It's like reporting that Spirit of Halloween employees accounted for the biggest retailer layoffs in November. It's a totally useless statistic that tells you nothing about what might really be going on that month that's abnormal compared to baseline.

You realize there are fully autonomous waymo cabs operating at "level 4" in several regions... today. Anybody can download the app and ride in them now.

Google didn't go around promising this "next year" every year for the last decade. They just went out and did the work to make the product work, and released it.

The local government has a degree of autonomy, but there are unspoken checks on it by a generally hostile state government. Nashville is represented by Democrats while the state is controlled by Republicans. There is an implicit threat that if the city fails to pursue the state's objectives the state will punish it.

Recently the city angered the state because it refused to hold the Republican convention. The state retaliated by passing a law to shrink the Nashville city council and another to take control of its airport. The state also took control of the Nashville tourism board that has a slush fund paid for by city taxes.

There is a direct line between state reps and local politicians. So for example when state lobbyists were angered at the loss of parking due to a bike lane they called up a council member to have the lane removed.

Which is all to say that the stadium deal was ordered by the state and not something the city had a real say in.

If the datacenter people go and cozy up to the governor or legislators the state will lean on the city again and they will ram it through as well.

So what might look like success in this environment might look very different if the bubble pops.

Obviously Oxide has a ton of experience on its side but it's so opaque where any of this goes that even they can't read the tea leaves.

It feels like semantics here. The US empire hasn't been defeated but its specific invasions have been.

But just knowing the US military is too weak to accomplish its objectives doesn't help much, because the US leadership is so mercurial it might decide to attack anyway. The deterrent effect only helps if you have rational leadership who is capable of evaluating the chance of failure.

Repelling such an invasion, while possible, is still a pyrrhic victory for whoever the US is attacking, because so much of their population will suffer in the invasion regardless.

If you didn't notice a difference between AI assisted coding in Jan 2025 and now beyond "marginal" improvements... well to be frank I don't think you've really been paying very close attention. It's a combination of the way we interact with the models and the models themselves, and this really did get much more effective in late 2025.

You can say the model improvements in that year were incremental, but the degree to which the overall capabilities have changed and what that means for software development is more than the sum of those parts.

So transcription can definitely be helpful when part of a broader, intentional process, while also being insufficient to do much on its own.

And the next logical question to ask is whether there's a better form of intentional behavior that might be more effective.

[0] https://en.wikipedia.org/wiki/Ludington_Pumped_Storage_Power...

This seems like a truly wild prediction to me.

LLMs can synthesize data in useful ways for all sorts of purposes, and I can see how they could be helpful in the process of a human editing and refining a work of fiction they are creating, but I can't imagine why I'd ever tolerate a work of fiction actually generated by an LLM.

The authorship and craft matters, the translation of human feeling and thought to the page matters. This isn't code where it either does what it's supposed to or does not.

Fiction from a machine is of no value.

Google acting like a normal but competent company that's just chugging away, meanwhile the hype cycle is propping up its upstart competitors to truly ludicrous valuations.

Buy and hold real estate, maybe?

I don't think you can really short this because you can't time it. But you can move assets to things that are 1) concrete 2) have utility and 3) that the US gov't won't allow to lose real value for political reasons.

The machine is really good at working the spec on its own now, which is amazing, science fiction shit. But you've still got a garbage in, garbage out problem at the end of the day, which is pretty much the only hope we who work in software have of remaining somehow employed.

And to elaborate on the stupidity, blue states send more taxes to the federal government, which then doles them back out to subsidize the low cost of living in red states.

So you end up with people in dense urban areas sending tax dollars to Washington, then begging to get the taxes their own citizens paid back, in support of building the infrastructure projects they want locally.

Like a lot of his peers, the first Trump administration seemed to unlock a permission structure to be increasingly reactionary in public, facilitated by the overt political bias of Musk's X. We've now reached a point where this sort of "vice signaling" is a great way to gain clout in the overtly nativist tech bro niche in the US.

Which is to say, he wasn't always like this, and it feels like each new blog post is getting worse in a way that, if I were at Basecamp, I might have initially overlooked as the frog was slowly boiled.

I feel like you can still be an AI maximalist while thinking the numbers simply don't add up.

The problem as I see it is that this is a zero sum game. Nobody has a moat. Even if AI eats all white collar work, where's the profit for these entities? They become commodities and they will compete with each other (and open weight models) only on price, not to mention the workers who lose their jobs to this stuff aren't going to be contributing to the economy at all.

It uses cloud-init for the initial setup and injects the agent's config file and some other stuff in a ro mount from my laptop. That single subdirectory is mounted rw into a conventional location inside the VM. I inject only credentials to e.g. test environments that don't matter if they're exposed. The agent context instructs it that it can use full sudo and install any utilities from apt that it needs.

So for the ux I do incus-dev-vm.sh shell or incus-dev-vm.sh opencode etc.

A fun thing about this is I used nested virt to have the agent write the script - so I spooled up an incus VM myself first, then the agent can run with full privs inside there creating nested VMs to ensure its functionality.

Right now I'm not filtering anything on the VM - with my threat model only limited access is granted and I don't give creds to touch real infra. But you should be able to filter it on the host (I haven't researched the best approach however).

I feel like the most pragmatic perspective is "trust but verify."

This is why they're so effective at coding: you can run the code yourself (or the test suite) to verify that it actually does what it's supposed to.

And maybe these people finding hallucinated results on Rachel's site are doing verification too.

It hardly matters that it's vibe coded and ugly if it works well... and it does (for me).

Saving a few CPU cycles and reducing RAM utilization on my local system isn't going to meaningfully improve anything in and of itself; the wall time is all bounded by the model or by the code it's invoking (i.e. compiling or running the test suite).