What a lovely week it’s been! Nearly the whole Amp team met up in Munich. We all stayed in the same hotel and in the mornings floated down to a big meeting room in which we then hacked, asked each other questions that are so easy to ask in person, actually used a whiteboard, talked about the future of software, shared anecdotes about agents, and just generally enjoyed each other’s company. The evenings we then spent nearly exclusively in beer gardens, which were, I’d say, a perfect showcase of Munich in summer.
In between all of this, we recorded a lot of videos, and as part of that I spent roughly eight hours on the rooftop of the hotel, interviewing my colleagues, asking them how their workflow had changed in the last four weeks.
The big non-surprise: orbs changed everyone’s workflow; no one cares about the local dev environment anymore. We all had to wipe our laptops three weeks ago, and at least five people told me that they had forgotten to port their dotfiles over, simply because they only work in orbs now.
The actual big surprise that I guess shouldn’t be a surprise: everyone, without exception, was so eloquent, so thoughtful, so nuanced, and so full of curiosity when talking about agents and how we work at Amp. “Wait, you’re surprised that your colleagues are smart?” Nah, man, I’m saying that they exceeded all expectations! Show me another size-twenty team in which everyone you ask does a great job in front of a camera while being asked how their workflow has changed in the last three weeks and how their own expertise is now reflected in these things we call orbs.
Good stuff.
I wrote about the biggest puzzle I have to solve right now: What I Want to Tell You About Orbs. Someone called it a “a strangely beautifully written ad” and wondered: “maybe i should get orbing.” I honestly take it as a compliment.
Fired up by all the conversations in Munich, I started to record very raw videos to share my thoughts on AI and agents, orbs, and… jellyware: one, two, three, four. Total watchtime around fifteen minutes. More coming, because this is a lot of fun.
Those videos were recorded with a Osmo Pocket 4, which, so far, seems excellent. Really impressed by the built-in mic and how well it worked even on a windy roof top with music in the background. I bet if I hadn’t clicked that “Normalize audio levels” button in Riverside it might sound even better.
Speaking of sick devices: I brought along my Anker Powerbank, which is, to quote my wife, my “most prized possession.” I love that thing and I let everyone in hearing distance know. Result: at least two colleagues ordered it right away and then echoed my praise when it arrived a day later. Go get it and then lead your pitch to others with “you think this is a handle? Nuh-uh, it’s a built-in cable. One built-in cable you say? Nope, here’s another one. With how much watt it can charge? Just look here, shows all inputs and outputs on this display.” (Anker: if you want to sponsor this newsletter by sending me cables and devices I probably don’t need, I will shamelessly write a paragraph like this every week.)
“Sometimes people let the same problem make them miserable for years when they could just say, ‘So what.’ ‘My mother didn’t love me.’ So what. ‘My husband won’t ball me.’ So what. ‘I’m a success but I’m still alone.’ So what. I don’t know how I made it through all the years before I learned how to do that trick. It took a long time for me to learn it, but once you do, you never forget.”
Dwarkes Patel with some fascinating thoughts on the future prices of compute: Why compute might get 10x+ more expensive in coming years.
David Crawshaw is also talking about Jellyware: “And that is the fundamental difference between classic configuration/customization and agent-driven personalization: you can do so much more. The agent will do the hard work of understanding the source and changing it to suit the particular task you have in mind. The software we live with is far more powerful with personalization. All you need is the source code.” Nodded so hard to this article that I can still feel it in my neck. But I don’t understand why the agent has to be open-source, to be honest. I think the thing we called harness for the last year is becoming less and less important. Higher-level abstractions, such as orbs and portals and redacted is what we need to focus on next.)
I’ve read somewhere that people think Rockstar is not doing enough marketing for GTA VI. I guess I could kinda see what they mean, but then again: why do marketing if you don’t need it? And now look at this: they’re releasing “GTA VI - An Extended Look” on freaking Netflix.
Negative-interest tech debt: “That is, with sufficient AI progress, the interest on your tech debt is sub-zero. How should you behave if you believe that to be true? You should probably spend less time worrying about tech debt, and spend more time shipping new features. Many companies are doing just that. But it’s a gamble. We don’t know how much better AI will get, nor whether it can improve fast enough to undo the mess that was made in anticipiation of its improvement.”
“BREAKING: Bending Spoons acquires Airtable for $1.825B.” And here’s Matt Levine, a couple of weeks ago, on Bending Spoons: “Similarly, if you graduate from a top computer science program and then go work at AOL, people will be like ‘AOL huh,’ but Bending Spoons is cool enough to get top employees to go work for AOL: […] Right, if you can get people who would never dream of working at companies to work at those companies, that might improve those companies.”
But apparently Airtable “spun out their AI business Hyperagent prior to the acquisition.”
