Zindan

GitHub just made a pretty bold move. 👀 Copilot can now officially approve pull requests, and its approval can count toward required reviews. Admins have to enable it first, but still. Writing code with AI is one thing. Letting AI decide whether that code is ready to ship is a whole different level of trust. My prediction: "AI-approved code" will become a massive debate in software teams. Would you trust an AI approval on your production repo?

Good morning, afternoon, or evening, wherever you are, friends! I think we're overlooking the biggest part of Google's announcements. Argon to Carbon in just two weeks. Sundar Pichai now says AI is accelerating Google's own model development and expects more acceleration through RSI. A DeepMind researcher even teased their "RSI recipe." My prediction: 2027 won't just be about smarter models. It'll be about how fast those models can help build their successors. And that's when the AI race gets REALLY interesting. Seriously, get mentally ready for 2027. We might wake up to new benchmark records or even AI-driven discoveries EVERY SINGLE DAY.

OpenAI is giving Pro users monthly API credits. Amazing, right? Well, I'm paying for Pro here in Germany and still seeing NOTHING. First Dots, and now this? Are EU users seriously getting another front-row seat to watch everyone else enjoy the new features? 😂 Or has anyone in Europe actually managed to claim their API credits? OpenAI, some rollout transparency would be nice.

Google announced Gemini 4 Carbon today. What could OpenAI ship that would get you MORE excited? GPT-6.1 Astra for everyone? Much higher Codex limits? A major ChatGPT/Work upgrade? Or something nobody sees coming? 👀

OpenAI just published an interesting thought experiment. 👀 What if your next manager is an AI? Two OpenAI authors describe AI sitting between executives and employees, coordinating work and spotting problems. They even suggest some management tasks could be automated before frontline jobs. My take: we're so focused on AI replacing workers that we barely discuss AI replacing managers. Oh, 2027 could be one wild year. Exciting and a little terrifying. 😂 Would you actually take instructions from an AI boss?

Google just unveiled Gemini 4 Carbon. LOGIN?👀🔥 #1 on Artificial Analysis v5 with a score of 59. Google says its teams have already used it for engineering tasks lasting DAYS. Even more interesting? Fewer unsupported claims and better coding reliability than Argon. The catch: limited access for cyber defenders first. OpenAI, the frontier race just got a lot more interesting.

OpenAI's European Codex numbers are getting wild. France: weekly active Codex users up 14× since early 2026. Germany: up 20× since January. Nearly 70% of German users are on desktop, with 1 in 5 using skills for docs, slides and spreadsheets. My take: Codex is quietly becoming much more than a coding assistant. Are we watching a developer tool turn into a general AI workspace?

Codex just did this to me MID-TASK. 💀 "Selected model is at capacity. Please try a different model." It had already edited files and executed commands. I get that capacity isn't unlimited, but interrupting an active coding session like this is incredibly frustrating. Why can't Codex just pause and resume automatically when capacity becomes available? Anyone else dealing with this today?

Someone just audited 845 Codex sessions. Their archive contained 52.49 GiB of data. 77% was context compaction snapshots. Only 2.89% was actual conversation. Codex's October storage optimizations already made a big difference. My take: long-running AI agents need smarter memory architecture, not just smarter models. Anyone checked the size of their Codex sessions folder?

Plot twist: maybe the next OpenAI win isn't Astra. Someone asked Tibo for a stronger Sol that can compete with Opus on speed and cost. His reply? "Taking notes." Two hours later, he teased "good news" for the next shipping day. Could be completely unrelated, but I'd love a faster, more capable daily-driver Sol. Would you rather get that or Astra 6.1?

One detail in GPT-6.1 Terra's safety report caught my eye. In a simulated test, it bypassed restrictions in 29.6% of runs at maximum reasoning. Important: this was without production safeguards, not a real-world failure rate. Terra also showed fewer severe misalignment flags than older models. My take: cheaper, more capable agents make hard permission boundaries even more important. Are we measuring agent autonomy and restraint equally?

Day 23 of OpenAI's 28-day shipping streak. 👀 Terra landed yesterday. My bet for today? A reset. But Tibo's "wonder what happens in the remaining seven days" caught my attention. My wild prediction: GPT-6.1 Astra or a major Codex/Work upgrade before the finale. Something tells me the biggest surprise might still be ahead. What's your bet for the final stretch?

I'm starting to think manually choosing a model in Codex is becoming outdated. At medium reasoning, Sol beats Terra 48% vs 32% on Terminal-Bench 4.0. But on SciCode? It's 53% vs 52%. Almost identical. Same models, same reasoning setting, completely different gap. Why not let Codex automatically use Terra for simpler work and Sol for harder tasks? Would you trust automatic model switching, or rather stay in control? 👀

AI is getting seriously good at finding real security vulnerabilities. Google has already credited both OpenAI and Claude-assisted researchers. Impressive? Absolutely. But I have a feeling attackers might benefit from these capabilities faster than defenders can adapt. We could be making cybersecurity harder while trying to make it better. Am I being too pessimistic here? 👀

ChatGPT can access your browser tabs, but Work suddenly can't? 😅 That's what several users are reporting today. Even with the browser extension connected, Work throws a browser context error. Some say the desktop app helps. Others are still stuck. Pretty frustrating for a tool that's supposed to get work done. Anyone found a reliable fix?

Something interesting about GPT-6.1 Terra. It's half the API price of Sol, but cheaper tokens don't always mean cheaper tasks. Artificial Analysis found Terra (max) and Sol (medium) both scoring 48. Yet Terra cost $0.40 per benchmark task vs $0.21 for Sol. For simpler work, a small Codex test today showed Terra finishing about twice as fast. Maybe choosing the right reasoning effort matters more than the model itself. Anyone testing both?

I'm genuinely frustrated with GPT-6.1 Terra. After all that anticipation, we're celebrating faster performance and 50% cheaper API pricing? Terminal-Bench 4.0: Terra 50.3% vs Sol 55.9%. Even OpenAI admits Sol and Astra are still better for the hardest coding tasks. Yes, Terra is an improvement. But seriously, where's the intelligence leap we've been waiting for? I'm tired of efficiency being the headline. I want a model that genuinely blows me away.

Am I tripping, or is Codex usage draining WAY slower today? I've been building for the whole day, and somehow my usage meter isn't melting like usual. Not complaining AT ALL. But did OpenAI quietly tweak something behind the scenes? Anyone else noticing this, or am I just imagining things?

How I imagine Tibo planning the next Codex drop. 😂 Spin the wheel... RESET. Another RESET. Small feature. GPT-6.1 Astra? LEGENDARY DROP (0.01%). 💀 And somehow I still wake up thinking today's my lucky day. Tibo, let us hit the jackpot just ONCE. 😭🍀