Some thoughts, because I write to think.

• Tech is about to enter a proper wealth extraction phase. One of my annoyances with “enshittification” is the lack of imagination it implies. Tech, as an industry, can be much more extractive than this. There’s ways to the bottom yet.
• Most large tech software products are, effectively, extremely shoddy front-ends to highly reliable back-ends. Look at Twitter. Their systems have taken what’s now two years of abuse where it’s now run with a fraction of its earlier workforce, but the overall engineering and design has held up. It has degraded massively and the front-end is the same clusterfuck it has always been, but the lesson CEOs took from it is that once your employees have put in the work of engineering a reliable back-end, they become disposable.
• The problem is that many of them are right – these platforms and products can run for a very long time without incident – and CrowdStrike taught them that you can have a software failure that literally disrupts the entire world’s economy without it being completely destroyed in the process.
• “Studies” like the one where Amazon claims that developers only spend one hour per day coding which, even if it were true only highlights how small the “coding” part is of the overall job, is just laying groundwork for using “AI” tools as an excuse for layoffs. We will absolutely see claims that LLMs improve productivity tenfold used as an excuse for mass layoffs.
• “AI” is only a part of the picture, but it’s already considered by management to be an effective enabler of downsizing. but I think it’s likely that it is going to disproportionally hit front-ends (web, iOS, Android) over servers, but the plan now is clearly to put the actual software development part of software into a managed decline and use LLM tools to juggle spurious “feature” additions to maintain a veneer of liveliness in the products.
• Much of the developer ecosystem on any given platform is downstream from big tech. Open source is dependent on either company goodwill or employees donating their free time – something they have less of when unemployed. Platform improvements are historically a big enabler for small- to medium-sized software businesses. Recruitment and corporate training has been the lifeblood of most developer education products. Much of the rest is dependent on tech worker disposable income or freelancer operating costs. If tech is turning properly extractive, it’s likely all of that will fall away over a few short years.
• Ongoing extraction is likely to make cascading failures like Crowdstrike more common, but tech companies know it’s survivable and for those with monopolies or oligopolies, is actually more likely to benefit them than harm. (Cascading failures are an opportunity to use captured regulators to set rules that benefit your business.)
• From rare earth materials, to battery components, to chip-making, to assembly, tech’s hardware supply chain is highly reliant on a globalised marketplace. A trade war will disrupt this. The only question is how much.
• Tech is now a much more overtly political project. Billionaires realised that they stand to make a whole lot more money if everything is deregulated. Between social media sites outright manipulating their products to favour a political party and their owners funding right-wing takeovers, it’s clear that both the organisations and products of tech exist to serve the political agenda of their owners.
• This shift is enabled by Trump’s win, but not dependent on it. Tech is a global political tool.
• This is worrying because the pressure to move our entire economies to centralised tech platforms – various forms of “cloud” software – means that those same economies are now under more direct centralised political control.
• More worrying is the widespread corporate adoption of generative models – irrespective of their effectiveness. Only a handful of companies actually make these models and the companies adopting them have neither the capability to properly vet them nor the interest in building those capabilities. This means it’s likely that in the near future all corporate writing will be mediated by a handful of language models controlled by a small number of organisations who all have an explicit political agenda.
• Even though you can’t force an LLM to stop fabricating nonsense or stop libelling random people, you can easily shift sentiment. If the political collaboration removes the worry of lawsuits or regulatory action, the companies themselves are free to manipulate language wholesale. If corporate adoption of language models is as widespread as it seems, then these political entities could change how businesses everywhere write – for example – about gay marriage, gender equality, or a woman’s right to vote. As long as it doesn’t have to be factual, it’s relatively simple to cause a model to be more negative about specified key words or phrases and more positive about others.
• Effectively, you assholes have given a handful of CEOs a racism and bigotry dial for the world’s English-language corporate writing.
• All of this puts those of us who work in tech and have a conscience in a pretty bad place. Act on your conscience and you’ll get laid off. Speak out and your freelance gigs will dry up. Those that remain will be in competition with thousands of newly laid-off tech employees stumbling into freelancing as an interim solution to their financial woes. Education products will rely mostly on small- to medium-sized businesses and freelancers instead of the much more lucrative big tech recruitment and training pipeline. Dev tools will shift into extractive automation if they are maintained at all.
• The dev tool ecosystem is another point of centralised control. Between GitHub, npm, TypeScript, and Visual Studio Code, Microsoft effectively owns modern software development.
• Amy Hoy has been right all along, basically. Sustainability comes from outside mainstream tech. “This is why my mission and my business since 2009 has been to help people create ethical “tech” businesses (businesses that use technology) that actually help people for a reasonable profit”. Serving the customer directly in ways that are conscious of their vulnerability and unlikely to be co-opted or controlled by a big tech companies is a radical act.
• Not only is it a radical act, it might increasingly be the only sustainable path left for anybody with a conscience who wants to continue to work in tech or a related field.

Went out and followed the local ravens around for a bit. A few more juveniles have joined the ones that arrived the other day. I’m guessing there are at least ten currently in town.

They love to perch on streetlights.

They were also very good about flying past the sunset.

Finally, starlings.

I’ve just started up a Patreon account…  Let me tell you about it. Since you’re here reading this, you already know me as the author of the Steerswoman series  (four books so far: The Steerswoman, The Outskirter’s Secret, The Lost Steersman, and The Language of Power). And you probably already know that I’ve been stalled […]

Here’s a fabulous three-song set by Gillian Welch and David Rawlings.

Nachruf auf Lothar König: Pfarrer und Punk, 1954–2024.

GIMPS Discovers Largest Known Prime Number: 2^136,279,841-1, first new Mersenne Prime found using GPUs.

Formalising Graphs with Coinduction, by Donnacha Oisín Kidney and Nicolas Wu (POPL 2025). “We argue in the paper that graphs are naturally coinductive, rather than inductive, and that many of the problems with graphs in functional languages go away once you give up on induction and pattern-matching, and embrace the coinductive way of doing things.”

The Design of a Self-Compiling C Transpiler Targeting POSIX Shell, SLE 2024 paper by Laurent Huberdeau, Cassandre Hamel, Stefan Monnier, and Marc Feeley.

getopt() but smaller, neat version by Tony Finch.

“Reflections on Trusting Trust”, but completely by accident this time, by duk.

Neural Networks (MNIST inference) on the “3-cent” Microcontroller, a PMS150C.

A line map for Milano’s trams in TeX, by terru.