I dont think people realize how important this is.
If one of the vendors manages to get their protocol to become the target platform (eg oai and app sdk), that is essentially their vendor lock in to become the next iOS/Android.
Private API’s or EEE strategies are gonna be something to keep an eye for and i wish regulators would step in to prevent them before its too late.
How is it any better if instead of _one_ vendor, _two_ vendors push an immature version of a standards extension that mainly caters to their needs and give it the official stamp of approval under the MCP umbrella?
Having a chatbot that drives websites inside of it is such an attempted monopolist play. Having a system agent that can interact with apps via API without being connected to the app is the pattern that's both elegant and preserves freedom.
I don't see how this is a step down from existing web applications. Should companies building web applications not be opinionated about their user interfaces? When I look at Notion, I should just get any view of the data inside it, regardless of whether it's the same view as my coworker gets? How is this preferable?
> Having a system agent that can interact with apps via API without being connected to the app is the pattern that's both elegant and preserves freedom
MCP was created so llm companies can have a plugin system. So instead of them being the API provider, they can become the platform that we build apps/plugins for, and they become the user interface to end consumers.
MCP defines the API so vendors of LLM tools like cursor, claude code, codex etc don't all make their own bespoke, custom ways to call tools.
The main issue is the disagreement on how to declare the MCP tool exists. Cursor, vscode, claude all use basically the same mcp.json file, but then codex uses `config.toml`. There's very little uniformity in project-specific MCP tools as well, they tend to be defined globally.
>but isn't this solved by publishing good API docs, and then pointing the LLM to those docs as a training resource?
Yes.
It's not a dumb question. The situation is so dumb you feel like an idiot for asking the obvious question. But it's the right question to ask.
Also you don't need to "train" the LLM on those resources. All major models have function / tool calling built in. Either create your own readme.txt with extra context or, if it's possible, update the API's with more "descriptive" metadata (aka something like swagger) to help the LLM understand how to use the API.
You keep saying that major models have "tool calling built in". And that by giving them context about available APIs, the LLM can "use the API".
But you don't explain, in any of your comments, precisely how an LLM in practice is able to itself invoke an API function. Could you explain how?
A model is typically distributed as a set of parameters, interpreted by an inference framework (such as llama.cpp), and not as a standalone application that understands how to invoke external functions.
So I am very keen to understand how these "major models" would invoke a function in the absence of a chassis container application (like Claude Code, that tells the model, via a prompt prefix, what tokens the model should emit to trigger a function, and which on detection of those tokens invokes the function on the model's behalf - which is not at all the same thing as the model invoking the function itself).
Just a high level explanation of how you are saying it works would be most illuminating.
The LLM output differentiates between text output intended for the user to see, vs tool usage.
You might be thinking "but I've never seen any sort of metadata in textual output from LLMs, so how does the client/agent know?"
To which I will ask: when you loaded this page in your browser, did you see any HTML tags, CSS, etc? No. But that's only because your browser read the HTML rendered the page, hiding the markup from you.
Similarly, what the LLM generates looks quite different compared to what you'll see in typical, interactive usage.
The schema is enforced much like end-user visible structured outputs work -- if you're not familiar, many services will let you constrain the output to validate against a given schema. See for example:
It is. Anthropic builds stuff like MCP and skills to try and lock people into their ecosystem. I'm sure they were surprised when MCP totally took off (I know I was).
Rechat | API Engineer, Front End Engineer | Remote
Rechat is the leading AI-powered operating platform for real estate brokers and agents. Think Shopify for real estate—a mobile-first super app with marketing automation, workflow management, and everything agents and brokers need to run their business in one place. We’re a small, sharp team building a big, sophisticated product without the usual corporate nonsense.
Why You’ll Like It Here:
Remote-first with flexible hours
Minimal meetings—we focus on real work
No tech hype-chasing—we use the right tools, not the latest trends
Flat team structure—no layers of bureaucracy
Who We’re Looking For:
Senior Frontend Engineer (TypeScript, React) – Own the UI and craft intuitive experiences
Senior Backend Engineer (Node.js, PostgreSQL) – Build and refine our robust, high-performance backend
If you thrive in small teams solving big technical challenges, reach out to emil+hn@rechat.com.
They have several actually, from computer vision in glasses (RayBan or Quest) to Speech To Text to get commands on such glasses, to "improved" translation via LLMs, to just chat bots in most of their chat solutions. They do integrate into products, it's not just research.
Is it good? No idea as I don't use them but I believe their angle is literally what Zuckerberg said publicly, roughly "Can't miss AI if it's real! Have to be first." which isn't exactly a very deep strategy but they have deep pockets.
More importantly, do these AI integrations they have make money or even have the potential to in the future?
It might surprise you to find out that Ray-Ban Meta glasses don't offer any sort of subscription service, not even as an option. Every Meta AI user is just costing Meta money, Meta isn't even giving them the option to buy the product from them.
I have no idea why. The kind of people who would buy Meta glasses would probably happily blow $10-20 on a subscription they forget about. You can get a subscription service for a robot litter box but you can't get one for AI glasses? Does Meta hate money?
Meta uses AI to search through Facebook and Instagram which...just makes searches cost them more money, I guess?
Sounds like they have pockets so deep that they are going into debt, which is an interesting sort of pocket depth.
IMO Zuckerberg's amateur founder status is more blatant as time goes on. He had his one moonshot and thinks he can do it again just as easily. Nobody told him that a large chunk of his success is owed to fortuitous timing.
I think there's been something of a cancerous ideology that you must be a first mover. It's a bit odd considering that Facebook itself was not a first mover in pretty much everything that it does that is successful and highly profitable.
