The AI Roundup - February 2026 - Part 1
Your fortnightly look at what’s happening in AI, from Scott Quilter at Techosaurus
Welcome to the very first edition of The AI Roundup. This is something new I’m trying out: a fortnightly write-up of the biggest AI stories, what they actually mean in plain English, and why you should (or shouldn’t) care.
If you listen to my podcast Prompt Fiction, you’ll know that every couple of weeks Reece and I sit down and try to make sense of everything that’s happened in the AI world. Some of what I cover here will come from those conversations, some of it will be things I’ve spotted in between, and all of it will be written the way I’d explain it to you over a coffee rather than how a tech journalist would write it up.
So here we go. First edition. Let’s see how this goes.
The Big Stories
You Can Now Buy Insurance Through ChatGPT
OpenAI continue their push to become the front door to the internet. The latest move? Insurance shopping, directly inside ChatGPT. Insurify launched what they claim is the first insurance app in OpenAI’s app directory, allowing users to compare car insurance quotes through a natural conversation. Experian followed shortly after with their own Insurance Marketplace app, covering over 37 carriers.
The pitch is that instead of filling out endless comparison site forms, you just have a chat. Tell it about your car, your driving history, your postcode, and it goes off and finds deals for you. It feeds back what it’s understood before searching, so you get a chance to check it hasn’t hallucinated your four-bed semi into a twelve-bedroom mansion.
And that was exactly my concern when Reece and I discussed this on the podcast. I don’t think a week goes past without AI getting something wrong in at least one of my conversations. If it’s making up facts about my driving licence or my address, that’s a problem when you’re signing off on a legally binding insurance policy. The takeaway? This is clever and it’s convenient, but double-check everything before you commit. That’s just good AI hygiene.
Sources: Experian Press Release (12 Feb 2026), Insurify via PR Newswire (10 Feb 2026)
Udemy Courses Land Inside ChatGPT
This one genuinely excited me. Udemy, the online learning platform with over 290,000 courses from 90,000 instructors, has partnered with OpenAI to embed their content directly into ChatGPT. That means when you’re having a conversation and ChatGPT detects a learning opportunity, it can proactively surface relevant Udemy courses.
But here’s what really got me going: it’s not just about finding a course. It’s about what happens when you combine proper educational content with AI’s ability to differentiate. Imagine you’re doing a course on data analysis but you’re not quite getting it. You could say to AI, “I don’t understand this. I’m really into WWE,” and the AI could go, “Right, let me teach you this concept using wrestling as the analogy.” Udemy don’t have the resources to write their courses for a thousand different audiences. AI can do that on the fly.
I went on a proper tangent about AI in education on the podcast this fortnight (I’ve written a longer blog post on it if you’re interested), but the short version is: I think this integration is a glimpse of something much bigger. If AI can personalise the learning experience while being grounded in proper, expert-curated course material, that’s a genuine step change in how we learn.
Sources: Udemy Press Release (11 Feb 2026), PYMNTS.com (11 Feb 2026)
OpenAI’s Codex Spark and the Cerebras Chip
This is one of those stories that sounds really technical but actually matters to everybody. OpenAI released GPT-5.3-Codex-Spark, a lighter, faster version of their coding assistant. The big deal? It doesn’t run on NVIDIA chips. It runs on Cerebras' Wafer Scale Engine 3, a single silicon wafer packed with 900,000 AI cores and 44GB of on-chip memory, delivering over 1,000 tokens per second.
Most AI today runs on NVIDIA graphics processors, which were originally designed to make video games look pretty. They work, but they’re power-hungry. I used the car analogy on the podcast: NVIDIA chips are like American V8 engines. Big, powerful, fuel-guzzling. Cerebras is more like a Japanese Suzuki with 750cc. Less doesn’t mean worse. It just means more efficient.
If you can get dramatically faster performance for less energy and less cost, the environmental argument against AI starts to look very different. OpenAI signed a $10 billion multi-year deal with Cerebras. Google are developing their own custom chips. The narrative around AI’s energy consumption is very much still being written, and I’ve written a longer piece on why that matters.
Sources: TechCrunch (12 Feb 2026), Cerebras Blog (12 Feb 2026), VentureBeat (12 Feb 2026)
Also This Fortnight
Google Docs Gets Gemini Audio Summaries
Google have added a Gemini-powered audio summary feature to Google Docs. Open a document, click Tools > Audio > “Listen to document summary,” and it generates a spoken synopsis of your entire document in under three minutes. You can pick voice styles (narrator, coach, educator) and adjust playback speed. It’s essentially NotebookLM’s audio overview capability, but built directly into your working documents. If you’re the kind of person who has to read a 20-page report five minutes before a meeting, this is for you.
