No idea. The eye test of this moderately-informed but not expert person suggests that Arsenal was far lower than that. So it makes me start to dismiss the visualization's relevance. 🤷🏻‍♂️

I propose that this meter isn't measuring the right things to answer the question "Who deserved to win?"

Wow. This article, while situated in a far more serious context with grave consequences - maps cleanly to risks and realities I see happening on teams I work in making design systems with AI.

You can choose to see a different future that isn't across-competition collapse.

Observation: White movement helping Saka be more dynamic?

Configuration Collapse As Figma Slots Push Composability, Be Ready to Drop Props

open.substack.com/pub/nathanac...

Well that is about as favorable a Champions League bracket side as Arsenal could have ever imagined.

I haven’t been this apoplectic about a performance in over a decade. Everything everyone says about mentality is, on today’s measure, deserved.

What I saw emerging before and taking hold after ManU’s first goal was sluggishness. When all the “random” balls start dropping at an opponent’s feet. How much of this is accumulated fatigue and do you rotate the ENTIRE side including bench for the CL?

SmashingConf in-person Conferences 2026 — Friendly, inclusive & fun events for web designers and developers. Looking for a practical conference about web design and development? We organize three hybrid conferences in 2026: Amsterdam, Freiburg, and Antwerp. Get your ticket now! Front-end, design, UX and…

SmashingConf is back for 2026! We’re bringing thoughtfully curated conferences to Amsterdam 🇳🇱 Freiburg 🇩🇪 and Antwerp 🇧🇪.

- Actionable talks on design systems, UX, accessibility, CSS and more
- Hands-on workshops
- Side events & networking that actually help you connect
smashingconf.com/conferences/

Either via plugin/AI or copy the layer, select it's parent, use Figma's convenient select all matching layers, paste.

"Code Only" Props in Figma Expanding component API shape in design assets

open.substack.com/pub/nathanac...

I've generally been using VS Code/Copilot in my work with:
- Claude Sonnet 4.5 for most spec'ed features
- GPT 5.2 for heavy refactors
- GPT 5 mini for basic things and
- mixing in Gemini and Claude Opus for cross-referencing/parallel takes.

Why are LLM plans so ridiculously conservative in time estimates? "Your POC will take 1 week and implementation 13 weeks."

It just finished the POC (with added scope) in 1 hour and says implementation is now 3 days.

Like, is it taunting me with "Without LLM" estimates?

So my now for a Saturday morning instructing an LLM to "Implement REST API conversion a Figma library file into my Anova data model based on Github spec kit's plan and tasks.", going to take a shower, coming back and realizing I didn't autoapprove the CLI commands and it got stuck on step 1.

I started using GPT 5.2 (instead of Claude 4.5 or similar) for plugin development yesterday. Today, I noticed the havoc this wreaked on my machine's energy consumption. Holy camole.

In my travels, I have rarely seen semantic spacing, when I do it’s for a few specialized cases rather than for day-to-day composition, and that lack of semantic spacing makes me sad.

I hate penalty shootouts in football. I would much rather see a “Remove a player for the pitch every three minutes” and see what happens

When do we have the Gyokeres conversation?

Patience is trying. I get that he’s a runner, but he does so much else so poorly, such as receiving aerial service. It’s just… bad.

Like my cousin’s fav windup whisper to an opponent: “why don’t your teammates pass to you?”

Stats like this make it obvious: when are we going start having the Victor G goodness-of-fit conversation? The eye test has suggested we do so since his very first appearance.

So what rate is an acceptable rate? (The answer is “When the rate is better than the human rate” although society won’t behave that way, such as with autonomous vehicles)

I agree. Although, also, as if humans are reliable and trustable in copious processes where AI isn't in the loop.

I also think trust and reliability depend on task/scenario and vary in significance relative to opportunity, risk and value. Me coding a plugin ≠ AI confirming nuclear launch codes.

Do you reframe things? "Trust but verify instead of don't trust?"

As I code, I find that AI makes mistakes, muddles things, but also learns. It can learn fast. Kinda like a junior coworker learning on the job. Despite the challenges, undoubtedly my output is still faster and better.

Love the article. RVP was for a bit, but he got out while can? Don’t give that sucker no statue. Give him guts!

Thus my favorite was long Ramsey. My prediction (after that injury)? Pain. Eventually, he was a fine champion, but his time was past. We wish him luck.

"Clock ittt, Dads an aura blud"

So, I won "Best Plugin" for the Figma Specs plugin at the @zeroheight.com design system awards last night, and am excited!

I told my family, and two my kids reacted this way. I'm now also slightly confused.

I am seriously tripping out. I just experienced two agentic "party mode" rounds of a team discussing an ADR of spec data schema updates for absolute positioning and Figma's Grid.

Their perspectives were so thoughtful, the decision correct. Is this my (short lived) future before I'm replaced?

Can you elaborate?

Temptations/Four Tops (1st)
Rush
Barenaked Ladies (in a strip mall in front of the Record Exchange)
Jonatha Brooke
Max Richter

Yeah seriously. A high school volleyball game has four refs. Basketball, three. Baseball, four. American football, so many more. And yet this game, so expansive and so many players, with one.

Now, I do need to get around to implementing slot prop evaluation (`body` 👀)