[HN Gopher] Wrapping my head around AI wrappers
       ___________________________________________________________________
        
       Wrapping my head around AI wrappers
        
       Author : nowflux
       Score  : 45 points
       Date   : 2025-11-15 17:44 UTC (5 days ago)
        
 (HTM) web link (www.wreflection.com)
 (TXT) w3m dump (www.wreflection.com)
        
       | jgalt212 wrote:
       | > But Cursor and other such tools depend almost entirely on
       | accessing Anthropic, OpenAI and Gemini models, until open-source
       | open-weight and in-house models match or exceed frontier models
       | in quality.
       | 
       | I'm not sure I agree with this because even though Cursor is pay
       | north of 100% of revenues to Athropic, Anthropic is selling
       | inference at a loss. So if Cursor builds and hosts its own models
       | it still has the marginal costs > marginal revenues problem.
       | 
       | The way out for Cursor could be a self-hosted much smaller model
       | that focuses on code, and not the world. This could have
       | inference costs lower than marginal revenues.
        
         | bogzz wrote:
         | I suppose supermaven is d'oing something to that effect.
        
         | xnx wrote:
         | Can you have a useful code model that doesn't understand the
         | world? It seems like such a model would be limited to little
         | more than auto complete.
        
           | esafak wrote:
           | I imagine so, through distillation. Start with an all-knowing
           | model, then extract the coding part.
        
         | simianwords wrote:
         | what makes you so sure of this?
         | 
         | > Anthropic is selling inference at a loss.
         | 
         | cost of models have gone down dramatically over time.
        
           | jgalt212 wrote:
           | True, but they keep using more tokens (Agentic FTW!) and the
           | currently most expensive models.
           | 
           | source: https://www.wheresyoured.at/
        
       | mentalgear wrote:
       | > But I think the insight lies between these positions. Even if a
       | new application starts as a wrapper, it can endure if it lives
       | where work is done, writes to proprietary systems of record,
       | builds proprietary data and learns from usage, and/or captures
       | distribution before incumbents bundle the feature.
       | 
       | Basically the same as MS & Social Media did: build a proprietary
       | silo around data, amass enough data, so it will become too big an
       | inconvenience to move away from the first provider.
       | 
       | It's good that the EU has laws now to ensure data
       | interoperability, export & ownership.
        
         | jrvarela56 wrote:
         | I agree with you in spirit but this harms the potential for
         | these new products to emerge. You're saying you don't want them
         | to be able to accrue a data moat. It sounds good for user
         | privacy and optionality later on but it makes it harder for
         | these services to get started as they dont see that model as
         | possible.
        
       | swyx wrote:
       | i recently framed this as "agent labs" vs "model labs" -
       | https://latent.space/p/agent-labs - definitely far from proven or
       | given that they are a lasting business model, but i think the
       | dynamic is at least more evident now than it was a year ago and
       | even that is notable as we are slowly figuring out what the new
       | ai economy looks like
        
       | _fat_santa wrote:
       | Currently working on a SaaS app that could be called an "AI
       | Wrapper". One thing I picked up on is once you start using AI
       | tools programmatically, you can start doing far more complex
       | things than what you can with ChatGPT or Claude.
       | 
       | One thing we've leaned heavily into was using Langgraph for
       | agentic workflows and it's really opened the door to cool ways
       | you can use AI. These days the way I tell apart an AI "Wrappers"
       | vs "Tools" is what is the underlying paradigm. Most "wrappers"
       | just copy the paradigm of ChatGPT/Claude where you have a
       | conversation with an agent, the "tools" are where you take the
       | ability to generate content and then plug that into a broader
       | workflow.
        
         | embedding-shape wrote:
         | > One thing we've leaned heavily into was using Langgraph for
         | agentic workflows
         | 
         | Probably my single biggest mistake so far with developing LLM
         | tooling so far has been to try to use Langgraph even after
         | inspecting the codebase, because people I thought were smarter
         | than me hyped it up.
         | 
         | Do yourself a favor and just write the plumbing yourself, it's
         | a lot easier than one might think before digging into it, and
         | tool calling is literally a loop passing tool requests and
         | responses back and forth until the model responds, and having
         | your own abstractions will make it a lot easier to build proper
         | workflows. Plus you get to use whatever language you want and
         | don't have to deal with Python.
        