What is NOT their angle; ads, UGC, entertainment experience (algo etc), Metaverse and gaming, communication (WhatsApp, insta etc) and I’m sure they’ll take advantage anything that’s close to their core areas of interest or anything else big. AI is definitely the tide that lifts all boats but if you’re one of the top 5 tech companies in the world then the prize is incredibly large and not yet known.
The investors don't seem to agree, it seems to be sinking rn... Ads? They already sell ads, is their "AI" algorithm better than the current one developed over years by some of the smartest phds on the planet? I very much doubt that.
Meta's goal is to stop OpenAI, Google and/or Anthropic from shutting them out of whatever AI ends up delivering. This is why they went with open weights for LLama - it prevents the other players becoming gatekeepers.
This is part of a pattern of tech leaders investing in order to avoid getting shut-out of whatever the next paradigm of computing is supposed to be.
- Google building Google+ and stuffing social into everything to avoid getting shut out of social networking. (The fear Larry/Sergey felt about this is why Vic Gundotra could bully and survive scandals until it became clear that Facebook wasn't an existential threat/Google+ was not going to really compete)
- Meta attempting to build an AI assistant because they were afraid Alexa/Google Assistant/Siri would be how everyone accessed computers in the future (due to technical failure, this product only ever launched as control mechanism for Oculus, but the ambition was larger)
Of course this always come alongside other factors that lead others to follow when a new concept is proven; however a tell-tale sign that leadership is worried about market dominance rather than a mere new line of business is that they spend or throw weight around above and beyond what the new line of business alone would justify.
I see a lot of negative comments here but to me, it was obvious this is where OAI should land.
They want to be the platform in which you tell what you want, and OAI does it for you. It's gonna connect to your inbox, calendar, payment methods, and you'll just ask it to do something and it will, using those apps.
If you ask it to build a headless frontend web app, it immediately starts generating code with Next.js. I’ve always wondered how it was trained to default to that choice, given the smorgasbord of web frameworks out there. Next.js is solid, but it’s also platform-ware, tightly coupled to commercial interests. I wish there were more bias toward genuinely open-source technologies.
There's probably different ways the LLM converged to it.
One could be for example: from people asking online which tools they should use to build something and being constantly recommended to do it with Next.js
Another could be: how many of the code that was used to train the LLM is done in Next.js
Generally, the answer is probably something along the lines of "next.js is kind of the most popular choice at the time of training".
To me it feels like the default choice in the industry, perhaps it's not and I'm wrong but if I could have that feeling I can see how the AI can as well.
I wonder what the ad labeling (according to EU law) would look like in that case.
In my (non-lawyer) understanding, each message potentially containing sponsored content (which would be every message, if the bias is encoded in the LLM itself,) would need to be marked as an ad individually.
Because the AI labs are just hovering up all internet text that they can, I've been seeing more and more marketing pilots that deliberately seed marketing material in thousands of fake, AI-generated blogs and tutorials. The intention here is to get new LLMs to train on these huge numbers of associations between specific use cases and the company's product. All in a way that gets their marketing information into the final weights.
You may have started seeing this when LLMs seem to promote things based entirely on marketing claims and not on real-world functionality.
They obviously want both. In fact they are already building an ad team.
They have money they have to burn, so it makes sense to throw all the scalable business models in the history, eg app store, algo feed, etc, to the wall and see what stick.
Consumer LLM apps have moat. As it is, ChatGPT (the app) spends most of its compute on Personal Non work messages (approx 1.9B per day vs 716 for Work)[0]. First, from ongoing conversations that users would return to, then to the pushing of specific and past chat memories, these conversations have become increasingly personalized. Suddenly, there is a lot of personal data that you rely on it having, that make the product better. You cannot just plop over to Gemini and replicate this.
Because it changes all the time. A few weeks ago, it was Gemini 2.5 Pro, then Claude Opus 4.1, GPT-5 Thinking, now maybe Claude Sonnet 4.5, etc[1]. Having a good model isn't enough when they're basically interchangeable now. You need something else.
[1] This is an example. Which model was the best when is not important.
Because it depends on how much better “best” is. If it’s only incrementally better than open source models that have other advantages, why would you bother?
OpenAI’s moat will only come from the products they built on top. Theoretically their products will be better because they’ll be more vertically integrated with the underlying models. It’s not unlike Apple’s playbook with regard to hardwares and software integration.
The sad part is that these "Israel Supporters" don't realize they've destroyed Israel in the process.
For 2 reasons:
1. Most young Americans, in both parties, are sickened by this. Next generations of Americans are not gonna support Israel anymore and the US support for Israel is the only security Israel has. In 20 years Israel will be left alone between all the countries that they once had a chance to make peace with, but decided to wage war against.
2. Israel is gonna deteriorate and become another authoritarian regime like the rest of of the Middle East. The ultra orthodox population grows and it's simply not gonna be a democratic state anymore.
So yes, go ahead, commit a genocide/ethnic cleansing (whatever you name it) in the name of a country that you destroyed in the process.
People on our circles are obsessed with model performance. OpenAI's lead is not there and hasn't been there for some time.
They do, however, have a major lead in terms of consumer adoption. To normal people who use llm's, ChatGPT is _the_ model.
This gives them a lot of opportunities. I don't know what's taking them so long to launch their own _real_ app store, but that's the game they are ahead of everyone else because of the consumer adoption.
Totally, I feel like though you do have to pay some attention for example in the context I'm working on, for the last while, Gemini was our gold standard for code generation whereas today, Claude subjectively produces the better results. Sure you can stick to what worked abut then you're missing the opportunity to be more productive or less busy, whichever one you choose.
I remember the days when I was looking for the perfect note-taking system/setup - I never achieved anything with it, I was too busy figuring out the best way to take notes.
