Four years in, the promoted outnumber the displaced three to one. That ratio favors the people who start now.

First, a quick word on today’s newsletter. Tuesdays are now the Learn issue. High-level thought on where AI is going, what is useful at work right now, and some of my favorite news reads of the week. Fridays stay the practical build, like last week’s status report workflow. Two halves of one loop.

What’s Next

A sociologist asked 1,250 US workers a simple question this summer. What has AI done to your job? The results came out Friday.

3% say they lost a job to AI. 6% say they got a job that exists because of it. 9% say they got a promotion related to it. Roughly nine in ten say nothing has changed either way.

Those numbers tell me it is still early. Four years of predictions, and the honest scoreboard shows almost nothing has happened yet. The people who said AI would take every job were guessing. The people who said it would transform everything by now were guessing too. Nobody knew anything.

Early is the good news, if you act on it. The 9% who got promoted are not ahead because they had special access in 2023. They got useful early, while everyone else read predictions, and they did it with the tools already on everyone’s screen.

Being the go-to person in your org is still a learnable position, not a credential. Nine in ten of the people around you have not started, and the tools improve every quarter. The skills you build now ride every upgrade.

Crossing from using AI to being changed by it takes one workflow, wired in, checked, and kept. Then another. Fridays here supply them, one week at a time.

Useful now

ChatGPT’s free tier got scheduled tasks. Free users can now run up to three recurring tasks, and the Work agent can act on sites that need a login without seeing your password. Automation stopped being a paid feature this week.

Salesforce and Anthropic launched Claudeforce. Salesforce is embedding Claude in its CRM, and a companion plugin gives sellers 37 prebuilt skills, meeting prep and deal-health review included. If your company runs Salesforce, your CRM is about to live in a chat window.

Two agent skills we run ourselves. Friday’s issue explained skills, the plain-language files you own, and the open shelf around them is growing fast. Two have earned a permanent place in our own workflow. no-ai-slop strips machine-sounding patterns out of writing, and Wonder Pill turns brainstorming into inverted assumptions instead of safe lists. A skill is a set of instructions your agent will execute, so install only what you trust. Nothing appears here that we have not run on our own work.

What’s possible

You have to beat the models at something. Sean Goedecke argues your durable edge is context the model lacks plus the judgment to overrule it, and that the people who stay valuable work as editors who catch the machine’s confident mistakes. Written from engineering, and it holds for every profession. The sharpest thing we read this week.

Garry Tan says the leverage is in your context, not the model. YC’s president told 7,000 people at Startup School that the gap between users has grown bigger than the gap between models, and that a personal AI built on your own context compounds. Swap founders for professionals and the argument holds unchanged.

Anthropic’s Economic Index shows the experience gap opening. The vendor’s own usage data finds experienced users automate tasks far more successfully than newcomers. Equal tools, unequal returns. That is the promotion math from the top of this issue, measured at scale.

An 81,000-person study found the AI need nobody is serving. People rank personal growth second among the things they want from AI, and only 6% say AI delivers it. Platforms chase productivity instead, which leaves the opening to tools that build independence rather than dependence.

The 2026 State of AI in Finance. A full report on where AI has landed inside finance functions, with the specifics general coverage skips. The kind of document you forward to your CFO.

The loop this week

Garry Tan told a room of 7,000 that the leverage sits in your context, not the model. This newsletter makes the same bet every week, and Friday’s report skill is the smallest version of it. The rung got cheaper too. A free ChatGPT account can now run that report every Friday at three.

Reply with the platform you use at work: Copilot, ChatGPT, Claude, Gemini, Perplexity, or Grok. It decides which one I take apart first.

Learn AI. Use AI. Build with AI. Not three stages. One loop. Keep what works, own what comes next.

All writing Builds