#artificial

Claude cannot be trusted to perform complex engineering tasks AMD’s AI director just analyzed 6,852 Claude Code sessions, 234,760 tool calls, and 17,871 thinking blocks. Her conclusion: “Claude cannot be trusted to perform complex engineering tasks.” Thinking depth dropped 67%. Code reads before edits fell from 6.6 to 2.0. The model started editing files it hadn’t even read. Stop-hook violations went from zero to 10 per day. Anthropic admitted they silently changed the default effort level from “high” to “medium” and introduced “adaptive thinking” that lets the model decide how much to reason. No announcement. No warning. When users shared transcripts, Anthropic’s own engineer confirmed the model was allocating ZERO thinking tokens on some turns. The turns with zero reasoning? Those were the ones hallucinating. AMD’s team has already switched to another provider. But here’s what most people are missing. This isn’t just a Claude story. AMD had 50+ concurrent sessions running on one tool. Their entire AI compiler workflow was built around Claude Code. One silent update broke everything. That’s vendor lock-in. And it will keep happening. → Every AI company will optimize for their margins, not your workflow → Today’s best model is tomorrow’s second choice → If your workflow can’t survive a provider switch, you don’t have a workflow. You have a dependency The fix is simple: stay multi-model. → Use tools like Perplexity that let you swap between Claude, GPT, Gemini in one interface → Learn prompt engineering that works across models, not tricks tied to one → Test alternatives monthly because the rankings shift fast Laurenzo said it herself: “6 months ago, Claude stood alone. Anthropic is far from alone at the capability tier Opus previously occupied.”

Spent today at MIT's Open Agentic Web conference. Six things worth thinking about. **We're in the DNS era of agent infrastructure.** Before agents can find and trust each other at scale, you need identity, attestation, reputation, and registry infrastructure — the same structural role DNS played before search was possible. This came up independently from multiple directions. It's the most underbuilt layer in the stack right now. **The chatbot framing is a local maximum.** The most interesting work wasn't better UX or smarter responses. It was agents as persistent actors that discover, negotiate, and transact across networks over time. People doing serious work have already moved past the assistant model entirely. **Coordination is the hard problem, not capability.** A room full of brilliant agents can still fail badly. This matches what I found running HiddenBench against frontier models earlier this year; collective reasoning is not the sum of individual reasoning. There's a real argument that the frontier is protocol design, not model scaling. **"Commerce of intelligence" is a real category.** Not buying things through agents. A market where intelligence itself (bundled, verified, priced, resold) is the object of exchange. Felt like the most underexplored idea in the room. **Data provenance becomes load-bearing.** What an agent knows, how it was verified, under what terms it flows: this is the actual architecture forming beneath everything else. **Partnership keeps outperforming replacement.** Demos that actually worked (healthcare, enterprise) was about helping experts operate at higher leverage, not substituting them. Autonomy theater keeps failing in the same ways.