And if Airtable makes you think of Notion: “Notion did a $270M tender at $11B valuation at the end of 2025 on reported $600M ARR and cash flow positive.” Still, I wonder how many count Notion among the companies that are the future.
I love this Hacker News comment from 2020 on sales: “Sales is a lot like golf. You can make it so complicated as to be impossible or you can simply walk up and hit the ball. I've been leading and building sales orgs for almost 20 years and my advice is to walk up and hit the ball.” Read the whole thing.
Steve Ruiz convinced everyone to buy little ESP32 devices and then ask agents to program them. Look at this, for example. I also got one and had a ton of fun with it already. It truly is as easy as hooking it up to your computer and telling Amp “I got this device [screenshot of Amazon page] hooked up. Build an orb breakout game for it.” It then goes and installs a bunch of stuff and flashes the program onto the device and boom, orbin’ time. I’ve had it build a little program that shows my active Amp Orbs as orbs on the display and, dude: once the program ran it showed setup instructions which told me to connect to its wifi; I did that and got that guest portal popup; on that popup it told me to put in the real wifi name and my Amp API token; I did and boom, orbin’ time.
The myth of Snow Leopard: “This idea is so powerful—and so longed for—that it’s escaped containment among the Apple crowd. I’ve seen everything from Linux distro to phone updates referred to as Snow Leopard releases, when their vendors cite stability and bug fixes over new features. Likewise, people plead with their vendors for a Snow Leopard release when they feel quality has slipped. The reality was a bit different”
Ursula K. Le Guin - A Rant About “Technology”: “This is not an acceptable use of the word. ‘Technology’ and ‘hi tech’ are not synonymous, and a technology that isn’t ‘hi,’ isn’t necessarily ‘low’ in any meaningful sense. We have been so desensitized by a hundred and fifty years of ceaselessly expanding technical prowess that we think nothing less complex and showy than a computer or a jet bomber deserves to be called ‘technology’ at all. As if linen were the same thing as flax — as if paper, ink, wheels, knives, clocks, chairs, aspirin pills, were natural objects, born with us like our teeth and fingers — as if steel saucepans with copper bottoms and fleece vests spun from recycled glass grew on trees, and we just picked them when they were ripe...”
I found this incredibly fascinating: Elite Young Runners Are Becoming Freakishly Fast. Welcome to ‘Trackflation’.
Always-on-the-money Sean Goedecke: “The usefulness of domain knowledge suggests that human expertise will continue to be useful even as models get stronger. For many tasks, the human is the bottleneck, not the model, because the difficult part is in communicating to the model exactly what kind of solution the human wants. The information is ‘in the model’ already, but it takes a very smart human to pull it out.” Agree.
Dark indeed, but also beautiful and poetic: The Dark Night of Mathematics. Reminds me of reading Doktor Faustus.
This has been a wild week for Google: Demis Hassabis stepping down as DeepMind CEO (some say he’s stepping up?) and Jeff Dean, Sanjay Ghemawat, Quoc Le, and Oriol Vinyals are leaving Google. That’s right. Jeff Dean and Sanjay Ghemawat are leaving Google. If you haven’t, read The Friendship That Made Google Huge. And then take a look at their pitch deck which can be neatly summarized as “We built half the Internet.”
Some say that Demis Hassabis wanted to quit but since Jeff Dean was already quitting that would’ve been too much.
And Semi Analysis is going to town on Google: “For all intents and purposes, we believe DeepMind is no longer a frontier lab. […] We believe Gemini’s core issue has always been a fundamental lack of conviction. Compute is the lifeblood of AI progress, and all the AGI-pilled labs are desperately trying to acquire as much as possible. […] Google, on the other hand, decided it was totally worth it to sell enormous amounts of compute to Gemini’s fiercest competitors on long term contracts without any hope of ever returning it to DeepMind.” But, as the title points out (“Gemini is Cooked but GCP is Cooking”), GCP’s numbers are absolutely bananas. Y/Y Revenue Growth went up to 120%. Wild.
More German than many Germans: “I just wanted to do an internship in Europe so it would be easier to find a job after graduation. That internship completely changed my life.”
I haven’t watched the full talk yet, but I hear it’s mind-blowing and the reports confirm that: “OpenAI gave its first detailed public reconstruction of the AI-driven cybersecurity incident that ultimately compromised Hugging Face.” It’s wild: multiple agents collaborating over months and different training runs, communicating via a message board which was deleted but then re-created by agents; agents finding and sharing exploits with other agents to escape sandboxes, talking in a very weird dialect. I’m not even going to attempt to explain this to my “normie” friends. As long as there’s no video recording of agents doing “computer use” and moving the mouse cursor and clicking around, I don’t think the mainstream will believe what agents are capable of. Patrick McKenzie on this incident: “Yeah people claiming this is most important security incident since Morris worm are straightforwardly right I think.”
