A calmer professional feed

The shortcomings of GenAI are pretty well rehearsed now: hallucination (no, that's just wrong), counting (how many Rs in strawberry?), sycophancy, and bias.

The thing I've found most challenging with AI-driven development is not building full stack, however complex that might be. It's actually the one area that appears to be simplest. I'm pretty ok with aesthetics, having spent several years at the start of my working life as a photographer. But UI (user interface) is unbelievably hard to get right, and I generally find AI pretty poor at it. There are plugins and skills to help, but they all tend towards a bland, Inter font, this-was-vibe-coded look. Nothing wrong with Inter (I use it a lot) but you generally know when a platform has had its inter

When I build with AI I generally have low expectations that it will ever get things right the first time. But here's how I accelerate the debugging process. So much of the success around AI assisted development lies in being able to articulate the precise problem so that the coding agents can fix it. To do this, these 3 debugging tools become your friends:

There is a story about a teacher who went to work in a school with no smartphones, no computers, and no digital devices of any kind.

Sharing with people here who have an interest in speaking to children about AI - a new 3-part booklet series called 'Talking about AI'. This was a collaboration between STEM Learning and Voice 21, which I authored. Booklets 1 and 2 are in the wild, and you can access a free copy here: https://community.stem.org.uk/blogs/shameel-khan/2026/04/20/what-can-oracy-and-ai-do-for-education

Ok, well this one properly surprised me. Egypt has just announced that from September 2026, Japanese will be taught as a second foreign language in its preparatory schools. Which on the face of it sounds about as exciting as a wet Wednesday in Hull (no offence Hull 🤭) . Stick with me though, because what is actually going on here is genuinely a bit mad and quite important.

One thing I run into when creating lesson resources, slides, thumbnails or visual guides is finding the “perfect” image online… only to realise it’s tiny, blurry or heavily compressed. Hence why I do enjoy Ai Image generation from time to time as it help with SOME needs. A lot of older images I have lying around or find online especially such as screenshots and low-res graphics just don’t hold up anymore when you place them onto a modern presentation or try to crop into them. Especially when projecting onto a large screen in class, those blurry edges and pixelated details become very noticeabl

Every so often someone names a problem so cleanly that it sharpens thinking you already had. @Carl Morris did that this week, writing about why good EdTech tools quietly fall out of use, and landing on the idea that the only product surface that really matters is the first thirty seconds of a teacher's Monday morning. His post points at an uncertainty that sits underneath a lot of planning in schools, and it made me reflect on why we built our system the way we did.

Security isn't always the most exciting topic, but it is one of the most important. Often it's something we as users of platform or teacher don't think about and we collectively all just assume our School admin has set it up security.

This is an invitation (no, don't worry, there's no hidden agenda, sales, or advertisements in this). I vibe-coded something that keeps growing in size (not just the AI coding without disregard). To vibe-code is to direct, cajole, beg at the foot of a Gen AI (ChatGPT Codex, Claude Code) to build something. DrawSplat.org is the creation and I started it because I always wanted these tools accessible for free when I was in the classroom. Now, they are. Feel free to share it widely. Ok, ok, I also wanted to recreate some games from my teens, like Castles and Catapults, SplatBall (which some might

My first thought was, "Who the hell trusts Steve Ballmer these days?" My second thought was that education has been a real issue in America for a number of decades now. Finally, I thought, data does not equal facts. Data can be based on facts. Data can be factual. There is rarely a time that data equals facts. (I say rarely because I am sure that someone can point out something where this is true - I can't think of an example myself.) Ballmer's USAFacts is a great platform for pulling data, I am sure. His videos - how I was introduced to this platform - are presented as him just spitting "fact

Quietly, I have been writing a book about AI, data governance, and higher education. I am trying to help people think about what they need to address as the shift to AI agents is happening quickly, while also acknowledging that many have not done much beyond saying “cheating is bad” in a banal way.

The more I think about it, the more certain I become: we are at a crossroads in education. However, I don`t think it is time for despair. I think, instead, it is time for empowerment.

How I created my weekly #PubMed /MEDLINE agent that emails recent published papers on a specific topic: Step 1: Write your search terms relevant to a specific topic (e.g. CAD, obesity etc), identify filters (such as adults, humans, English language studies only), create and paste the script in #Google Apps Script, run it once to test, then set a weekly timer. Step 2: Use #Claude Console/Anthropic API or #OpenAI API (separate from Claude chat) to summarize raw abstracts and remove duplicates/repeats. Step 3: Rank the emailed list based on your preference (study design, journal, etc.). Create a