The Commoner

Wednesday, October 7, 2026

Sup, freaks.

Somebody put open source rebuilds of seven Adobe apps on GitHub last week. If you do knowledge work for a living, I'd pay attention.

WHAT TO EXPECT IN THIS NEWSLETTER

• AI agents are rebuilding Adobe's apps at a pace that should be a shot across the bow of every knowledge worker.
• BlackRock, Strike and Tether's frozen wallet make the case for spending stablecoins and saving in bitcoin.
• Sullivan's sentiment data says bitcoiners have checked out, which has historically been a good time to stack.
• A cloned voice allegedly sent a SWAT team to Larry Taunton's house.
• Google pays for more nuclear, while fuel costs show up across the services economy.

Marty's Bent

Four Decades and $35 Billion vs. a Week-Old Repo

Here's a post from Eric S. Raymond, the guy who wrote The Cathedral and the Bazaar, that highlights just how quickly AI is flipping the software industry on its head. "Adobe just got nuked. And closed source is dead, dead, dead," he wrote yesterday in reaction to PhotoCraft, an open source rebuild of Photoshop written in Rust. It's one of seven apps the ArtCraft team pushed to GitHub on September 30th and October 1st. The other six go after Illustrator, Premiere, Lightroom, Acrobat Pro, After Effects and InDesign. They're free, they're built for Mac, Windows and Linux, and every one of them is built so an AI agent can drive it. PhotoCraft and PrintCraft are in early alpha and the other five are still in development. The developer says he used Claude Opus 5.5 to build them and hasn't written code by hand since February. ESR assumed they decompiled Photoshop and fed it to an LLM, but a community note on his post pointed out that the project forbids that. Their contributor rules say not to copy code or assets from Photoshop and to match its behavior through public specs and observation.

Adobe was founded by John Warnock and Chuck Geschke in December 1982. Illustrator shipped in 1987, Photoshop in 1990, and over the next three decades they built or bought Premiere, Acrobat, After Effects, InDesign and the rest of what became Creative Cloud. The first version of Photoshop was just over 100,000 lines of code. By 2013 the Computer History Museum said the current version was well over 10 million. Last fiscal year Adobe did $23.77 billion in revenue and $8.7 billion in operating income. It reported 34,206 employees last quarter and was worth about $93 billion at yesterday's close, down from roughly $328 billion at its 2021 peak.

I wanted to get a sense of how much human work went into all of that. If you add up the research and development expense in Adobe's annual filings from fiscal 1991 through 2025 you get roughly $35 billion, and that doesn't count what Adobe paid to buy companies like Macromedia. At a fully loaded cost of $200,000 per engineer per year, $35 billion is the equivalent of something like 175,000 engineer-years, or about 350 million hours at 2,000 hours a year. This is a very rough napkin calculation, not a count of hours actually worked. Plenty of that money went to Adobe's marketing software business and to things that aren't salaries, so cut it in half if you like. You're still talking about the cost of well over a hundred million engineering hours, even if nobody can count exactly how many were actually worked.

Now look at what the PhotoCraft team says about its own pace. Their planning doc says a single session wired up about 70 Photoshop menu items and several major subsystems "in a handful of wall-clock hours using 8 parallel agents." Their tracker went from 224 of Photoshop's 625 menu items wired up to commands on September 30th to 532 the next day. They're upfront that a wired-up menu item isn't the same thing as a feature that works like Photoshop. Their own roadmap says "real Photoshop parity is still well below 50%" and that true one-to-one parity is many months of work. A separate planning doc estimates another 140 to 220 agent-hours to get to roughly 95% of Photoshop's functionality, not counting video and the AI features, and the roadmap itself warns that doc is on the optimistic side. Those aren't apples-to-apples numbers and we don't know how much human time went into directing the agents. But even if that estimate is off by a factor of ten, the gap is absurd.

Adobe spent four decades figuring out what a photo editor should do, and now that the answer is out in the open, a small team with a fleet of agents can work through it at a pace that would've been unimaginable two years ago. This should be a shot across the bow of every knowledge worker out there, whether you're in tech, finance or design. A lot of what we get paid to do is take something that has already been figured out and turn it into a finished product. If you're not using these tools, you're going to fall behind.

The video game decompilation crowd has been doing the same thing to old games. The original Halo is now running natively on Windows, Linux and Android with system link games of up to 128 players, and you can fire it up in a web browser. Completely insane. On the same day ESR posted, Boris Cherny, who leads Claude Code at Anthropic, asked Opus 5.5 to build an interactive website companion for Acquired's Home Depot episode, with watercolor illustrations drawn by Claude. Not long ago you'd have hired a small agency to do that.

I don't think we've even begun to see what happens to the economy when more people start using these tools to build things. Last week I wrote about a16z's State of Markets report, which showed that only 2% of S&P 500 companies disclose an AI metric they track over time. This week a16z put out new consumer spending data showing that in August just 4.5% of eligible consumers in YipitData's U.S. e-receipt panel had an active paid personal subscription to ChatGPT, Gemini or Claude. Across all the consumer AI spending they tracked, the top 1% of payers spent an average of $903 a month while the median payer spent $25. According to a16z, those top spenders look less like regular consumers and more like people buying software to build things and get work done. Imagine 10% of people using these tools as intensely as that small group of power users. What are they going to build?

