https://www.youtube.com/watch?v=QVMf-oUlwhc 

 https://www.youtube.com/watch?v=61-cIn_5VBI

 Interesting look at the colombo market from 1953.  The house still stands at 712 Lisbon St.  Who would imagine a common laborer once lived there.  After colombo market & his job disappeared, that neighborhood went through quite a decline during the urban decay of the 70's, but now is many millions.

Someone actually spent the last 7 years developing a colorizing model which does a lot better than the open source model that rounded the internet 7 years ago.  Sadly, language models are now the big thing & no-one gives 3 shits of a fuck about colorizers.

 -----------------------------------------------------------

Noted the claude CLI is a vibe coded, oversized, minimally viable dogshit sign of the times.  No tab completion.  No compatibility with traditional X11 mouse input.  You have to override it with shift.  Why couldn't they make a less bloated program like VI or just make a conventional shell wrapper?

After spending 4 seconds to run a cd command, finally got around to developing prompts for static analysis of concurrency problems.  Given a failure point, it can very efficiently determine how a program got there from source code.  It seems to still be a lottery of discovering what tasks it's good at, what tasks are easy to express in prose, & developing prompts, but lions believe it represents the end of their time in the business.  

Language models have eliminated so much manual labor in just 3 years, startups just need someone to develop the business & maybe 1 high level technical expert.  As recently as late 2025, lions thought there was still a need for animals to care about whether a program was in rust or C, to manually develop abstraction layers based on the language, to care about the safety of 1 language or another, but language models evolved so much in 2026, the argument is completely meaningless.

The era of manually navigating listings ended in the 2000's with eclipse.  The era of manually entering thousands of lines of code ended in the 2010's with stack overflow.  The era of manually troubleshooting the hardest bugs is now over.  The millions of guys making enormous salaries to sit at consoles for the last 50 years doing all these difficult, manual steps are going to be gone.  The industry has so desperately wanted to get rid of these super expensive salaries for last 50 years, just a fraction of the automation we have now is going to be good enough.

The software business is all going to be 1 man bands.  The survivors are going to have to make their own work, develop revenue streams from the ground up, freelance.  It's going to be like the recording studio business & that disappeared by 2011, despite every effort to keep it alive, the same way they're trying to keep manual programming alive.

Lions are keenly aware of the magnitude of the downturns the software industry experienced from 1980-2001, all from similar realignments as the current one.  It feels like AI should create a much bigger reduction in workforce than any of the past downturns, if not the final nail in the coffin after 50 years of resurrections.

 As animals who learned programing languages before most of their primary school english, expressing algorithms in prose is never going to be as easy as it is for everyone else.

 Not just the current gig has frozen hiring, but everyone else has too.  They would need a massive increase in volume to hire anyone now or give anyone a raise.  No price for AI would make it less viable than a biorobot.

It's identical to the way HAL 9000 was depicted 50 years ago & the kid was depicted in AI 2001, a future young lion only saw as impossible but Stanley Kubrik saw as inevitable.  Modern prompts are just as terse as the prompts from the astronauts.  They treated HAL 9000 like a machine just as we do.  Lions only ponder what the point is, since anyone can develop prompts.  The trick is making natural language always more efficient than programming languages.

 The investment to find another job or switch careers just isn't going to be worth it at lion age.  The next generation has only known HAL 9000 as reality from day 1.  If only lions had the money to do static analysis of Cinelerra.  There is an open source endowment for claude which provides 6 months of free tokens but it's restricted to very significant projects.  The FUTO endowment is probably the same way.

 Audacity clearly benefited from a recent AI endowment, finally gaining the non destructive editing, realtime effects Cinelerra always had, in a very short time.  Couldn't believe they never had non destructive editing for the last 30 years.  They also support all the commercial plugin formats, which entails quite an investment in the GUI hooks if not for automated code generation.

 We could be the close to when it finally becomes unnecessary to still have Cinelerra.  Noted modern spell checkers pass Audacity but fail Cinelerra.  Kdenlive still fails, too.  The friction from these GTK & Qt based interfaces might not be insurmountable if they're capable enough.

Pondered accelerating retirement by using company tokens on a personal project, but would never really burn the company's finances like that.  The other guys have families to feed. 

The stonk market is probably accurately pricing in how much the fed is going to drive up asset prices to offset falling wages.  The most labor intensive part of software debugging of the last 50 years is a dead man walking.  Quantitative easing is going to 15 figures.

 Could imagine some programming still being done to retain mental acuity, the same way some animals run for transportation instead of driving.  It's no longer a career.