Some lighter "non-hype" reflections

Again, I posted this on LinkedIn and wanted to share it here as well.

Every year I go to North America for a month or so in Feb because of conferences.

I also take this time out to reflect on the previous year, take a step back from my life in the UK and recalibrate.

This time I was in beautiful British Columbia 🇨🇦, and once I kept to myself like Tony Stark in a bunker, I took the opportunity to play, build and revisit things I’d bookmarked over the past few months.

Here are some of my lighter, non-hype (and maybe slightly controversial) thoughts during that time away:

“Vibe coded” websites can look good… when you know what you’re doing. Vibe coding does pose its problems, but for the purpose of this post, I’ll put that aside. I’ve been working on my own websites (using Claude Code) and it’s clear that model outputs have improved over the past year. Designs are more aesthetically pleasing, although without strong direction they likely converge towards the same design patterns which probably explains why so many sites look similar.

“Open-source” AI still isn’t really open-source. You’d think by now we’d have an influx of fully open models, and while there are a few, what we mostly have are open-weight models, which isn’t quite the same thing.

Small Language Models are being slept on tbh. On-device, lightweight AI just makes sense for a lot of practical use cases.

Most static LLM benchmarks are useless. This isn’t a new discovery, but it’s becoming clearer that real-world evaluation is what truly matters. I like what the team at Arena. AI are doing here, using community comparisons and feedback to build leaderboards for real-world use....check them out.

Consumer “self-driving” cars are starting to feel more real. I rented a Tesla Model 3 with FSD for a month (FSD's not available in the UK) and was genuinely surprised at how well it handled real-world driving. Curious to see how BYD’s "God’s Eye" tech compares.

AGI is still a useless term. It’s ill-defined and everyone means something different by it. It’s also become a kind of red herring, distracting folks from the more immediate concerns.

It’s become increasingly clear that scaling LLMs alone isn’t sufficient in the quest for general intelligence (I know, I know 😬) and so we now have alternative approaches (world model approaches JEPA etc.) gaining attention (and funding).


There is nothing groundbreaking in this post, but merely a few non-hype observations from my time away.


(Pictured below with my cousin Neeta Nagra Ed.D, MBA and her dog Koko, at Tunnels Bluff, Lion’s Bay, BC 🇨🇦. Awesome hike, much recommended, the pics don’t do it justice).