Last week I was pinned to the couch by a sleeping newborn. My wife and our four-year-old were asleep too, and I remembered that I still had to add our son to my health insurance. He was six days old. Interacting with an insurance company was about the last thing I wanted to do.
I told my agent to fill out the paperwork and stop before submitting. It did. I checked everything and submitted it myself, without waking him. Deeply unpleasant job, finished from my phone, one handed.
That is a good use of AI. I use this stuff constantly, and I want it to get better because I have a life I would like to spend more time living. Apparently that also means I'm supposed to clap for every model release and welcome whatever the people selling it want to build up the street.
The launch got ahead of the customer
On Thursday, OpenAI announced GPT-6 Astra. I opened Codex and couldn't find it in my model picker, while people with early access were already telling the rest of us how wonderful it was. I wanted to sit down and test the damn thing. Instead, I got a launch to watch.
OpenAI's announcement describes the distinction: access began with a limited set of organizations, with broader paid access to follow over the coming days. That's a staged rollout. The impression that everyone could sit down and use it on launch day ran ahead of what ordinary customers could actually do.
Choosing who gets early access also means choosing who gets to speak with authority first. Everyone else is reacting to someone else's experience, and by the time we can run our own tests, the first round of coverage is already written. A model can be excellent and the launch can still be a mess.
The useful review is the one that helps you decide whether this thing can do YOUR work. Show me the task it finished, where it failed, and how much checking the result needed. People trying to keep a job or run a business need that information considerably more than they need another man announcing the arrival of a new civilization.
Agreement has a production budget
On August 31, Bloomberg reported that Build American AI was launching a multimillion-dollar advertising campaign, initially targeting Kansas, Ohio, and Wisconsin. The group is backed by the pro-AI super PAC Leading the Future. They have money to spend on changing what people think about data centers.
There is a more revealing piece of history here. In May, WIRED reported offers of $5,000 per TikTok video for influencers to promote Build American AI's message about the threat from Chinese AI. The reporter had received an invitation to participate, and other creators confirmed similar outreach. The opinion arrived with a price and suggested talking points.
OpenAI acknowledges that Greg Brockman and his wife Anna have supported Leading the Future in a personal capacity. The company says it does not direct the group's activities. That distinction matters, and so does the fact that people who stand to benefit from AI's expansion can finance campaigns telling the rest of us how to feel about it.
I went up to Mesa to film a data center recently. Security chased me away, and I came home thinking about the houses across the road, the noise, and what living there would actually be like. If you own one of those houses, you deserve answers about that specific facility. A patriotic video about beating China tells you precisely fuck all about whether you will enjoy sitting in your backyard.
My contract keeps the criticism in
When Moonshot approached me about reviewing Kimi, I asked what would happen if I didn't like parts of it. I wanted my right to say so in the contract. They signed it, paid me for the work, and I disclosed the sponsorship in the video.
I am looking to make money. I have children and bills, and making this stuff takes time. A company can pay for that time, but I will never whore out my opinion of a product I'm testing. If the criticism has to disappear for the deal to happen, the deal can disappear.
The same standard applies when I use the product myself. Getting an insurance form filled out does not oblige me to defend Sam Altman's launch, and liking a model does not oblige you to support a particular data center. You are allowed to make those decisions separately. These companies are selling software, and I would like them to get comfortable being treated that way.
I filmed the Mesa visit and my argument about data centers because I wanted to look at what the buildout means near actual homes. If you watch one thing after this, watch that. The people living across the road have a better reason to care than the people getting paid to tell them to stop worrying.
The move this week: when an AI review convinces you to spend money, look for who paid for it and whether the reviewer was free to criticize the product. I owe you those answers too. My own test for whether the technology helped me was simple: the paperwork was done, and my son was still asleep in my arms.
- Dr. J
Check your employer's story
Bring three readers to this newsletter through your personal link below. Once all three confirm their subscriptions, I'll send you my Employer Investigation Kit free: a seven-page guide to checking layoffs, visa filings, wages and company disclosures, plus a worked Microsoft example and reusable worksheets.