Tax Code Change Caused Mass Tech Layoffs.
Also OpenAI’s Paperwork Nightmare and an Open-Source Tax Prep App.

The Annual Budget Meeting Where Accounting Changed The Law Of Physics

The most catastrophic business failure of the modern tech era was not a server crash, but a line item on a spreadsheet. Tech companies—including giants like Microsoft and Meta—have been shedding hundreds of thousands of jobs since 2023, and the quiet culprit is a delayed amendment to the U.S. tax code's Section 174. This provision, originally slipped into the 2017 Tax Cuts and Jobs Act to make the bill look cheaper on paper, finally went into effect in 2022, forcing companies to spread R&D deductions—like programmer salaries—over five years instead of deducting the full cost immediately.

In the grand theater of corporate office drama, this is essentially a change in the expense report policy that has bankrupt the entire department's hiring budget. A company that previously spent $150,000 on an engineer could immediately write that off; now, they can only claim $30,000 in the first year. The remaining liability, for many startups, is an immediate tax bill where there used to be none. So, when the CFO looks at the books, the easiest way to make the numbers work is to simply make the engineers go away, turning what was an incentive to invest in domestic talent into an active punishment for growth. It is a masterpiece of bureaucratic incompetence, perfectly translating the language of financial regulation into the universal language of a pink slip.

The Chatbot Must Keep All Its Doodles

OpenAI is currently in a very public spat with a U.S. federal judge because it was ordered to preserve all future ChatGPT output log data, which is essentially forcing the company to keep a permanent record of every single text message the AI generates. The order comes as part of a copyright lawsuit brought by *The New York Times*, which wants to see proof that the AI copied its content verbatim. This mandate is problematic for OpenAI because it overrides its own privacy commitments to users—the ones who clicked the 'Delete Chat' button and assumed the company’s internal shredder actually worked.

In a remarkable display of legalistic paper-pushing, the court has decreed that every thought the AI has had, even the ones marked "Temporary Chat" or "Privacy Mode," must now be saved to a special, tamper-proof folder for the lawyers to look at later. OpenAI objected, naturally, pointing out that this runs counter to a long list of international privacy laws and that keeping petabytes of sensitive chat logs for months is apparently an engineering task of baffling complexity. It appears that having a system that deletes data is much harder to build than a system that simply makes things up.

The Government Finally Open-Sources Its Internal Knowledge Base

The Internal Revenue Service (IRS) has caused a minor stir by dropping the source code for its pilot tax preparation software, Direct File, onto GitHub for the world to inspect. This move is a victory for the concept of government transparency, but a potential migraine for anyone who thought the tax code was complicated enough already. The service utilizes a proprietary declarative system called the "Fact Graph," which attempts to translate the entire Internal Revenue Code into a set of logical questions and answers, and is written in the Scala programming language.

The good news is that taxpayers can now verify exactly how the government calculates their liability—if they are one of the estimated 3% of the population who can read both Scala and 26 U.S.C. without having an immediate breakdown. The bad news is that the whole process feels less like true public service and more like the Treasury department decided to run its crucial, billion-dollar intranet project on the same shared drive the interns use. The repository is technically open-source, which means a truly dedicated enthusiast could theoretically spot a bug in the code that says, "You owe $7,000," and submit a pull request changing it to "You owe $0."

Briefs

• Internal Tooling Update: The ubiquitous FFmpeg project has quietly finished merging WebRTC support. It’s like the IT department finally approved a new version of Adobe Reader that doesn't constantly crash your browser.
• Avian Intelligence: Cornell's Merlin Bird ID tool remains a success, proving that the only practical use case for modern AI is definitively identifying a common backyard species by sound.
• Confidentiality Policy: Tesla is aggressively trying to prevent the public disclosure of self-driving crash data sought by The Washington Post, citing "competitive harm". Apparently, seeing how many times the car failed to notice a concrete barrier is a closely guarded trade secret.