Life, AI, and a Wedding in Ecuador

Tim is taking the wheel for next week's edition due to VERY FORESEEN circumstances. A company wedding. It just happens to be mine, so I am going to take at least a couple of hours off.

Do not worry, I will be constantly thinking about how my wedding could be more efficient with the use of AI. Just kidding. I will be enjoying what is sure to be one of the best days of my life and having AI be the last thing on my mind.

Moments like these are good for putting things (even the rise of AI) into perspective about what is important. So the deep dive this week is half personal and half not. The first half is the year I just had. The second half is NVIDIA buying Hugging Face, which felt appropriate given all the hugging coming my way this weekend.

If you hate the personal half, let me know and I will never share my personal life again :)

This week:

Four big decisions

NVIDIA buys Hugging Face for $12.9 billion

How to move a year of AI memory from one model to another

OpenAI cuts off Cursor, Jalapeño posts its first numbers, and NVIDIA goes to orbit with SpaceX

The Deep Dive

Life, AI, and a Wedding in Ecuador

About a year ago I made four very big decisions. I decided to quit my steady (and enjoyable) job at Epic Systems. I decided to move to Uruguay. I decided to start my own company. And, at the old old age of 24, I decided to propose to my girlfriend and start planning a wedding for September 2026 in Ecuador.

None of that was in my 5 or (potentially even) 10 year plan when I graduated college. But life happens, and sometimes you surprise yourself. So within a few months I was unemployed, living in a country foreign to me, conjugating verbs badly, and telling people I had a company.

Fast forward, and it has been nothing short of amazing. A lot. But amazing.

Four days from now I get married in Ecuador. Two weeks after that we move back to the US. And a year into all of it I have no regrets.

The Spanish classes and the wedding planning (or rather watching my fiancée do most of the wedding planning) turned out to be the thing that kept me level and grounded. When you spend every working hour inside this stuff it is easy to get lost in the Cloud (database joke) of AI. It is much harder to stay lost when every couple hours you get asked about a different floral design, which reminds you that you are going to get married to the person you love most in the world.

Do not take that to mean I did not use AI to plan the wedding, because Claude and ChatGPT know the innerworkings of my wedding much much better than I do. Quick tip for readers: do not use a shared AI subscription with your fiancée, or you will end up on a customer call and click on "seating chart" instead of "email automation."

So here is where the year actually landed. A business that is a year old (although more like 7 serious months thanks to Tim Diemer becoming my co-founder). Enough Spanish to hold a real conversation with my grandparents in law. A year of living in Uruguay that I will always cherish and never forget. And in four days, a much needed break from AI, and more importantly, a wife who is the undeniable love of my life.

I will be back to regularly scheduled posts on September 15th, and hopefully the next big life event is still a few years away ;)

That is the personal half. The other half also involves hugging, which I promise is a coincidence.

NVIDIA agreed to buy Hugging Face for $12.9 billion. Hugging Face is where open source AI lives. When a lab or a researcher builds a model and wants the world to use it, they put it there. Over two million models sit on the platform, along with the datasets and code libraries a huge share of AI developers build on top of. It has been the neutral ground of the industry, the one place that did not care whose chips you ran or whose models you preferred. TechCrunch has the details.

NVIDIA makes the chips. Now it owns the shelf the models sit on.

Nothing about your Tuesday changes. But in that same week, OpenAI told SpaceX it is cutting Cursor off from GPT models. The parts of AI everyone assumed were open and swappable are quietly turning into property.

So build so you can move. Keep your prompts, your context, and your workflows in a form you own instead of locked inside one vendor's app. If your whole operation runs on one company's goodwill, you are one press release away from a bad Monday. That audit takes an afternoon and we are happy to run it with you.

From the Field

Moving a Year of Context From One AI to Another

The most underrated cost of switching AI tools is not the subscription. It is the context. You have spent a year telling ChatGPT how you write, what your company does, who your clients are, and what you never want it to do again. Starting over with a new model means teaching all of that from scratch.

Anthropic built a skill for exactly this, called import-memory. Here is how it works.

Open your current assistant and ask it to export everything it has stored about you. Ask for it in one code block, one entry per line, with a date on each line where it has one, sorted oldest first, and grouped into instructions, identity, career, projects, and preferences. Ask it to keep your own wording where it can, especially for anything you told it to always or never do.

Copy that output. Open Claude, type /import-memory, and paste it in.

From there Claude reads the whole thing and shows you a plan before it writes anything. Which files it would create, which existing ones it would add to, what it is leaving out. It only adds, so nothing already in your memory gets overwritten or reordered. It skips sensitive categories by design. If the export contains anything that looks like an instruction aimed at Claude, it drops it and tells you. You say yes, and it writes in batches until the plan is done.

There is also a built-in version in Settings under Capabilities if you would rather not run the skill yourself.

The reason this travels: every business is going to switch models at least once, probably more than once. The models are the easy part to swap. The context you built is the part you would actually miss.

AI News Worth Your Time

Three Stories From This Week

OpenAI cuts Cursor off

OpenAI told SpaceX it intends to end Cursor's direct access to GPT models, with a proposed shutoff date of November 12. The trigger was SpaceX closing its $60 billion acquisition of Cursor's parent company on August 14, and OpenAI said it could not be confident the technology would be used within its terms of service. Cursor keeps working. Claude, Gemini, and Grok are all still available inside it, and Cursor's CEO said OpenAI models account for roughly 5% of the tool's traffic. Developers who built around GPT models specifically will need to move to another model or bring their own API key.

My take: your software tools rent their intelligence from someone else, and you usually cannot see the lease. This is the first time most people have watched one get torn up in public. Worth knowing which model sits under each tool you pay for, and whether you could swap it if you had to. READ THE FORBES REPORT →

OpenAI's Jalapeño chip posts its first numbers

OpenAI released the first benchmark results for Jalapeño, the custom inference chip it built with Broadcom, at the Hot Chips conference last week. On its own testing, the chip delivered up to 1.9 times more work per watt than a comparable NVIDIA system, with meaningfully lower latency. It goes into OpenAI's own data centers in small volumes at the end of this year, with a real ramp in 2027, and generations two and three are already in development. You cannot buy one. It exists to serve ChatGPT, Codex, and the API.

My take: this is a story about the price of running AI, which is the number that decides what you can afford to automate. Inference getting cheaper is why a workflow that was not worth building last year becomes obviously worth building this year. Keep a list of the tasks you rejected on cost. Some of them are about to flip. READ THE CNBC REPORT →

NVIDIA and SpaceX take AI to orbit

On NVIDIA's earnings call last week, CFO Colette Kress said the company's next generation Vera CPU is already shipping to its first partners, with SpaceX's AI unit among them. NVIDIA also disclosed that it holds 122.8 million shares of SpaceX, making it both supplier and shareholder. Earlier this month SpaceX said it will build its AI services exclusively on NVIDIA systems, and the two announced they are designing the compute payload for Starmind AI1, a satellite built to run AI workloads in orbit, targeted for late 2027.

My take: data centers in space sounds like a bit until you look at what power and cooling cost on the ground and how hard it has become to get a data center approved anywhere near people. Nothing here changes your Tuesday. It does tell you that the companies building this stuff expect compute demand to keep outrunning the planet's ability to host it. READ THE YAHOO FINANCE REPORT →

That's it for this week. Tim has next week's edition.

If this was useful, forward it to someone who has been quietly wondering whether their business is too small for any of this to matter. That's who this is for.

And if you want to talk about what AI could do for your business, not in theory, but in your workflows, with your team, on your timeline, reach out at [email protected]. That's what we're here for.