History suggests otherwise. railroads, telecoms, search all consolidated. The natural equilibrium for transformative infrastructure is winner take all. AGI/ASI won’t be different but will be nearly every vertical and governments will legislate too little too late.
Nothing natural about it. Such monopolies were propped up by the state using public funds and profits captured by the capital class. Many benefitted by the arrangement and so it became normalized. But it’s a choice people made to structure things that way.
The car industry, oil and gas… all could have played out differently if different players had gained wider adoption or if governments used a different economic model.
local models are going to win and therefore the hardware providers, Apple and nvidia.
There isn't going to be any moat for the hosted providers besides hardware scale. They can run your request on shared 1TB memory hardware, or whatever.
But local hardware is going to catch up, the hosted providers are going to become commoditized, and the costs are just going to be compute whether its your hardware or theirs.
And your laptop is going to be powerful enough to be good enough for most cases.
Local hardware catching up doesn’t matter if the thing worth having never leaves the building. Enterprise services are hard, moat is in distribution and know how.
Honestly, music generation is solved. You don’t have fine control at this point but people are unable to tell the difference anymore. There are tons of YouTube videos with blind tests between real artists and AI and people have no idea.
IMO assisted coding (auto complete style) has more flow state than the old days of getting stuck on obscure bugs (as satisfying as those were to crack).
Full agent coding however is the complete opposite, you’re in constant damage control of a junior who moves fast and breaks everything. They’ve got better but still do dumb mistakes.
lot of engineers are discovering firsthand what it’s like to manage a team of eager but useless employees. Not fun at all
> IMO assisted coding (auto complete style) has more flow
I find this impossible to wrap my head around. Any time I autocomplete more than a single line, my flow is destroyed when I have to review what was just autocompleted. The only way I could remain in "flow" in this circumstance would be if I didn't review the code... I'm not ready to trust the LLMs that much yet
The other thing is as far as I can tell, the power-tool style modes where you use AI to boost focus and do quick research is both much cheaper and, by the time you account for all the damage control/prompt tuning/other externalities, roughly as fast as full agentic.
I suspect with the prices going up, that realization is going to be pretty appealing.
Prices are not going up. DeepSeek V4 Pro is 5-10x cheaper than Claude 4.7.
As some of us have been predicting, model capability had already mostly plateaued, and the Chinese have and will continue to relentlessly push cost down. Chinese models will be used for 95% of things, with nation-native models for security/sovereignity-sensitive workloads. Eventually (5+ years from now), efficiency gains and hardware progress will make running local models the dominant way of doing things.
And yes, that puts the investors of Claude and OpenAI in quite a pickle.
I want the frontier on prem to be true but IMO not good enough yet unless async.
What started as all-you-can-eat $50 buffet has quietly become a $6k bill, frontier models that don’t ship your codebase to Beijing don’t come cheap anymore.
Yes, that’s basically what I’m talking about. Without subsidies, that’s a lot of incentive to use more efficient/open weight models, but also to use them in ways that are less compute intensive—fewer tokens, shorter reasoning chains, less nonstop, and generally tie up less hardware for a lot less time.
I agree, I also think there are long term costs to agentic coding that we won't see yet for a while, gradually and then suddenly.
It feels like nobody is even trying to iterate on the "power-tool style" usage of language models, everyone jumped straight to agents. It's not clear to me that removing a human from the loop is strictly more efficient though. Imagine an editor with an embedded language model (either running locally or using cheap cloud models) that is constantly churning, analysing code, reading debug logs, offering suggestions. Refactoring is not b̴e̴g̴g̴i̴n̴g̴ asking a chat interface to make changes, but structured refactors utilising the language models powered by the underlying ast representations to pull off much larger or involved refactors that surpass the abilities of current IDEs. Or doing codegen in your editor in a more structured (and thus repeatable) way compared to an agent spitting out code that you have to then review.
> IMO assisted coding (auto complete style) has more flow state than the old days of getting stuck on obscure bugs (as satisfying as those were to crack).
I strongly doubt that. Flow state happens when you can disassociate thinking and doing. Most of your brain is on the thinking while the doing lags behind. It’s like driving on a well known road. You don’t pay that much attention to the act of driving. Instead maybe you start planning what you want to do on your next stop
Using ai auto complete would be like your well known road, but turned into an obstacle course. You need your full focus on the road. Not that different from someone that is unfamiliar with the road, but not a flow state either. Full agentic coding is you being in the passenger seat with a driver known to be sometimes erratic. Less thinking but more worrying.
There are still BBS you can access via telnet (and actual dial up if you really want), after the fifth one asks you for your full name, street address and phone Humber it gets a little old.
reply