           | CjHuber wrote:
           | I can recommend looking into DSpy, I haven't felt the need to
           | use any other LLM based frameworks though I'm open to
           | suggestions
        
             | embedding-shape wrote:
             | There really isn't need, all they add is additional code to
             | be responsible for, building the same abstractions yourself
             | but focused on your use case will be something like 50-100
             | lines of code, hard to beat the simplicity, and the
             | understanding you'll get.
        
       | TrackerFF wrote:
       | What do we call companies that spin up open-source ML models, and
       | basically just sell access to said models?
       | 
       | One example is audio / stem separation, object segmentation.
       | 
       | They're not wrappers, but whatever that is one step deeper down
       | in complexity.
        
         | jabroni_salad wrote:
         | I think those are just inference providers unless they are
         | adding something on top of it.
        
       | lubujackson wrote:
       | Almost every startup is a wrapper of some sort, and has been for
       | a while. The reason a startup can _startup_ is because it has
       | some baked in competency by using new and underutilized tools. In
       | the dot com boom, that was the internet itself.
       | 
       | Now it's AI. Only after doing this for 20+ years do I really
       | appreciate that the arduous process and product winnowing that
       | happens over time is the bulk of the value (and the moat, when
       | none other exists).
        
         | amelius wrote:
         | > Almost every startup is a wrapper of some sort, and has been
         | for a while.
         | 
         | The difference is that with AI they will send your data to a
         | third party.
        
         | max002 wrote:
         | Cant help myself and compare to frameworks, libraries and
         | oop... cant we built so fast because of them?
         | 
         | I think of wrapper more as a very thin layer around. Thin layer
         | is easy to reproduce. I do not question that a smart collection
         | of wrappers can do great product. Its all about idea :)
         | 
         | However its if ones idea is based purely on wrappers there's
         | really no moat, nothing stopping somebody else to copy it
         | within a moment
        
       | btown wrote:
       | OpenAI is just a wrapper around NVIDIA, which is just a wrapper
       | around TSMC, which is just a wrapper around ASML, which is just a
       | wrapper around Zeiss optics, which is just a wrapper around EUV
       | photons, which are just wrappers around quarks, which are just
       | wrappers around quantum fields...
       | 
       | A Large Language Model is just a Large Hadron Model with better
       | marketing.
        
         | random3 wrote:
         | The processor is _built_ of transistors, _built_ of silicone.
         | The paper that _wraps_ the box that _wraps_ the processor is
         | simply a mindless container. Yes, there's nuance when it comes
         | to "wrapper" companies but it in the end they may just be
         | wrappers. Back in the "web 2.0" things were called mash-ups and
         | everybody didn't try to make them look like companies.
        
         | esafak wrote:
         | I would argue that a "wrapper" denotes a product that could be
         | replicated with little effort, not that one relies on the
         | other. None of your examples fit this definition.
        
           | M4R5H4LL wrote:
           | I believe the controversy arises from the notion of "little
           | effort" and critics who have never independently pushed
           | anything to market. It comes across as dismissive and
           | arrogant simply because someone exudes excessive confidence
           | in a limited set of skills. I can personally attest to the
           | immense demands of building a successful business, and it's
           | evident that very few individuals possess the capability to
           | achieve that. Therefore, while it may provide comfort to
           | avoid challenging oneself and dismiss others' total work, it
           | ultimately doesn't benefit anyone and feels more like a self-
           | serving "I could, but I never will" attitude.
        
             | chris_pie wrote:
             | The potential customer rarely cares whether a service
             | provider is running their business well. What matters is
             | the product's value added and risks added, as compared to
             | just using the underlying tech directly.
        
       | keiferski wrote:
       | Marketing, UI, and brand matter a lot. Especially when all of the
       | products are functionally "the same" to the average consumer, who
       | doesn't care about technical details and benchmarks, etc. It
       | reminds me of this great scene from Mad Men:
       | 
       |  _This is the greatest advertising opportunity since the
       | invention of cereal. We have six identical companies making six
       | identical products. We can say anything we want._
       | 
       | https://youtu.be/8SsnkXH2mQY?si=SWPOsGBel1yh3kMd&t=198
        
       | darepublic wrote:
       | as others have mentioned -- I think wrapper is a fair term. It is
       | not trivial and took untold man hours of research and labour to
       | go from nvidia gpus to modern llms. some of the ai products
       | really do feel like minimal engineering around calls to openai
       | (or claude or what have you)
        
       ___________________________________________________________________
       (page generated 2025-11-20 23:01 UTC)