The models are already doing things that would have sounded like science fiction not long ago. In May, an OpenAI model disproved a conjecture about the unit distance problem that Paul Erdős posed in 1946. Last month Anthropic said Claude wrote the first complete computer-checked proof of Fermat's Last Theorem, working largely on its own for 11 days. Think about how much these tools have improved this year and last year alone, and then imagine another year. Two years. Five or ten years. Things are going to be completely different. This change isn't around the corner. It's upon us, most people don't realize it, and it's going to be discombobulating.

Believe it or not, the Unabomber manifesto talked pretty deeply about this. The passage below from Ted Kaczynski's 1995 manifesto asks the questions we're going to have to grapple with. Are the machines going to take over control and decision making for us, or are we going to control them? Can we even control them once they reach a certain level of intelligence? Will they just keep getting smarter? His argument was that we wouldn't hand over power on purpose. We'd drift into depending on machine-made decisions because they produce better results, until turning the machines off was no longer a real option.

Paragraphs 172 to 174 of Industrial Society and Its Future (1995). The full passage is also reproduced in Bill Joy's 2000 Wired essay.

These are very existential and philosophical questions, and they're going to get answered whether we're ready or not. There will certainly be existential breakdowns and some ugly reactions as people come to grips with what's happening in AI, because these tools are only going to get better.

My advice to anybody reading this is to ride the wave. Don't get too fearful and don't doom. Lean into it, stay on top of it and try to apply these tools to your job, your work or your passion to get ahead if you can. And make sure you stay well grounded while you do it. Stay connected to your faith, your family and your community.

SIGNAL

Spend stablecoins, save in bitcoin

On September 28 I argued that once agents are doing our spending, stablecoins win payments and bitcoin becomes the savings account. BlackRock's September paper, The Machine-Native Economy, landed in the same place. Citing a Bitcoin Policy Institute study where simulated AI model responses favored bitcoin in 79.1% of store-of-value scenarios, it describes "a potential AI-native monetary architecture in which stablecoins serve as transaction money and bitcoin as a store of value." BlackRock calls that evidence preliminary.

Yesterday Strike launched Bitcoin Interest on Cash: eligible U.S. users can opt in to earn a variable 3.6% on cash at Cross River Bank (balances below $102 currently earn nothing), with the interest converted into bitcoin daily, paid out monthly and held by Strike until you withdraw it. And Conduit sued Tether, saying Tether froze roughly $2.76 million of its USDT a year ago on its own initiative. Spend with whatever rail is fastest. Save in bitcoin you hold yourself.

Nobody cares, and that's bullish

Michael Sullivan tracks the language bitcoiners use online, and his latest piece focuses on what he calls desire: how badly people sound like they want bitcoin to pay off, compared with their usual tone. In his data, desire tends to peak late in bull runs. Buying when it was in the bottom fifth of readings did much better a year later than buying when it was in the top fifth. He says it hit the lowest reading in his series in early August, and that the low-desire backdrop was a big reason he was so bullish around bitcoin's roughly $58,000 low in late June.

It's his own metric and he's upfront that he doesn't know exactly why it works. Yesterday was one year since bitcoin's record near $126,000. "People are checked out. People don't care," Sullivan writes. That's usually when you want to be stacking.

Taunton says a cloned voice sent a SWAT team to his house

Larry Taunton described being swatted in March of last year. He says his dog alerted him while he was in bed and he grabbed his gun. This week he posted an update: "According to my latest FBI update, the person initiating the 911 call to trigger a SWAT team response used A.I. to impersonate my voice to say that I had been shot."

Last week I wrote about how convincing deepfakes are getting and what that means for multisig co-signers and fake kidnapping calls. Somebody allegedly used a copy of a man's voice to send armed police to his house. Set up a code word with your family, and make sure anyone who signs for your bitcoin verifies requests some other way than a phone call.

Google is paying for more nuclear

Google and Constellation announced a 20-year deal to add 890 MW of new nuclear capacity to the PJM grid by upgrading 11 existing reactors across Illinois, Pennsylvania and New Jersey. The first uprate is expected by 2028 and the rest before the end of 2032. Constellation is putting in more than $4.3 billion, with Google as the anchor customer. There's a separate 15-year agreement for 2,700 MW from plants that are already running.

Constellation pitched it as a direct response to PJM's "Bring Your Own Power" proposal, and Google says other grid customers "bear none of the associated costs." Same idea as DarkHorse in Hopkinsville yesterday, at a much bigger scale. If you're going to use a lot of power, pay to add supply.

Fuel costs are showing up everywhere

The ISM services survey for September had its prices index at 74, the highest since July 2022. Not one industry reported paying lower prices. Fuel costs were "mentioned twice as often as any other single issue impacting performance," according to the survey's chair.

Saudi Aramco CEO Amin Nasser doesn't see that easing soon. He said nearly 3 billion barrels of supply have been lost since the war began, more than a third of that has been covered by drawing down stockpiles, and the world's "supply resilience cushion is scarily thin." Even after Hormuz fully reopens, he thinks rebuilding inventories could take up to two years. Brent settled at $100.58 yesterday. Higher fuel costs reach a lot further than the gas pump.

⚡ FREEDOM TECH CORNER

DATA SNAPSHOT

Spot price, block height and hashrate estimate: October 7, 2026, 6:20 p.m. ET. Bitcoin Lab daily series: October 6, 2026. U.S. spot ETF flows: October 6, 2026.

Sources: Kraken spot quote, mempool.space block height and hashrate, Bitcoin Lab daily metrics, and TFTC ETF flows. Daily metrics use the provider’s October 6 observation, not live intraday estimates.

Talk tomorrow, freaks.

Marty Bent · TFTC · Nostr

YouTube

Podcast