AI Ecosystems
Anthropic proposed the MCP standard in 2024, since then the development of AI interoperability & LLM saas integration took off - with different foundation model companies adopting this method and, rewiring their products to build the best integrated & multi-modal tools for their customers.
LLM applications got the attention of the main public through chatbots just a few years back. Now the focul point is no longer on building the perfect image generator or UI interface on the web, but rather API systems that can integrate, evolve & act semi-autonomously within the life context of the user.
Its been inferred that the next logical step forward for getting humanity closer to AGI is to make systems more contextually aware and modal (sort of wired to all environments).. that concept translates across the entire spectrum from software to physical AI.
However defining AGI use cases in the world is still a far stretch..for now (we want to achieve more than just autonomous pizza ordering).
What companies are currently striving for is to build entire ecosystems around their models, for the web, influencing the traditional way people have been using the internet for the past couple of decades.
A lot of news sources are portraying this as aggressive web tool or browser competition... but at this point of time the hype is misinformation at best. Those VC funded initiatives are still in experimental phase, designed in-part for accessing more user behavior metadata to fine-tune proprietary models.
Companies like Open AI currently build narrow scoped platforms (like Atlas for Mac OS users for example) in order to test their APIs & analyze tons of networking issues before deploying at massive scale.
This proves that the rumored "AI Bubble" reports are not accurate in evalutating the intrinsic value & fundamentals of these systems ...we are still in early stages of this computational progress, with the ongoing massive expansion of infrastructure involved.
After the GPT store launch, Open AI analyzed the use case of Plugins & GPTs & realized the core interoperability problem. They redesigned their app integration blueprint and also introduced multi-agent workflow building (a kind of light version of N8N)... acknowledging that tracking navigation habits & long-term user goals is crucial for scaling their ecosystem, and be in sync with AI UX.
Despite these developments, the line is often blurred to see which incentive is prioritized between innovation & control, and the latter warrants a big leap of faith from the user.
Although building workflows and agentic systems is getting less clumbersome, companies are motivated to gather all metrics of their products through some telemetry backdoor... giving them constant authority of the interface, You can sort of relate the Apps SDK presented on October 6th as an Apple store, where the consumer is indirectly locked in.
Silicon Valley is advertising this by saying AI will be useful & enable productivity... i completely agree, yet, to what extent are consumers making themselves useful to someone else's AI through their browser activities?
That is the catch-22.
We are not only talking about new features for coding assistants (although Codex is still the main engine with respect to Open AI's stack), this is about setting the stage for the next digital economy and user interaction for the entire www.
While Perplexity tried to buy Chrome for roughly 35 billion dollars, Open AI understood the networking challenges of re-engineering web browsing and designed Atlas based off a Chromium fork.
Is there an alignment? Its hard to say.
Anthropic yells that out loud like a bitter english professor... while Google keeps building with the traditional ad revenue incentive. Open AI is convincingly trying to shape your primate brain into a data pipeline, meanwhile, Zuckerberg is trying to convince every engineer he has the best snack bars at work.
Go figure.
But what does alignment even mean? Who gets to define constructive human potential and set policies for that?
Will we stop building apps, and instead agents will auto-build tools for us based off our cognition data?
Who knows.
Maybe,.. right now, experts are discussing a "context clairvoyance system" over some snacks while sipping Paso Robles wine.