Source: Google Workspace Updates Blog (12 Feb 2026)
Apple Opening CarPlay to ChatGPT, Claude, and Gemini
Bloomberg reported that Apple is planning to allow third-party AI chatbot apps to integrate with CarPlay for the first time. That means ChatGPT, Claude, and Gemini could soon have icons on your car’s dashboard. Siri stays as the default (you can’t replace the steering wheel button), but you’ll be able to tap into a proper AI assistant while driving. I was genuinely excited about this one. The idea of getting into the car and being able to say “What emails did I miss? Draft a reply. Book that appointment” — all hands-free, all voice-controlled — that’s the future of AI as a proper personal assistant.
Source: MacRumors (6 Feb 2026), 9to5Mac (6 Feb 2026)
Malwarebytes Scam Detection in ChatGPT
Malwarebytes, one of the most trusted names in cybersecurity, is now available as an app inside ChatGPT. You can paste in a suspicious email, share a dodgy link, or even upload computer logs and get an AI-assisted assessment of whether it’s legit. It’s free, it’s simple, and it’s cybersecurity 101: if you’re ever unsure about something, don’t click it, don’t respond to it, and now you can ask AI to help you check it.
Meta’s AI Afterlife Patent
Meta were granted a patent in late December that describes technology allowing AI to continue posting, commenting, and even simulating video calls on behalf of a user after they’ve died. The patent, authored by CTO Andrew Bosworth, was first filed in 2023. Meta quickly stated they have “no plans to move forward” with it.
But the questions it raises are genuinely uncomfortable. Who gives consent? Does your family decide? Do other people consent to interacting with your digital ghost? Reece and I didn’t have any good answers when we discussed this on the podcast, and I’m not sure there are any yet. I’ve written a longer piece on this one because it deserves more than a paragraph.
Sources: Fast Company (17 Feb 2026), Dexerto (16 Feb 2026)
Scott’s Soapbox: Anthropic Want Feelings, Not Firewalls
Anthropic co-founder Daniela Amodei said in an interview with ABC News that when they hire, they look for people who are “great communicators, who have excellent EQ and people skills, who are kind and compassionate and curious.” She said studying the humanities will be more important than ever, not less.
This absolutely echoes what we teach at Techosaurus every single day. AI is a communication and delegation skill, not a technical one. The future doesn’t belong to the people who can operate the machine. Anyone can operate the machine. The future belongs to the people who can think around problems, question what the machine is telling them, and push it to do better. I’ve written a full blog post on why this matters so much.
Sources: BusinessToday (8 Feb 2026), National Today (7 Feb 2026)
AI Uncovers How the Romans Played
My favourite story this fortnight. Archaeologists used the Ludii AI game system to crack the rules of a mysterious 1,700-year-old Roman board game found in Heerlen, Netherlands. The game, now named “Ludus Coriovalli,” was carved into a small limestone slab and had sat unidentified in a museum for over a century. Researchers trained AI agents on over 100 known game rule sets, had them play tens of thousands of simulated matches, then cross-referenced the results with microscopic wear patterns on the stone’s surface. The conclusion: it’s a blocking game, and this pushes the history of blocking games in Europe back by centuries. You can actually play the reconstructed game online.
Sources: Antiquity Journal, Science News (10 Feb 2026), Archaeology Magazine (16 Feb 2026)
Try This
This fortnight’s challenge: If you use Google Workspace, try the new Audio Summary feature in Google Docs. Open a long document, go to Tools > Audio > “Listen to document summary” and see how well Gemini captures the key points. If you’re not on Google, open ChatGPT and try the Malwarebytes app. Paste in that email you’re not sure about and see what it thinks. Both are free, both take seconds, and both are a practical way to start building AI into your daily routine.
Listen
Reece and I covered all of these stories (and went on several tangents) in the latest episode of Prompt Fiction. Chapter 11, Part 1 is available now on all podcast platforms. Part 2 is coming soon with a story about machines with autonomy and a deeper look at AI advertising.
Listen to Chapter 11, Part 1 → prompt-fiction.show
Scott Quilter | Co-Founder & Chief AI & Innovation Officer, Techosaurus LTD