The skinny on what's happening in AI - straight from the previous live session:
Highlights:
🚀 Claude Opus 4.5: The Model You Can Actually Use
Anthropic dropped Claude Opus 4.5 and frankly it’s brilliant. It’s also the third state-of-the-art (SOTA) model in a single week after Grok 4.1 and Gemini 3.
Benchmarks are predictably excellent - 80.9 on SWE-bench (versus 76.2 for Gemini 3 Pro), 37.6 on ARC-AGI-2 (versus Gemini's 31.1). But here's what actually matters: they've finally made it usable on a day to day basic.
Kyle's take: The old Opus was basically an (unusable) tease. Use it a handful of times, then "come back tomorrow." Useless for actual work. I’m on the $200 plan and it works fine but on the $20 plan? Forget about it - it was basically a lure to bump you up to $200.
Now they've removed the Opus-specific caps entirely - you get roughly the same tokens as you had with Sonnet.
The real win here isn't the benchmarks. We knew they would be good. No, the win here is that Anthropic's best model is finally something you can integrate into your daily workflow without it randomly abandoning you mid-project.
Plus they've added automatic conversation compacting, so chats can run indefinitely. It’s a small but super important thing. Oh and Cursor's offering Opus 4.5 at Sonnet pricing until December 5th - go take advantage of that whilst it lasts!
💀 "Software Engineering is Done" - Anthropic Employee
Adam Wolff from Anthropic posted that with Opus 4.5, he's routinely having the agent work autonomously for 20-30 minutes. When he comes back, tasks are done. His conclusion: coding as we know it is basically solved, likely by 2026.
Kyle's take: This tracks with what I experienced overnight - I put coding tasks into 4.5 overnight, woke up, and they were done clean. If you've got problems you've been battering your head against, throw 4.5 at it.
The Reddit crowd is attacking Wolf personally for saying this, but betting against AI is a genuinely stupid bet at this point. We saw it with Nano Banana Pro and text in images - everyone said that was years away. Then it just... happened. Ta-da: that’s a solved problem.
I was chatting to a designer yesterday in a cafe who'd just seen Nano Banana Pro and said "yeah, that's kind of the ball game." The uncomfortable truth is: if your entire value proposition is a skill that AI keeps getting better at, you need to be building for optionality. Have your nine-to-five, but also have something on the side. Because we genuinely don't know what's coming.
🎓 Andrej Karpathy: Homework is Dead, Schools Need to Adapt
Karpathy (OpenAI co-founder) spoke to a school board about AI and dropped some uncomfortable truths. His main points: AI detection in homework will never work. Full stop. That’s over.
You have to assume all work done outside the classroom used AI. The solution? Flip the classroom - do learning at home, do assessment in class where you can monitor students.
Kyle's take: The detection game is over before it started. The money going into AI generation versus detection is orders of magnitude different - it's an unsolvable problem.
I sit in cafes and watch students with their textbooks open and ChatGPT running beside them. That's just reality now.
The interesting bit is Karpathy's point about calculators (go read his full tweet) - we still teach arithmetic so you can gut-check when your calculator gives a wrong answer. Same principle applies to AI, except AI is confidently wrong in ways that are much harder to spot.
If you grew up having to do things manually, you've got that critical faculty. Kids growing up now? The temptation to skip building that foundation is massive. Not sure I would have been able to resist it! The flipped classroom isn't just nice-to-have anymore - it might be the only way forward.
🛒 OpenAI's Shopping Research: The Ads Are Coming
OpenAI launched "Shopping Research" in ChatGPT, timed perfectly for Black Friday! It's deep research but specifically for purchasing decisions - comparing products, finding deals.
Kyle's take: This is how they're going to sneak advertising in. Start with shopping - nobody complains about sponsored product recommendations when you're already trying to buy something.
Once that's normalised, it spreads to regular queries.
Google's in the same position - 70% of their $350 billion revenue comes from search advertising, and they're actively cannibalising it with AI. Nobody wants to be first to stick ads in AI responses, so they're all waiting for each other. See who blinks first.
My prediction: we'll see proper ads in all major AIs by end of next year. They have to make money somehow!
Member Question: "Any thoughts on AI CRMs?"
Kyle's response: The tool has to be good first. Just because something has "AI" slapped on it doesn't make it useful. Every CRM is adding AI features anyway, so that's not a differentiator. What matters is whether it fits your workflows, integrates with your existing systems, and whether your team will actually use it. Migrating off a CRM is an absolute nightmare, so make the right choice based on fundamentals - not because the marketing material mentions AI forty times.
Member Question: "I'm a new headhunter and plan to use AI to find new clients. Any ideas?"
Kyle's response: The obvious answer is scraping LinkedIn with Apollo/Apify, getting emails, using AI to write personalised outreach. Problem is: everyone's doing that!
I delete every AI-generated LinkedIn message I receive. The ones I respond to? Short, casual, slightly messy grammar - because they're clearly written by an actual human.
Sounds mad coming from an “AI guy”, but I'd lean into a "person-first" approach. Use AI for backend admin, but do actual human outreach. Any agency role - headhunters, estate agents, even some solicitors - is getting automated. You can ride the wave short-term, but full automation is coming. Building a human-first brand gives you something defensible.
Want the full unfiltered discussion? Join me tomorrow for the daily AI news live stream where we dig into the stories and you can ask questions directly.
Streaming on YouTube (with screen share) and TikTok (follow and turn on notifications for Live Notification).
