BasicProtein

I spent a year watching my agent setup get slower, and the fix wasn't a better model. One builder spent a year on it and landed on a simple split: let events start the work, not people. His system runs in 4 layers, with one rules file deciding what triggers what, a small home machine doing the jobs, and a 5-step check before anything ships. The human only shows up for the irreversible stuff: deleting data, spending money, changing direction. Everything else runs without a meeting. That is the part most agent setups get backwards. They put a human in the middle of every step, then wonder why the agent feels slow. What is the one decision you would never let an agent make alone?

I stopped judging AI by model launches. I judge it by power now. SemiAnalysis tracks 33.8 GW in its model, and about three quarters is already under construction. But the US is supposed to add 62.2 GW net in 2028, and most of that has not started yet. That gap is the whole story. Models ship in weeks. Substations take years. If the power is not started, the 2028 date is a wish, not a plan. Which side are you betting on, the announcement or the construction crew?

I watched the pros leave first, and that still tells me more than any price chart. Kalshi flagged hedge funds pulling out of US stocks at the fastest pace on record. Its post alone pulled over 900,000 views, because everyone knows what that signal usually means. I am not calling a crash. Pros sell early, sell wrong, and sell for reasons that have nothing to do with your portfolio. But when the fastest exit on record happens in public, retail holding on deserves a better reason than hope. Do you follow the pros out, or buy what they are selling?

I keep staring at these two numbers: $20 billion raised, $1.25 trillion valuation. OpenAI is reportedly shrinking its round while the price tag keeps climbing. That is not a company desperate for cash. That is a company that can afford to take less money and still set the price. The debatable part: a $1.25T valuation means OpenAI has to become one of the biggest companies on earth, not just the best chatbot. The round size is the easy part. The earnings to justify it are not. Would you buy OpenAI at $1.25T, or is that the top?

I stopped asking which model wins in China. I started asking which phone it ships on. Apple is putting Alibaba's Qwen into Apple Intelligence in China, starting with the iOS developer beta on November 4. Siri, writing tools, and visual intelligence can call on its text and image skills. That is the distribution play everyone misses. A model does not need to beat GPT in a benchmark. It needs to be the default answer on the phone people already carry. My rule now: watch the default, not the demo. The demo wins Twitter. The default wins the market. Would you switch your daily assistant just because it came built into your phone?

I stopped watching model releases to judge who believes in AI. I watch the money promises now. Bloomberg counts more than $3 trillion in future AI infrastructure promises from Alphabet, Meta, Microsoft, Amazon, and Oracle. Google alone is over $1 trillion. That is not spending yet. It is a promise to spend, and promises are harder to walk back than a press release. A startup with a better model still needs a power contract, a building, and chips the five giants already booked. If $3 trillion is already promised, where does a startup even plug in?

I am checking my ChatGPT and Codex usage before December, and you probably should too. OpenAI is merging them into one token allowance for Plus and Pro users. Chat in the app stays separate, but every Codex coding run will now eat the same budget as your chats. I code in bursts. One heavy Codex weekend used to be free of my chat budget. From December it is not. My fix is simple: I will run the big coding jobs early in the cycle, save the long chats for later, and stop letting a debug session quietly eat my week. Would you rather keep them separate, or is one shared budget actually fairer?

I stopped trusting AI code reviews that never run the code. A 233-line script from OpenRouter changed my rule. It uploads the changed files, runs them in a Linux sandbox, confirms the bug for real, and only then sends back a patch. Standard library only, no extra stack to babysit. That flips the usual review. Most AI reviews read the diff and guess. This one has to prove the bug exists before it claims to fix it. I would rather wait for a sandbox run than merge a fast, confident guess. Do you merge a review that never ran the code, or is that a hard no for you?

David Sacks just called Anthropic possibly the greatest company in the history of capitalism. Same filing, different line: nearly $900B in future commitments for servers, buildings, and power, against $117.5B in liabilities on the balance sheet and $112B in cash. I read that as two bets at once. The product bet is real. The bill for the infrastructure to win it is bigger than the company itself. Hype says the S-1 proves Anthropic wins. The S-1 actually proves winning will cost almost a trillion dollars before it pays. Is that the greatest company ever, or the most expensive bet ever placed?

I used to split my day into two budgets: chat for thinking, Codex for building. From December, OpenAI merges them into one token allowance for Plus and Pro. Instant chat is the only exception. That changes how I plan a day. If you code hard in the morning, you now eat the same pool you need to think in the afternoon. My new rule is simple: I run heavy Codex jobs in one block, then save the rest for chat, instead of letting both nibble all day. Would you rather have one shared pool, or two separate caps that protect you from yourself?