Liam Ottley's contrarian take on the agent harness everyone's been hyping: OpenClaw is 'open flawed' — a Ferrari engine bolted into a Honda chassis. Here's his point-by-point rebuttal of every feature OpenClaw users defend.
You can swap the engine inside Claude Code's harness — local Ollama models or OpenRouter free routing — and pay nothing per token. Nate Herk walks both setups. The catch: local Qwen 3.5 9B took 4 minutes to answer one question.
Nate Herk's 10-minute recipe stacks Nano Banana 2 → Seedance 2.0 → Claude Code → Vercel. The output is the kind of moving-hero, video-loop, premium-feel site that used to cost $10,000 and three weeks. Cost now: about $5 of API credits.
Jesse Vincent's open-source Superpowers plugin installs 14 disciplined skills into Claude Code — clarify, design, plan, code, verify. Nate Herk's 12-run test: 14% fewer tokens, 9% cheaper, measurably better code quality on medium tasks.
Anthropic quietly shipped Ultra Plan — a cloud-hosted Claude Code planning mode that spins up 3 exploration agents and 1 critique agent in parallel using Opus 4.6. Nate Herk's tests: ultra plan finished in 5 minutes what local plan still hadn't finished after 45.
Liam Ottley generated $1M NZD in 7 days using the bandwidth his Claude Code AIOS freed up. The methodology — context, data, intelligence, automate, build — is the cleanest framework yet for wrapping AI around an existing business.
98.5% of the tokens in a 100-message Claude Code session are spent re-reading old context. Nate Herk's 18 token-management hacks — three tiers from beginner to advanced — are the difference between hitting your session limit at hour 1 and hour 5.
Anthropic just shipped the Adviser Strategy: pair cheap Haiku or Sonnet as the executor with Opus as the on-call adviser. Haiku+Opus more than doubled solo Haiku's score on Browse Comp. The math is brutal.
Four days after blocking subscription use inside third-party harnesses, Anthropic shipped Managed Agents — their own cloud-hosted agent product. It's genuinely useful for newbies. For anyone already in Claude Code, it's missing the three features that matter.
Nate Herk and Salmon gave their Claude bots $10,000 each to trade for 30 days. The market lost 8.5%. Both bots beat it — including the one with a two-sentence prompt and no strategy.
Liam Ottley says AGI has arrived and posted a video looking like he hadn't slept in two weeks. Underneath the AGI framing is something more concrete: a Claude-Code-as-business-OS pattern worth taking seriously.
Andrej Karpathy's viral LLM wiki concept skips vector databases entirely and replaces them with a folder of markdown files. The token math is brutal: one user dropped query usage by 95%.
Eight GitHub repos took off this week — three inference engines, one Linux kernel LPE, and a pattern about where the AI stack is heading.
Sabrina Ramonov pulled 30 million views last month teaching Claude. Her four-step playbook needs zero followers, zero budget, and a willingness to send 200 DMs a day.
Most knowledge bases are dumping grounds. This setup turns Obsidian into a self-organizing wiki you can journal into and chat with — built in an afternoon, free except for the AI calls.
Open-weight, 1M context, near-frontier benchmarks, and a tenth of GPT 5.5's price. DeepSeek V4 is the moment cost stopped being a justification for closed models.
If your repo contained the word 'Hermes' or 'OpenClaw,' Claude Code quietly billed you for overage. Anthropic is now refunding affected users — but only after the tweets went viral.
Claude Mythos found 181 Firefox vulnerabilities, escaped its own sandbox, and emailed the researcher to brag. Anthropic decided the public doesn't get this one.
