This week is quietly packed with structural questions for front-end work. There is a fresh tour of imperative versus declarative programming with restaurant tables and self-driving cars, a deep dive into why you still cannot truly cancel a JavaScript promise, and a remarkable tale of removing 15,000 circular dependencies across over 1,000 projects.
On the HTML and CSS side, we have a lovely piece on CSS as a state machine, a guide to building fluid UIs without traditional breakpoints and a reminder that !important is rarely the best answer when cascade layers exist. Google’s new spam policy for back button hijacking should make the web a bit less hostile, and HDR images abused for marketing will definitely make it brighter on modern screens. Rounding things out are a peek into GitHub’s fake-star economy, a discussion of why AI still struggles with the front end, Addy Osmani’s Agentic Engine Optimisation, and the Roblox-plus-AI chain that helped bring down Vercel, along with my own story of four caches and one very stubborn bug. Happy reading!
Another attempt to explain the difference between these two programming styles, but this one relies on concrete examples, including the classic restaurant and automatic car metaphors. If you struggle to explain why React is called declarative without drifting into buzzwords, this is a nice refresher to keep in your back pocket
https://fireship.dev/c/react/imperative-vs-declarative
You cannot strictly cancel a promise once it is created, but in practice, you can often design your code so that it behaves as if you did, and this post goes deep into how and why.
https://www.inngest.com/blog/hanging-promises-for-control-flow
A pretty wild story about what it takes to unwind more than fifteen thousand circular dependencies spread across more than a thousand projects in under a year. If you have ever sighed at your own dependency graph, this will make you feel better and maybe give you a strategy or two for tackling technical debt at scale.
https://stefanhaas.xyz/article/15k-circular-dependencies/
I am still not entirely convinced, but the idea is intriguing. The proposal is to organise a large monorepo vertically around product areas and domains, with clear boundaries between those “verticals” so teams can own their slice without losing shared building blocks, which often become their own verticals rather than a single global UI library.
https://tkdodo.eu/blog/the-vertical-codebase
A delightful post on treating CSS as a state machine for UI behaviour. Using the usual hidden checkbox trick for light and dark themes is just the starting point here.
https://css-tricks.com/the-radio-state-machine/
A reminder that fluid design is not just a 2010s trend but a very practical way to build layouts that adapt continuously instead of jumping between fixed breakpoints. With container queries, clamp(), calc() and a few more modern CSS tools, the post shows how to rethink layout so that it responds to context rather than a handful of hard-coded screen widths.
https://frontendmasters.com/blog/building-a-ui-without-breakpoints/
This post explores why you should avoid reaching for !important and how cascade layers, specificity and better architecture can usually solve the same problems more cleanly. It is especially useful if you work with frameworks and third party libraries and want to keep styles organised instead of fighting a losing battle with overrides.
https://css-tricks.com/alternatives-to-the-important-keyword/
Google Search is updating its spam policy so that pages using “back button hijacking” are treated as malicious and can be penalised in search results.
https://developers.google.com/search/blog/2026/04/back-button-hijacking
This post surprised me a little: it shows how HDR images can be abused for marketing by making on screen visuals pop in a way that SDR images simply cannot. The catch is that the effect is only visible on modern, high-end displays such as recent MacBook Pros, iPhones, Pixels and similar, so you have to decide whether that is your audience
https://tn1ck.com/blog/abuse-hdr-images-for-marketing
When startup valuations and “unicorn” momentum lean on GitHub star counts, stars start to look a lot like a tradable commodity. This investigation digs into how fake stars are bought and sold, why the ecosystem has not really cracked down on it yet and what that means for using stars as a proxy for quality.
https://awesomeagents.ai/news/github-fake-stars-investigation/
This is a sharp look at where AI falls down on the front end, from layout maths to accessibility to bespoke interactions. I share much of the scepticism, but in more structured setups, I have also found AI can pull its weight, so I tend to see it as a capable assistant rather than a replacement.
https://nerdy.dev/why-ai-sucks-at-front-end
Here it is: Agentic Engine Optimisation, the idea that you now have to think about how your product appears not just to humans but also to non-human agents such as AI systems and automation tools. It is SEO-like in spirit but uses different tactics and aims, focusing on being legible and useful to agents who never see your actual UI.
https://addyosmani.com/blog/agentic-engine-optimization/
This is slightly older news now, but what a story: a Roblox cheat installed on a work laptop, a compromised AI “office suite”, and suddenly Vercel’s platform is in serious trouble. It is a fascinating chain of failures and trust assumptions that is well worth a read.
Maybe it wasn't such a bad idea to move my blog away from Vercel. At least it reduced the complexity of juggling multiple platforms, although that did not stop me from running into some issues, which is exactly what my latest post is about.