Munching our popcorn (while we still have fuel to heat it up) & waiting for Trump to crash and burn ...
We've discussed before the AI bubble, its seemingly inevitable forthcoming bursting, and the system-crashing potential it might hold. Well here's something to give anyone pause:
 |
| Source: CHARTR / date from Bloomberg |
It's pretty remarkable to me that Uber needed such a long runway - surely its only major assets are software? Tesla looks pretty outstanding value & commercial discipline[1] by comparison - and Musk was shooting for the moon, with serious hardware involved, as well as software.
But why am I gazing at the foothills? Because then we step back and look at the mammoth on the mountain, and - wow! What type of revenues do they "forecast" - and persuade investors of - to make that even vaguely sane? And OpenAI isn't alone out there. What kind of economy can afford their products on the necessary scale?
Sometimes you have to look at something - a bullion price spike, a US president - and just tell yourself the obvious conclusion is the correct one. That Is Insane.
And the US economy[2] has been wagered on this? Hello China - the rest of the century belongs to you.
ND
PS: for something more edifying on AI, try this:
__________________
[1] One would need to interrogate that tapering-off of the black bars: was it Tesla that bought X? Had greenie car buyers lost faith in Musk even before 2025 and his rogue DOGE behaviour?
[2] And I'm the one that has said, many times: in the long run, never bet against the US economy ...
20 comments:
@ND - Every Uber ride ran at a loss for years to undercut existing taxi firms, for the longest while it seemed profitability for them was a dream.
As for the bubble bursting, prior to that we're seeing an exodus from private finance of software companies, fearing AI will undermine them.
I still reckon the bubble will pop next year, although Trump looks to be accruing all the crises he can get before the midterms, so who knows.
CH
In the long run, we are all dead..... but you already knew that
The thought occurs - what will Israel do when US/China hits the inflexion point, yuan becomes global currency and Beijing (see - what happened to Peking?) becomes the new Washington? They'll never get the power over the Politburo that AIPAC have over Congress. Seems to me they should be turning all their efforts to sustaining the States, not encouraging them to burn their treasures in Eastern Europe and the Middle East.
"Trump threatens to stop arms supplies to Ukraine, if Europe does not help him in the war with Iran - FT"
UK and EU really are being squeezed.
Small thing on Uber but a few year back had someone senior renting a flat I kept in Berlin and they gave me their contract as part of the tenancy checks. All blunt language, no legalese, could have been written by a teenager into rap music. Main point was to worm into politics and build the monopoly, little to do with software. Open AI may try to aim for dominance but it'll still be left open to competition.
Bob - your story reminds me of the instructions given by Team Blair to a new ambassador headed for Washington:
"your job is to get right up POTUS's arse - and stay there!"
There seem two parts to AI - will it result in 'Einsteins in a box' and will it push white collar wages down or out. Either way looks good to Capitalists. In the short term there is liitle else to invest in and there is FOMO. Whether AI does do the Einstein bit is mildly interesting but what counts and seems to be working is pushing white collar wages down and out.
That is where the money is made - pushing wages down and eliminating jobs. Which leaves governments to pick up the tab but governments can't do much about that. Rock - Hard Place, bless. Now about that 'Next Big Thing?'.
Settle in for a long miserable summer and lay in some candles and logs for next winter.
The AI boom has basically been created to satisfy the God complexes of the borderline psychopaths who run the US tech industry. You only have to read the BS these people write about what they want to do to society to realise they are bat shit insane. And because (so far) they have made a lot of money with their techy stuff, the finance bros have just said 'How much money do you want?', without bothering to ask if its a good idea in the first place, let alone going to work.
AI could be the US's Darien Scheme.
Interesting piece in the Economist:
"Why AI has not yet upset India’s IT industry ... Deploying the technology is the real world is proving tricky"
https://archive.ph/LCiFh
@anon 10:45
I use Claude quite a bit, and if you just ask to build something out of the box, you'll get something.
What you have to do is first give it guardrails against how you code, what patterns it is allowed to do, etc. in order to make things readable and maintainable. This can be done via installing a Skill, or just having a document for it to read.
You then need to add several more files explicitly telling what not to do.
