[blululu]

Out of curiosity how much of this is a manifestation of the utility of LLMs? I get the current political impetus right now but also the barrier for swapping out an infra stack was also much higher 2 years ago. From my own projects major swaps are now relatively trivial which means that vendor lock in is weak.

[grey-area]

Precisely none of it is related to LLMs. It's related to the political situation and the possibility of trade war and tariffs.

[Scarblac]

And actual war, given the threats to Greenland.

But especially that Microsoft blocked the work email of the ICC lead prosecutor for political reasons, that has all alarm bells ringing.

[tcp_handshaker]

Given how bad the US military performed against Iran, its pretty clear that any hostilities started by the US against NATO, would finish with a takeover of Washington within...2 weeks...

[hvb2]

What makes you say that?

Where does this weird expectation come from that you can basically achieve your goals in a 2 week air campaign?

If the US wanted to they could probably do a lot of damage but, as you can see in Ukraine, taking over a country is a whole different thing. Unless you're willing to go in with the army and are willing to lose a LOT of people. And even then it'll take months or years for a single country

[antonvs]

> Where does this weird expectation come from that you can basically achieve your goals in a 2 week air campaign?

From every US administration that’s started a war since about 1945.

[Scarblac]

Even if that fantasy was true in any way, the US can do enough damage to European digital systems to utterly cripple society on day 1.

[crote]

And of course the actions in Iran and Venezuela, demonstrating that even the most braindead threats aren't just empty bluffing.

[embedding-shape]

Besides some companies that were deep into the weeds of AWS and been pushed to enable and use every single AWS service by their reps, I don't think it's much harder/easier today than it was two years ago.

Sure, LLMs help a bit with the actual typing, but the hard part is still planning, alignment and actual execution, all which are best done by humans talking and working with other humans.

I don't think we're in the days of "Hey Codex migrate our AWS stack 100% to Hetzner VPS stat" yet, without issues along the way. Wouldn't claim it's impossible either, but again the easy parts were already easy, and the hard parts are still hard.

[ghaff]

Architecting to make portability easier should absolutely be a thing. But people in the weeds of a specific public cloud provider today will absolutely need to make tradeoffs between getting to a position where they can be more portable and devoting those resources to other things.

[embedding-shape]

> Architecting to make portability easier should absolutely be a thing.

It is, but it's really hard to make the right trade-offs. Usually I'd start by basically making a list of what could potentially be done to make it easier, check lightly how hard each one of those will be, then sit down with stakeholders and figure out the balance between how fast they want to move, and how risky they want the migration to be. Some opts for less safeguards and moving sooner, others for absolutely less risk but it takes the time it takes.

[vga1]

Some. Many companies have relied in the past on the fact that doing things is freaking complicated. Such as maintaining your own services instead of using something from a provider like AWS.

What used to be a half year transition project that will be half-assed due to resource constraints, can now be properly done in a month by a skilled engineering team works on it with LLM leverage.

Of course, if America was still a trusted ally (like if Harris was president), we still wouldn't be doing it. Even if it was easy now.

[ambicapter]

How do LLMs help with the mechanics of switching infrastructure stacks? Does writing code faster make infrastructure swap easier?

[riccardomc]

zero.

It is a manifestation of the commoditization of Cloud Computing Interfaces.

Every provider offers a blob storage, kubernetes clusters, queues and what not.

I'd argue that covers 90% of SaaS needs.