Hacker News Reader

Comment by socratic_weeb | Hacker News Reader

This industry's complete outsourcing of its core business value on a third party proprietary subscription based tool, Claude, made by an unprofitable company, Anthropic, is very concerning. You are all lunatics, sorry.

View on HN

I don't follow this line of reasoning. Would it have been meaningfully different if OP had used open-weight models like GLM or DeepSeek? Does it really matter considering we'll have superior models next quarter?

The meaningful difference is that you will not experience EOS if (when) anthropic/openai/etc fails to become profitable and is no longer subsidized by capital funding.

While "vibe-coded" apps do help lots of people who didnt have the time/money/skills to create their projects, you should be aware that currently the compute is being subsidized so that users become reliant/used to the service.

I believe vibe coding in general is a bad business strategy whether you use FOSS models or not because it means your product probably doesn't have any "secret sauce" (leaving code maintainability problems aside). By that I mean a carefully researched innovation that gives the edge to your product. Nevertheless, using FOSS models is clearly better for the reason you mentioned. I believe serious businesses will transition into using AI on-premise in a very restrictive manner (eg: AI only for tests and reviews policy, etc.). We'll once the dust is settled.

Internal business tools are not innovative products. The potential edge comes from things like being better aligned with the business process or eliminating tasks from the process.

You can inject "secret sauce" through your domain knowledge or life experience.

You could vibe code the "tedium" out of your app with little to no care about it using AI while paying close attention to the critical aspects of your product. Of course, the fact that all of your AI code usage is being monitored by the company that provides you the model/harness is also still means they can just steal your product whenever they want

Stricter use would remove the primary benefit while not really giving much upside so I don't think companies will move in this direction

> Of course, the fact that all of your AI code usage is being monitored by the company that provides you the model/harness is also still means they can just steal your product whenever they want

Also anything which isn’t kept private can quickly be cloned. I think it’s going to be hard for a SaaS to stay profitable unless there’s a real-world tie-in to keep someone from pointing a bit at your app and cloning the observable behavior with just enough changes to claim they didn’t.

The question is if your "secret sauce" will stay secret as you use Anthropic's products, considering they've been launching specialised models like Claude Legal and Design. What happened to Figma with Claude Design should be a warning sign.

You're making the mistake of assuming that just because LLMs make things easy, all of the value is in the LLM and not in the prompting and glue work.

Put on the hats of a risk manager or supply chain manager. Yes, that would have mattered. What you insource today, you can repeat tomorrow. If you know the price of an input will multiply in the near future, you should at least develop processes that can handle disruption. If you expect dirt cheap LLM’s to remain available, or even better: improve in quality, feel free to make that assumption explicit and keep using the external supply. I don’t know what the future holds, but the vibe your software organizations cannot go back to manual development once LoC has exploded. I hope firms are making these choices deliberately.

You seem to be referencing as core business value a combination of (1) the ability to maintain a (2) business criitical resource, but stating it as if catastrophizing from a hasty generalization. Can you clarify?

It's really not that simple. A lot of what's involved is fungible. Of course, the answer to "Was the greatest intelligence harnessed to make the greatest decision?" Is always no.

AIs can be swapped for one another, run locally, and implemented in ways that are less prone to loss of function. Loss of anyone who understands the working code is a type of risk, but that kind of issue tends to bounce between losing skill from lack of foresight due to economic savings, and overhiring / bringing back lost employees as consultants.

The industry is a joke.

Meta requires tens of thousands of engineers to maintain a social media site. Google even more for an ad platform.

Never have so many achieved so little and the joke is all those clowns think they are “10x engineers”. Meanwhile WhatsApp got to global scale with less than 30 people (before Meta bought it and piled on the inefficiency).

Vibe coding is many orders of magnitude more efficient than the industry standard and that’s why it’s so disruptive.

I think the comment said they were outsourcing it to "whatever".

Reply on news.ycombinator.com