Portfolio 2.0, soft launched
Okay so I shipped a rewrite of this site over the weekend, and most of the front-end design was done by Claude Designer. Pretty stoked with it.
I try to stay across all the latest AI tools, mostly because I'd rather jump headfirst into something and try to break it than read the documentation first. Designer had been on my radar for a while, I'd seen the marketing, never tried it. Then a YouTube video popped up of someone pushing Claude 4.7 through Designer end to end, and I figured what's $24 against actually hiring a web dev. Re-upped my Claude sub, in we went.
Quick context. The v1 of this site was a Flask/Jinja thing I'd iterated on with Cursor over the back half of last year, mostly using composer models or earlier Sonnets, and occasionally falling back to ChatGPT 3.5/4 for the harder bits. It worked. It also basically looked like a markdown file with some colours, which, to be fair, is exactly what it was. No React, no real animation, no front-end grammar to speak of. The visual language I'd been thoughtful about, but the actual implementation was nowhere near anything you'd call best practice, and the agents I had at the time weren't really teaching me what I didn't know I didn't know. You know how it goes.
First thing Designer did when I imported the project was prompt me to codify the whole design system. Which I didn't expect to find as interesting as I did. Colour ramps, type scale, spacing, motion tokens, all the stuff that had been kicking around in my head pulled out into one place I could actually push back at. Made the conversations afterwards heaps sharper.
Then I asked for three concepts and it built me three landing pages. I picked the bits I liked from each, mashed them together, handed the whole thing over to Claude Code for the local implementation, and just kept pushing it. More features, more polish, asking what best practice was on whatever animation or accessibility thing I was on at the time, doing my own web research alongside. The interesting bit wasn't that the models could do this. It was the thoughtfulness of how Designer scaffolds the work, and then how Code picks up the baton from there.
Which is the same thing I was banging on about in the last post. The models are fine. It's the implementation that's actually bringing the ROI back now.
Same energy in some other stuff I've been pulling at lately. Been optimising my Hermes gateway, watching where tokens get spent, asking whether my agents are getting proper deep personalisation context or whether I'm just brute-forcing big payloads at them and hoping. I also built another LOTR agent called Elrond whose whole job is to go research online, sweep for best practices, and look for efficiencies across the rest of the setup. An agent whose entire purpose is to make the other agents better. Which is sort of recursive and sort of just neat.
A couple of weeks ago Anthropic dropped Claude for Small Business, which is the same thing again. About fifteen agentic workflows hooked into QuickBooks, PayPal, HubSpot, Canva, Docusign, all the Google and Microsoft stuff. Invoice chasing, payroll planning, month-end close, lead triage. Take something a small business owner had been doing badly with the wrong tools, point a thoughtful implementation at it, give them back the hours. Each one of these on its own is small. The rate it's all compounding at is the weird part.
Anyway. v2 is live. Looks like a website now instead of a styled README. Same content, same markdown source of truth, same Flask deploy pipeline, the layer that changed is the one between the data and your eyeballs.
Soft launch, baby.
← kipjordan.com