
fun links: @paultoo @paulg @patrickc @lisperati @techemails @arampell @naval @levelsio
Joined January 2009
- Tweets5.7K
- Following3.5K
- Followers502
- Likes12.7K
Pete XL Pizza retweeted
there are only three ways of allocating scarce goods:
-prices
-lotteries
-lines
prices and ONLY prices encourage more supply and therefore DECREASE scarcity. the cure for high prices is high prices.
Pete XL Pizza retweeted
Replying to @jukan05
Adjusted Earnings Before Interest, Taxes, Depreciation, Amortization, Revenue Share with Distribution Partners, and Model Training Cost
EBITDARSWDPMTC
Pete XL Pizza retweeted
If the Chinese government were sufficiently organized, they'd be funding opposition to data centers in the US. But the Chinese government is very organized. So they presumably already are.
Pete XL Pizza retweeted
Replying to @Suhail
The most successful ones. Some are in it just for the money. But for the most successful founders, the company is also their project in the same way a painting or book (or a piece of software) is for its creator.
Pete XL Pizza retweeted
.@HBO series based on Robert Moses when?
Pete XL Pizza retweeted
New article: Oatmeal Spice
I explore what makes a procedurally generated output interesting and memorable and give some tips on tweaking your generator to help.
Link in reply
Pete XL Pizza retweeted
Of my friends who use coding agents heavily, the happiest seem to fall into two distant camps:
a) Controlled fast loops: 1-2min cycles, mostly single-threaded, still totally in control of the code, using the agent "to type faster"
b) Delegated slow loops: Nudging along in the background a couple times a day, while something else (design, writing, etc) is their primary focus; paying ~no attention to code; it's fine if agents get stuck for hours
Like many, I've been trying to make some middle ground work: trying to delegate more than the first camp, while giving the work more focus and technical oversight than the second. My role is planning, technical guidance, and code review. This leads to 10-30min cycles, which lead to parallelism (to avoid busy-waiting), which leads to context-switching and fragmentation, which leads to working memory churn and poor comprehension, which leads to situations where neither the agents nor I understand what's going on.
It sucks and I hate it. I'm moving faster, but the work is unpleasant and unrewarding. It seems awfully hard to exert *partial* technical control—much easier to exert ~full or ~none.
The ideal, maybe, is something like what @simonlast outlines in xitter.gq/simonlast/status/20579…: teams of agents making technical plans, reviewing each others' work, autonomously and adversarially testing, etc. You still get robust "technical oversight"—just by other agents. Unfortunately in my domain (mobile interface with heavy gestures and animation) that's not yet tractable, even with lots of homegrown scaffolds and probes. But things will probably look very different in a year.
I'm curious if others have found happy middle ground between these poles?
Pete XL Pizza retweeted
It's Sunday June 12th, 2016 and the world is mourning the victims of the Pulse nightclub shooting. Trump, who recently clinched the Republican Party nomination, posts on Twitter that Obama should leave office over the shooting.
You walk to a nearby cafe thinking about the possible implications of Move 37 (played a few months prior) and how neat the SpaceX reusable rockets are.
On the way, a mysterious gentleman pulls a newspaper out of his cloak and hands it to you. It's dated June 12th, 2026.
Top Headlines:
- Elon Musk Becomes World's First Trillionaire as SpaceX IPO's
- Department of War Publishes Third Release of Unidentified Anomalous Phenomena Files
- Citing Safety Concerns, Trump Administration Places Anthropic's New 10T Parameter AI Models Under Export Control
Huh? Still holding the newspaper, you look up... but the mysterious gentleman has already vanished. You look down... the headlines are also gone. Sipping your coffee, you think to yourself "Wow, these art students sure are are getting stranger and strfr!"
Before heading home, you find yourself checking the price of BTC... honestly, $600 per BTC seems pretty high. Maybe it's time to sell??
Pete XL Pizza retweeted
“Programming, like writing, is an activity, where one iteratively sharpens what they're doing as they do it. (You wouldn't believe how many drafts I've written of this essay.)”
Pete XL Pizza retweeted
Extraordinary rate of progress in image generation from text. If I were an illustrator I would be a little nervous right now.
arxiv.org/abs/2112.10741
Pete XL Pizza retweeted
I can't help answering my own rhetorical question. I think the reason politicians think you can't get rich without doing bad things is that they treat it as an axiom that economic inequality is bad. And if it's bad to be rich, how could you get there by doing good things?
Pete XL Pizza retweeted
Systems of Record
What is a System of Record, and what makes some stickier than others?