With over a quarter of a century learning from my own screwups, and indeed others, I'm quite good at giving it those rails from inception to continual delivery and testing.
You then need to define some actual requirements - what are the actual outputs - along with grouped tasks. This is the bit where *regular* software development falls down, let alone using AI tools.
Once you have those in place, you then iteratively improve them against the first few tasks until you get it outputting what you want in general. You then start iterating through what you actually want.
It's a lot of preliminary work to do, and it'll take a good 50-100 hours to get it at an acceptable state. If you've planned it out, you can get one person doing that and everyone can use the results, but realistically, it's going to be your whole team because virtually no one is doing the planning bit. Here's some funky new tech, use it and deliver stuff quicker...
From a development point of view, it's also fascinating watching it work, but all the guardrail setting up is tedious, but turn it into a Skill and you need only do it the once. The problem is the dev no longer really codes, they become the Product Owner (thus removing one job) and basically write up requirements in a manner Claude will consume sanely. It's very boring.
My next step is to setup something like NanoClaw, so I can do all this via email. One agent will me email me about a project, and then back and forth about requirements, until I'm happy, then another agent will trigger Claude's agents, along with remote control, and I can then run a whole development stream from my phone, one that runs 24/7, token expiration allowing.
If you don't know how to optimise your Claude messages, you can quickly chew through tokens & come up against usage limits. (Grandmother. Suck. Eggs.)
Anthropic have apparently just 'accidentally' released the entire source code for Claude. That should do a lot for business.....
https://www.theguardian.com/technology/2026/apr/01/anthropic-claudes-code-leaks-ai
Take a look at the share price for SAP, INFOSYS and CAPGEMINI over the last 5 years. I reckon AI will rattle the cages of the programmers. Sure some roles will change and ride the AI boom but the overall direction of travel is likely down. If AI is any good at all it will make corporate IT much more generic.
I hold no candle for Reddit but they have an intriguing blog on the Civil Service. If half of what is visible is true the CS has and is bought into every tickbox management BS ology going. Looks a truly ghastly poisonous environment captured by every kind of consultancy fad. Very expensive and nothing good will come of it.
It's beautiful to watch. Watching it talk through a debugging process is particularly satisfying.
I've been more on the product owner side than the engineering side, though always closely integrated.
I reimplemented my favourite productivity app (Todoist) core functionality in 5 minutes. Quality of life and whistles probably took another hour. (Though I quite quickly worked out that I needed to start applying some development rigour, backups etc as part of the iteration process as it routinely wiped the data without migrating to the new schema.).
I was blown away I had something working, and useful so quickly.
Harder yards to come with integrations etc.
The "coding" part of engineering will go away almost entirely, especially for boiler plate work.
@ND it goes a little deeper than that. For truly hands-off working, on top of your agents writing code, you'll have other agents analysing pull requests, yet more deploying and testing... The costs mount up even with well designed prompts.
Add in combatting context rot, and what looks to be simple at a top level begins to get increasingly complex.
@RS - I don't think it will get rid of coding, easily replicable code maybe, but we've had mechanisms to do that for years and most places don't, so maybe not even that.
Add in a lot of work is brownfield, and AI is almost useless at that at scale. Munged together monoliths, where no one has bothered with separation of concern, and every layer is riddled with hacks, and objects that don't belong there, is guaranteed to confuse an AI.
I still believe cost will be the eventual undoing of AI, with people either using less capable models with less value, or a negative ROI with the good models.
Sure, AI bubble will burst, same as the dotcom bubble did, internet and internet companies never to be seen again, LOL!
The dotcon bubble did burst.
As AltaVista. Boo.com. And that Dutch one that went from 60bn to 12bn in a weekend.
As BQ says, a lot of the pre-dotcom bubble businesses are no longer around.
Will AI disappear? Probably not, but given the exposure, and funny money financing schemes (e.g. nVidia handing OpenAI a few billion to buy nVidia chips) that have a whiff of the Crazy Eddie's about them, I'm not sure if it'll pick back up like the internet business did.
The internet being a network, the startup costs are low, AI, not so much, even if you use something like ollama, you'll still need plenty of time + money to train it on anything hefty.
Yep, so what happened to the internet and internet companies afterwards?
Post a Comment