6 Months Using AI for Actual Work: What's Incredible, What's Overhyped, and What's Quietly Dangerous Six months ago I committed to using AI tools for everything I possibly could in my work. Every day, every task, every workflow. Here's the honest report as of April 2026. --- **What's Genuinely Incredible** 1. First drafts of anything — AI eliminated the blank-page problem entirely. I don't dread starting anymore. 2. Research synthesis — Feeding 10 articles into Claude Opus 4.6 and asking "what's the common thread?" gets me a better synthesis in 2 minutes than I could produce in an hour. 3. Code for non-coders — I've built automation scripts, web scrapers, and a custom dashboard without knowing how to code. Cursor (powered by Claude) changed what "non-technical" means. The tool has 2M+ users now for good reason. 4. Getting unstuck — Talking through a problem with an AI that can actually push back is underrated. Not therapy, but something. 5. Learning new topics fast — "Teach me [topic] like I'm smart but completely new to this. What are the most common misconceptions?" is my go-to for rapid learning. --- **What's Massively Overhyped** 1. "AI will do it for you" — Everything still requires your judgment and context. The AI drafts. You think. 2. AI SEO content — The "publish 100 AI articles and watch traffic pour in" strategy is even more dead in 2026 than it was in 2024. Google has gotten much better at identifying low-value AI content. 3. AI chatbots for customer service — Unless you invest heavily in training and iteration, they frustrate users more than they help. 4. "Set it and forget it" automation — AI workflows break. They require monitoring. Fully autonomous workflows exist only in narrow, controlled cases. 5. Chasing the newest model — New model releases happen constantly now. I've learned to stay on a model that works for my tasks rather than jumping to every new release. --- **What's Quietly Dangerous (Nobody Talks About This)** 1. Skill atrophy — My first-draft writing has gotten worse. I outsourced that skill and I'm losing the muscle. I now intentionally write without AI some days. 2. Confidence without competence — Frontier models give confident-sounding answers to things they don't know. If you're not knowledgeable enough to catch errors, you can build strategies on wrong foundations. 3. The "good enough" trap — AI output is often 80% there. If you stop at 80%, your work looks like everyone else's. The 20% you add is the differentiation. 4. Over-automation without understanding — I automated a workflow without fully understanding it first. When it broke, I couldn't fix it. Understand before you automate. 5. Vendor dependency — My workflows are deeply integrated with specific AI tools and APIs. Pricing changes, policy shifts, and service disruptions are real risks at this point. --- **The Honest Summary** AI tools have made me more productive, creative, and capable than I've ever been. They've also made me lazier in ways I didn't notice until recently. The people winning with AI in 2026 aren't the ones using the most tools or running the newest models. They're the ones using AI to amplify genuine skills and judgment — not replace them. What's your honest take after 6+ months of serious AI use? Curious whether others have hit these same walls.

Everyone wants to be a content creator. nobody wants to actually create anything worth watching. and the internet is slowly suffocating under the weight of it. i want to be very clear upfront, i'm not talking about people who are genuinely trying. i'm not talking about the person in their bedroom at midnight editing their 30th video because they actually love what they make. i'm talking about the other kind. the ones who downloaded CapCut on a Tuesday, pointed their phone at their face on Wednesday, and by Friday were telling people at family dinners that they're a "content creator." the internet used to be where you went to find something you couldn't find anywhere else. now it's where everyone goes to show you something you've already seen just slightly worse. and i think i finally understand why this is happening. somewhere along the way, the word "content creator" got completely detached from the word "content." the creator part became the goal. the actual content became an afterthought. a necessary inconvenience between you and the fame you've already decided you deserve. people don't ask themselves "what do i have to give?" anymore. they ask "what do i have to post?" and those two questions produce very, very different things.

Why are you hopeful about AI? While AI does provide some value, humans rarely operate in the best interest of others. I have no doubt that governments, businesses, and criminals will use AI for nefarious reasons. I think people need to quit comparing it to lesser technological innovations. It’s not the same. Not even close. Why are you optimistic? And why should I be less pessimistic?

What's your "When Language Model AI can do X, I'll be impressed"? I have two at the top of my mind: 1. When it can read musical notes. I will be mildly impressed when I can paste in a picture of musical notes and with programming sets up instruments needed to play music, and then correctly plays the song it reads from the notes. 2. My jaw will drop when finally with a simple prompt an AI can create a classic arcade style fully functioning and fun to play Pinball game. Each new version of models that become available I give that one a go. None have been even remotely close to achieving this goal. So what are your visions for what will impress you to some extent when an AI can make it for you?

bad grammar is literally the last proof that ur human. and i think thats actually terrifying ngl we're in this weird era now where everyone suddenly writes perfectly. every message, every email, every caption. no typos. no "lol sorry typed that too fast." no lowercase chaos. just. clean. polished. structured. english. and it's freaking me out bc clean polished english used to mean someone was smart or educated or careful. now it just means they hit "improve with AI" before they hit send. here's the part that actually keeps me up we spent years being embarrassed about bad grammar. teachers corrected it. bosses judged it. people got roasted in comment sections for it. and now? bad grammar is basically a flex. it's the handshake that says yeah, a real disorganized sleep-deprived actual human being typed this with their actual thumbs and didn't stop to clean it up bc they had something real to say and just said it. ur typos are ur fingerprints now. dont let anybody take them from you.