A system of record keeps track of the atomic units of a business. Calendars for resources and people (busy vs free), inventory (in stock vs not, delivery dates, costs), personnel (hours worked, vacation days, and salaries), shareholders (ownership, options, and types of stock), etc.
Some of these items are “static.” Some are read constantly. Some are written and changed constantly. Sometimes there are far more reads than writes, and sometimes the opposite. Some of these are connected to many different “pipes” through APIs or corporate processes; some are used once and virtually forgotten. Some are intermittently read but of tremendous value (e.g., shareholder records!), and some are frequently read but of relatively low value (e.g, conference room calendars).
Sometimes systems of record bundle these things together, in something that we typically refer to as a “vertical operating system.” ServiceTitan, for example, represents multiple systems of record (or a system of multiple records!). An Air Conditioning business must know its customers (CRM), when its technicians (personnel) are free (calendar), and parts (inventory) that are available for sale, and at what prices (accounting).
A system of record for the busy/free status for conference rooms in your Miami office? Perhaps a pain to replace, but effectively costless.
A system of record for tables in your Miami restaurant that grosses $20M a year? Just dropping one hour of records might cost more than the yearly savings of switching to a lower cost system.
And then, of course, there are systems of record for raw data — literal databases that underpin virtually all systems of record. Ask a CTO running dozens of tables with billions of reads and writes per day how hard it is to swap out a mySQL database for Postgres and you’ll elicit visible anger.
The stickiness of a system of record is a function of all of these different inputs, against the cost (often front-loaded/one-time) and risk (lost revenue, profits, and mistakes) of moving. Imagine paying $100/month for software where a $1/month option exists, but the one-time labor cost of switching is $3000, and there’s a risk that something goes terribly wrong. It’s quite logical to keep paying $100/month!
This is part of why greenfields are often superior to brownfields (a16z.com/the-greenfield-stra…). It’s why finding things outside of the goldilocks zone of pricing (arampell.org/2022/12/08/the-…), or outside of acceptable satisfaction levels (too hated), is a possible unlock. And as services become the new battlefield, “migration as a service” could *finally* bend the cost/benefit analysis of switching.
It’s great to be a system of record. I’ve always said that the best companies have hostages, not customers — but the hostages have a lot more tools to free themselves in 2026. Take note.
Pete XL Pizza retweeted
Specific example that came up recently. Had Claude port an old complex gtk3 rust app to iced. Had "working" code in about half an hour, probably 100x faster than I would have done it. Most features worked, felt like "am I releasing this today". LOL, no
This app has a big complex spreadsheet like table widget that subscribes to network data, LOTS of network data. Found that Claude decided it was a good idea to create a new mpsc channel, and spawn a new tokio task FOR EVERY CELL in the table, and then throw all that at a single mutex to update the view.
Both Opus 4.7 and Codex 5.5 reviewed this code multiple times and saw no issues doing this. If it had gone into production it would have exploded the first time someone looked at a table with 1 million rows and 100 columns 😂
That was just the beginning, the api of gtk3 is very different from the api of iced, and that created a lot of opportunity to increase the compile time safety, clarify the code, and generally prevent problems. Neither Claude nor GPT were able to spot most of these opportunities, or if they did, they didn't feel like I told them to make big changes like that, so were conservative in their changes.
Pete XL Pizza retweeted
Principle-driven development (“what would this need to work”) vs. Empiricism-driven software (“this doesn't work but if it did it would do X”)
Pete XL Pizza retweeted
a mentor once told me that there are three kinds of projects in life: two days, two months, or ten years
Pete XL Pizza retweeted
Replying to @nivi
Humans ask good questions because they create good problems.
Unlike LLMs (for example), humans formulate the problems they wish to solve.
One problem humans like to solve is lack of novelty.
Humans are wired to create novelty (its own reward), frontier-seeking.
Not sea-faring, but frontier-faring!
Pete XL Pizza retweeted
Replying to @levelsio
The US spends more (as a percentage of GDP) on social welfare than China by a large margin, about 2X.
The total government spending is about the same, but in China it goes into infrastructure and megaprojects. Very little handouts for the people.
And no, it's not military spending which is only around 3% of GDP vs 1.5% of GDP.
Pete XL Pizza retweeted
Equation editor with *color* and *interactivity*: p.migdal.pl/equations-explai… - from @pmigdal
You can edit with Markdown + LaTeX, and then export to HTML, LaTeX, Typst, Beamer.
Then in the HTML page you can hover over any section of the equation *or* text and get an explanation.




















