Even if this was written "by"/with the help of AI, the point the post is actually pretty spot on.
You can use AI to create slop using very simple, uninformed prompts, but you can also "collaborate" with it to write thoughtful pieces with good points.
One point not made here though that I think is one of the most important about the space: these companies are being valued like software companies but defense work doesn't produce software-like margins and it never will.
A company like Anduril being valued at $60+ billion is insane. Most of the major defense primes in the US trade at around 2x sales. Supposedly Anduril has a few billion in sales.
For comparison, Northrop did $42 billion in sales last year and has a market cap of around $75 billion. Even factoring in growth, which reasonably deserves a premium, there's really just no way to get to some of these valuations in the defense market because it's structurally limited on the margin side.
Even companies that primarily sell software to the US Government don't get software like margins. Functionally, the business of selling software to the US Government scales like plumbing and not like commercial/consumer software. A plumbing business can only stack so many plumbing jobs into its plumbers, so to scale it needs to hire more plumbers. Most software businesses scale beautifully because the first copy is really expensive and all future copies are somewhere between free (distributed over the internet) and very cheap (shrink-wrapped box with a CD inside). But software companies that sell to the US Government are actually selling developer time to the US Government, not finished products. So that scales with the number of developers you can hire.
And that actually makes sense. Even for something like hospital management, the government is going to be different. Epic's EHR system can sell to bunches of hospitals and amortize most of their development work across all of those hospitals- there is only VA system and it is very different from how even a large hospital system functions (no insurance claims! lifetime tracking!), so adapting it is going to be painful.
At this point there's no reason to read things when in most cases these articles should be summarized to 5 sentences (which is probably what the article's prompt was)
For those who haven't worked in defense, a lot of contracts are "cost plus" meaning the company's profits are legally limited to 15% of declared expenses.
> You can use AI to create slop using very simple, uninformed prompts, but you can also "collaborate" with it to write thoughtful pieces with good points.
If you're working that closely with the piece you plan on publishing then you can go ahead and pick your own words too. This is a social skills test. You need to signal respect to your readers if you want them to engage with your output. That sense has clearly been eroded, or perhaps never developed, in the author if they consider this style acceptable.
Do you realize how much content today is being produced using AI in some fashion? Even for editing and concept development.
You can spot the tell-tale signs even in major publications.
You're free to throw the baby out with the bath water if you'd like, but it's increasingly unrealistic to stop reading the second you encounter a suspected AI footprint unless you're willing to just stop reading.
> You're free to throw the baby out with the bath water if you'd like, but it's increasingly unrealistic to stop reading the second you encounter a suspected AI footprint unless you're willing to just stop reading.
Strongly disagree. It's incredibly nice, actually, for the mask to come off and get such an easy negative signal on low-effort content. There's more quality content than I could ever read in a lifetime, and it constitutes the slimmest fraction of all content that I will encounter. LLMs are highly effective honeypots for people who want to waste my and others’ time and attention.
Spam bot
Claude slop. "durable base", "That is real money", "those are requests, not appropriations", etc etc.
Even if this was written "by"/with the help of AI, the point the post is actually pretty spot on.
You can use AI to create slop using very simple, uninformed prompts, but you can also "collaborate" with it to write thoughtful pieces with good points.
One point not made here though that I think is one of the most important about the space: these companies are being valued like software companies but defense work doesn't produce software-like margins and it never will.
A company like Anduril being valued at $60+ billion is insane. Most of the major defense primes in the US trade at around 2x sales. Supposedly Anduril has a few billion in sales.
For comparison, Northrop did $42 billion in sales last year and has a market cap of around $75 billion. Even factoring in growth, which reasonably deserves a premium, there's really just no way to get to some of these valuations in the defense market because it's structurally limited on the margin side.
Even companies that primarily sell software to the US Government don't get software like margins. Functionally, the business of selling software to the US Government scales like plumbing and not like commercial/consumer software. A plumbing business can only stack so many plumbing jobs into its plumbers, so to scale it needs to hire more plumbers. Most software businesses scale beautifully because the first copy is really expensive and all future copies are somewhere between free (distributed over the internet) and very cheap (shrink-wrapped box with a CD inside). But software companies that sell to the US Government are actually selling developer time to the US Government, not finished products. So that scales with the number of developers you can hire.
And that actually makes sense. Even for something like hospital management, the government is going to be different. Epic's EHR system can sell to bunches of hospitals and amortize most of their development work across all of those hospitals- there is only VA system and it is very different from how even a large hospital system functions (no insurance claims! lifetime tracking!), so adapting it is going to be painful.
At this point there's no reason to read things when in most cases these articles should be summarized to 5 sentences (which is probably what the article's prompt was)
> it's structurally limited on the margin side
For those who haven't worked in defense, a lot of contracts are "cost plus" meaning the company's profits are legally limited to 15% of declared expenses.
Yeah, but if you didn't take the time to write it why should I read the full thing? Here's summary:
> Defense-tech startup valuations vastly outpace the real, contestable market.
> As valuations meet procurement reality, mergers, recaps, rollups, and acquisitions will reshape the sector
> You can use AI to create slop using very simple, uninformed prompts, but you can also "collaborate" with it to write thoughtful pieces with good points.
If you're working that closely with the piece you plan on publishing then you can go ahead and pick your own words too. This is a social skills test. You need to signal respect to your readers if you want them to engage with your output. That sense has clearly been eroded, or perhaps never developed, in the author if they consider this style acceptable.
Do you realize how much content today is being produced using AI in some fashion? Even for editing and concept development.
You can spot the tell-tale signs even in major publications.
You're free to throw the baby out with the bath water if you'd like, but it's increasingly unrealistic to stop reading the second you encounter a suspected AI footprint unless you're willing to just stop reading.
> You're free to throw the baby out with the bath water if you'd like, but it's increasingly unrealistic to stop reading the second you encounter a suspected AI footprint unless you're willing to just stop reading.
Strongly disagree. It's incredibly nice, actually, for the mask to come off and get such an easy negative signal on low-effort content. There's more quality content than I could ever read in a lifetime, and it constitutes the slimmest fraction of all content that I will encounter. LLMs are highly effective honeypots for people who want to waste my and others’ time and attention.
> There's more quality content than I could ever read in a lifetime, and it constitutes the slimmest fraction of all content that I will encounter.
And yet here you are engaging in an apparently pointless discussion about a post you apparently wouldn't care to read in the first place.
Such is life
The author is literally a VC partner
That explains why they have such contempt for their readers.
My mental model suggests that makes it more likely to have been crapped out by Claude.