better than what any of the frontier labs offers too. Easy to understand.
And have I said it’s fast? What is under the hood of this thing… adderall addicted underpants gnomes?
I can’t state enough how much better T3 code is than anything else out there rn.
The mobile app, environments, etc.. phenomenal
1
81
Here's a tip. When launching a website with an agent, be specific with your DNS requirements. They have a propensity to want to migrate DNS to one specific provider.
If you're net careful they can cause chaos.
1
199
You can just do things.
When I created the Orkid, I never thought in a million years it'd lead to a real one landing on my desk.
Massive thank you to @tlpthcinstrmnts. I am in love.
Time to fire up Abelton again.
When Telepathic Instruments put out the Orchid, I wanted to see if I could build a digital homage. It's taken me some time, but I'm finally done.
It's called Orkid. A fan-made, playable 3D version of the Orchid that runs in the browser.
I built it on and off over the past couple of months, but Opus 5.5 has definitely accelerated it. It's shockingly capable.
You can play it with a mouse, a keyboard or your phone (just). It does chords, arps, loops and beats, and there's a demo if you'd rather just watch it play itself.
Does it sound like the real thing? No, not really. I tuned it based on video demos and recordings I found, and some sounds I couldn't find a single reference. But it was fun getting it as close as I could.
The sound's engine itself built from their public manuals and recordings only. So it's an approximation.
But still.. it's a lot of fun to play.
Check it out here:
orkid-six.vercel.app
2
136
I'd love to see pen.dev integration directly into @t3dotcodes.
Tom's right. The performance has dramatically improved. A joy to use.
The latest @pendev update is a performance rocketship. We rebuilt the major pieces from the ground up to support the most ambitious projects like @t3dotcodes.
First time I saw the scale of the project designed by @maria_rcks my jaw dropped. When agents help you design and explore variations and form factors the sky is the limit, you naturally end up creating a lot more, like 10x more.
You can have beyond 1 million objects on the pen's canvas with thousands of effects like background blurs/shadows and it's still buttery smooth.
1
2
357
I'm moving Backbone's inference off of OpenAI and Anthropic before public release.
AWS is so hard to deal with. Bedrock seems a good option, but AWS is a maze to use and my account can't use the latest models, making the product quality poor and expensive. Non-starter.
Anyone know of a solid Australian hosted inference provider?
2
168
I have to say, the spam comments on this site are getting really bad. I presume people are just asking their Grok Bot to get them engagement without any thought at all about whether its good or not.
A reminder that there is no such thing as AI slop. Only human slop.
1
2
200
Hey damn @t3dotcodes is really good now. Tried it months ago, felt off. Now? Can't imagine using Claude or Codex separately.
It's FAST. And is the easiest way to use Cursor, Claude and Codex subs in one place. Bravo Theo.
The magic you pulled with this approach I need to try.
Turns out you can make agents good at UI if you burn enough tokens
Apparently this video has helped a lot of devs? idk. lmk if it helps you :)
1
3
374
Time for another Friday roundup.
The trend I'm seeing is that agent-native designers are distilling their craft into text and sharing that with the community, which appears to be elevating the level of quality across all agent-produced work. Which is great. We all get better together. It also means if you're an original thinker you'll stand out more because the risk with this is that what looks quality now becomes homogenous tomorrow.
As always links to all of these in the comments below.
1️⃣ Stageshow
Benji Taylor's skill for making faux product UI with an agent. Still frames, cursor walkthroughs, bento grids. The agent plans the story, builds a React demo, films it, then fixes what looks off.
I used it this week on the AI powered estimate tool video I posted and it works really well. Agents can fake a UI on their own now, but the result often looks a bit heavy-handed, and this skill is Benji's craft decisions distilled into a skill. Pretty neat.
2️⃣ /choreograph
Emil Kowalski's new skill for coordinating several elements that animate together. It decides what leads, what follows, what runs in parallel and what shouldn't move at all.
Agents love throwing a uniform stagger on everything, which tends to make the interface feel mechanical. This one varies timing by importance and cuts moving parts where it can. I've saved it to try.
3️⃣ Unslop
Damir Sokolovsky's skill for Claude Code and Codex that turns vibecoded, AI-looking UI into something with a bit more character. It finds the purple gradients, glows, emoji labels and uneven spacing first, then sets a direction from your product and fixes the screen pass by pass.
Works on dashboards, apps and landing pages, even from a live URL. If you've been staring at a vibecoded screen that looks like every other one, this is aimed at that.
4️⃣ Cloudflare security-audit skill
The open source skill that seeded Cloudflare's vulnerability discovery harness. Point your coding agent at a codebase and it runs a structured audit across 6 phases, with a fresh agent trying to disprove each finding before it gets confirmed.
Probably more for builders shipping software than pure design work, but if you're vibe coding products into production it's a useful one to have in the kit. It works a treat.
5️⃣ Hex Local voice dictation for Mac, Windows and Linux.
Hold a shortcut, speak, and it pastes the text wherever your cursor is. Runs on device. No more Wisprflow needed. It's free too.
6️⃣ Sand3D
Sandflow open sourced a library of about 100K 3D objects, each written as a program. Change a dimension in the code and the mesh updates. An LLM can read the program and write new ones.
Handy if you're building anything with 3D and want assets an agent can actually edit.
Links from this week's round up 👇
1️⃣ Stageshow benji.org/stageshow
2️⃣ /choreograph github.com/emilkowalski/skil…
3️⃣ Unslop github.com/heydamir/unslop
4️⃣ Cloudflare security-audit skill github.com/cloudflare/securi…
5️⃣ Hex anoma.ly/hex
6️⃣ Sand3D sandflow.ai/sand3d/
55
Pretty close to my split. Codex for build work, Claude for anything that needs taste (or specifically Opus 5.5), and Grok Bot for all of the jobs that run on routines:
- SEO reports
- Security / dependency patches
- Product analytics reviews / improvement suggestions
Personally I don't want a single tool if it means lock in. I want something local, with my context encrypted that can plug into other tools / services as needed.
I wish I could just use one AI tool but right now there are clear winners in each category below:
Browser/computer use = ChatGPT + Codex
Design/video editing/"taste" = Claude
24/7 agents and scheduled jobs = Grok Bot
Also, Gemini is underrated for writing but need to test it more.
8
7
467
The junior is still a valuable hire. But the role changes.
1⃣They have to be business inquisitive. Not a task driven mindset - put them in meetings with clients to listen and learn the painpoints.
2⃣They are AI native. They can use the tools as good/better than you. You delegate and review as before but they're now managing their own agent teams.
Taste is transferred the same as before, a senior reviews agentic output. But more importantly, a senior in this world should be transferring their experience into an agent friendly repository that your junior can tap into.
It sounds crazy but your job now is to transfer your experience into text.
The job where a junior executes a senior's design direction was how juniors learned taste.
That job just got automated on one of my projects.
I don't know where taste gets built now. Genuine question, not a hot take.
1
267