Posts by Paul Thaler-Williams

We won't even see the next era of business coming. One day it will just appear. We look backwards for inspiration. Forwards for escape. Recently, I was inspired by an answer from Roman Mars of 99% invisible to the question "which design hill would you die on?" His response: "efficiency makes society worse." All our attempts to carve the perfect process leave us with hyper-efficient, but inhuman monoliths. People, the real elements of society, get left by the side of the road. Valencia's Juan Roig, well-known for his entrepreneurship, had a similar response to the Classic American Capitalist perspective that calls for endless streamlining working towards the capitalist ideal of running a billion-dollar company with a single employee. Roig's approach was instead: maximum employment. The highest good is the highest engagement with society. Is it efficient? No. Is it human? Very much yes. I would like to see business become more human.

https://hashatar.io/ Hashatar is an experimental project to represent information, such as bank numbers and crypto addresses, as small, easily comparable images for quick visual validation. A hashatar could be printed on an invoice, displayed in a banking app, or integrated into a crypto wallet to represent critical information. Like a hash, it doesn't reveal the underlying information, which could be sensitive, but it changes radically when small changes are made to the input. This shows mistakes in circumstances when data must be reproduced faithfully either by copy-pasting, or by entering data manually. The inspiration came from a recent attack on an NPM package, which substituted any crypto addresses it found for visually similar ones, attempting to trick users into sending crypto to the attacker's address. I see Hashatar as a new standard for sensitive data entry.

I believe that European leaders sense that inaction in the face of Russia does not have a positive outcome. However, action without buy-in from the US is also not worth contemplating, so acquiring that buy-in is of primary importance. Recent reports of drone incursions into Poland and Romania have resulting in an NATO Article 4 invocation by Donald Tusk. Presentation in the media is amplified compared to similar incidents such as the errant Ukrainian missile in Poland in 2022, during which Article 4 invocation was also discussed. In a 2025 interview, Andrzej Duda, when asked if Zelenskyy had pressured him to say that the missile had been Russian, replied "you could say that". Various casi belli are continuously laid at the feet of the Western public in the hopes that they percolate up to Donald Trump, although it does not seem likely that he will give his blessing to such action. Perhaps soon, a casus belli will be found that makes war with Russia a moral imperative.

Bob Nystrom wrote a famous post entitled "What Color is Your Function?" - https://journal.stuffwithstuff.com/2015/02/01/what-color-is-your-function/ It discussed the tension in software between how sync and async functions behave. A new category of data is emerging that I call "purple". This is produced by LLMs. The property is transferable. Any data or function that uses LLM-generated data is purple, and any data or function based on the same is also purple. Keeping track of AI provenance is crucial going forward, as is providing checks on the data produced by purple functions. So, I present a new approach: https://llmsmr.io It summarises text using an LLM, but also provides tools to quickly check the result by extracting direct quotes and presenting them to the user. The summary is purple, but the quotes can be verified simply to ensure that they do indeed appear in the original text. I believe this approach should be applied to all purple data going forward.