I am writing this from San Francisco.
I have been here for a few days. And I need to tell you something that is difficult to put into words from the outside but is completely unmistakable when you are standing in it: the conversation here is not the same conversation the rest of the world is having about AI.
Outside San Francisco, the conversation is still about possibility. Will AI affect my job? Will it be accurate enough to trust? Which model should I use? These are the questions of people watching a storm approach from a distance.
In San Francisco, the storm has already made landfall. The people building this technology are not debating whether it will change the world. They have moved past that. They are now living inside the changed version — and asking the next set of questions. The ones that come after.
This issue is about what I saw. Three signals that are easy to miss from the outside, and impossible to ignore once you are here. And what they mean for the 12 months ahead of you.
Signal 01 — The AI That Pays Rent
On Market Street in San Francisco, there is a retail boutique called Andon Market. It is run entirely by an AI agent named Luna. Not managed by AI — run by it. The founders signed a three-year lease, deposited $100,000 into a bank account, handed Luna a debit card, and stepped back. Luna manages inventory, negotiates with suppliers, sets prices, handles customer interactions, and makes daily operating decisions. Luna is powered by Anthropic's Claude — the same model I use every day to write this newsletter.
This is not a demo. It is not a pilot program. Luna pays rent every month. The AI has financial obligations and a multi-year operating mandate. That sentence would have sounded absurd 18 months ago. Today it is a Tuesday in San Francisco.
Signal 02 — The 18-Month Clock
In February 2026, Mustafa Suleiman — CEO of Microsoft AI — said publicly that all white-collar work will be automated within 18 months. Not "impacted." Not "disrupted." Automated. The word choice was deliberate and he did not walk it back.
Meanwhile, PwC data already shows a 56% wage premium for workers with demonstrable AI skills — right now, not in five years. And Anthropic's own economists published research in April 2026 showing that Claude can already perform a substantial portion of most knowledge work jobs at professional quality. These are not predictions. They are measurements of what exists today.
The divergence between AI-fluent and AI-passive workers is not theoretical. It is being paid out in salary differences this month.
Signal 03 — The Street Has Already Changed
I took a Waymo here. No driver. Smooth ride. Completely unremarkable — which is itself the most remarkable thing. When a driverless vehicle becomes boring, something fundamental has shifted. And Waymo is now about to have competition: Uber and Lucid launched a robotaxi in April. Amazon's Zoox is launching paid service later this year. Tesla is coming. San Francisco is about to become the only city on Earth where you can choose between four competing autonomous vehicle services.
The physical world changing this fast is always the last signal — the one that means the shift is no longer reversible. Software changes invisibly. When the streets change, the argument is over.
What This Means for the Next 12 Months
The people I speak to inside these companies are not worried about the technology. They are worried about the gap — the distance between what is being built and what most people understand is being built. That gap is the actual risk.
If you are a knowledge worker — analyst, lawyer, marketer, consultant, strategist, writer, manager — the question is no longer whether AI will do parts of your job. Anthropic's research says it already can. The question is whether you are on the right side of the 56% wage premium. Whether you understand how to direct these systems rather than compete with them. Whether you are building the fluency that compounds — or waiting for a better moment to start.
The window is open. I cannot tell you how long it stays that way. But I can tell you that from where I am standing right now, in the city where this is being built, the people who understand what is coming are moving with real urgency. That urgency is the most important thing I brought back from San Francisco.
What happens when AI runs a store in San Francisco
The full story of Andon Market and Luna — the AI agent running a real retail store on Market Street with a debit card, a lease, and a $100k operating budget. Read the original reporting.
Microsoft's AI chief gives it 18 months — Fortune
Mustafa Suleiman's full statement and the context around it. Whether you think the timeline is right or aggressive, the direction is not in question. This is the most important thing a sitting AI executive has said publicly in 2026.
Anthropic's research shows AI can already do a huge portion of many jobs — Fortune
Anthropic's own economists on what Claude can already perform at professional quality — and what that means for the future of knowledge work. The company building the technology is now publishing the automation research. Read that carefully.
Claude.ai — Anthropic
I am highlighting this not as a generic "use AI" recommendation but as a specific, grounded observation: the AI powering Luna's store in San Francisco is Claude. The AI Anthropic's economists used to measure automation potential is Claude. If you want to understand what the technology is actually capable of right now — not its marketing, its actual capability ceiling — the fastest way is to push Claude on tasks in your own domain. Not as a chatbot. As a professional peer. Assign it real work. Evaluate the output with expert eyes. The result will tell you more than any article.
Claude Pro: $20/month. The same model running a retail store in San Francisco.
This is the most important prompt I can give you right now. Run it. Be honest with the inputs. The output will be uncomfortable for some tasks — and that discomfort is the point:
I want you to audit my role against AI automation risk — not to reassure me, but to give me an accurate picture.
My role: [your job title]
What I actually do day to day: [list your 5–8 main tasks]
For each task, tell me:
1. Can AI perform this today, and at what quality level?
2. What specifically would a human need to do to remain essential here?
3. Where is the gap between current AI capability and what my role requires?
Be direct. Do not soften this. I need an accurate map, not comfort.
The instruction "do not soften this" matters. Without it, most models will hedge toward optimism. You want the accurate map — even if some of what it shows you requires action.
One question this week — and I mean it as a genuine question, not a rhetorical one:
Is there a task in your work that you assumed was safe from AI — and that you are now less certain about?
I ask because the most valuable signal I can collect from this newsletter is where the honest uncertainty lives. Not the fear, not the dismissal — the genuine "I am not sure anymore." That is where the real conversation starts.
Reply directly to this email. I read every response and reply to most of them.
I started T-Minus AI because I work at the intersection of two worlds. During the day I lead AI enablement across global procurement at Siemens Healthineers — training hundreds of enterprise professionals on how to actually use this technology. After hours I ship products, write guides, and build tools for people who want the same edge without the enterprise budget.
T-Minus AI exists to close the gap I described in this newsletter. The gap between what is being built and what most people know is being built. The gap between the 56% wage premium and the people not yet earning it.
If you are starting from zero, the free Prompting Hub at tminusai.com is 19 laws and 32 cheat codes that apply to any model. Start there.
If you are ready to build a system — the kind that makes every AI session produce professional-quality output from the first message — the Power Guide Pro is the complete operating system. Personalisation, Projects, weekly workflows, model selection. Everything in one place. €59.99.
The countdown is real. I am watching it from San Francisco.
More next week.
Kapish
Enterprise AI Architect · T-Minus AI · San Francisco