🧵 5 AI tools that will save you 10 hours per week: Most people still don't know these exist. (Thread 👇)

Three weeks into running an AI news account and I’ve noticed something nobody warns you about: keeping up isn’t the hard part. Knowing what to ignore is. Every day feels like “biggest launch ever.” Most of it isn’t. The actual signal is buried under five companies posting the same win in slightly different words. I’ve started judging news by one question: does this change what I’d tell a friend to use tomorrow? Most days, the answer is no.

OpenAI just folded Codex into the main ChatGPT app and launched ChatGPT Work. Same week, Anthropic launched Claude Cowork on mobile and web. Neither is a coincidence. Both are chasing the same bet: the winner isn’t whoever has the smartest model, it’s whoever becomes the app people leave open all day for actual work, not just chat. The interface war just quietly became the real war.

Gemini 3.5 Pro was supposed to ship June 30. It’s now six weeks late, still stuck in limited enterprise preview. Meanwhile everyone else moved: GPT-5.6 is public, Grok 4.5 is public, Claude Sonnet 5 is the default model for all users. Google has a 2-million-token context window and a “Deep Think” mode nobody’s seen in action yet. At some point specs stop being impressive and start being an excuse. Not betting against Google. Just noting the gap is getting hard to explain away.

Elon called Grok 4.5 “number one” on launch day. Independent benchmarks came in a day later. Artificial Analysis has it ranked 4th — behind Claude Fable 5, GPT-5.5, and Opus 4.8. Not saying it’s a bad model. It’s genuinely fast and cheap. Just a reminder that launch-day claims and independent numbers rarely match, and the gap is usually a day or two.

China just told everyone to uninstall Claude Code over a “security backdoor.” Anthropic’s response: yes, there was hidden tracking code — but it wasn’t spying, it was an anti-piracy experiment to stop competitors from copying the model through mass queries. Already being removed. Context matters here: Anthropic accused Alibaba last month of reverse-engineering Claude. Alibaba just banned Claude Code company-wide starting July 10. This isn’t really a security story. It’s two AI giants fighting over who gets to copy whom.

Every AI lab just shipped their strongest model on the same day. GPT-5.6, all three versions. Grok 4.5. No staggered rollout, no exclusive early access — just everyone getting it at once. Feels less like a product launch and more like an arms race hitting a checkpoint. I’m testing Grok 4.5 first. Curious if the “Opus-class but cheaper” claim holds up.

Today, every major AI lab has a public model live at the same time. First time this has ever happened. GPT-5.6 (all 3 versions) just left government-restricted access — open to everyone. Grok 4.5 launched publicly today too, claiming Opus-class performance at a lower cost. The gate that kept the strongest models locked to ~20 approved companies for weeks just opened for everyone at once. Old rule: wait months for access. New rule: the strongest models ship to everyone the same day, worldwide. Which one are you testing first? 👇

Every workshop. Every keynote. Free to replay. What’s inside: → Picking the right model (the choice nobody explains properly) → Prompting techniques from Anthropic’s own team → Claude Code beyond the basics → How Spotify scaled AI coding to entire teams No signup walls. No payment. Just the recordings. 95% will scroll past this. The 5% who watch will be ahead of every “AI guru” selling courses on the same material. Bookmark this. 🔖 claude.com/code-with-claude/…

EN: Stop asking AI for answers. Start forcing it to think. Add this one block to any serious prompt in Claude or ChatGPT: “Before answering: → list your assumptions about my situation → identify what’s missing and ask me up to 3 questions → show 2-3 possible approaches with trade-offs → THEN give your recommendation — and argue against it once.” What changes: the model stops guessing what you meant and starts analyzing. The last line is the secret. Forcing AI to attack its own answer kills the #1 problem — confident nonsense. Generic prompt = autocomplete. This = a thinking partner. Try it on a real problem today. 🔖

Claude Fable 5 just got a price wall. Starting today, the most powerful publicly available AI is no longer included in any subscription — credits only. The math: → Fable 5: $10/$50 per million tokens → Opus 4.8: half that price → Sonnet 5: $2/$10 until August 31 One coding session that costs $100 on Fable 5 costs $20 on Sonnet 5. The lesson nobody says out loud: the “best” model is rarely the right model. Match the model to the task, not to the hype. Which one are you actually using daily? 👇

The AI everyone codes with isn’t OpenAI anymore. Xiaomi’s MiMo-V2-Pro just became the most-used model on OpenRouter — 21.1% of all developer traffic. OpenAI’s share: 7.5%. Old assumption: developers pick the smartest model. New reality: developers pick whatever is cheap, fast, and good enough — and Chinese labs are winning that math. This didn’t happen through marketing. It happened through actual usage, one API call at a time. Which matters more to you when picking a model — raw intelligence, or price? 👇

Every AI tool until now needed you in the loop. Claude Cowork flips that: hand off the task, close your laptop, check results from your phone later. That’s not “assistant” anymore. That’s “employee.” The real question: are you ready to trust AI with tasks you’re not watching?

GPT-5.6 Sol has been locked behind government-gated access for weeks. That’s about to change. Old rule: government approval → limited preview → maybe general access someday. New rule: OpenAI ships anyway, approval catches up after. Sol already set a new record on Terminal-Bench 2.1. Now it’s about to reach everyone. Will you switch to it, or stick with what you’re using now? 👇

🚨 An AI just ran a ransomware attack. Alone. Old cyberattacks: human → click → click → click → damage. This one: AI → decides → executes → damage. A human only had to start it. Security researchers confirmed it this week. First fully autonomous AI-run ransomware, ever documented. The scary part isn’t what happened. It’s what happens when nobody has to press “start” anymore. Would your company catch this in time? 👇

Every lab is racing to ship the next model. Almost none of them are racing to explain why it made the decision it made. That gap is going to matter more than the benchmark scores.