OpenAI – Tale of Two Cities

Sam builds while the EU fiddles.

• OpenAI has re-launched its offensive for the consumer and enterprise AI ecosystems, and, as its models are now far more mature, it has a much better chance of winning developers to build on its ecosystem.
• However, all these new tools and services for consumers will not be available in the EU, as a result of the AI Act, which, despite promises made 4 months ago (see below) to revise it, has seen no progress that I am aware of.  
• This means that the EU continues to send a message to its innovators and start-ups that says, “if you want to develop AI, don’t do it here”.
• OpenAI’s developer day has attracted very little attention as it rather foolishly announced the gigantic deal with AMD at the same time, meaning that most people didn’t even notice that there was another event running at the same time.
• This is unfortunate, as the potential impact of what OpenAI had to say is profound and potentially far more relevant than buying $60bn of chips from AMD.
• This is because OpenAI is laying out how it intends to become a giant consumer and enterprise ecosystem and generate hundreds of billions of dollars in revenue.
• OpenAI has already made some progress as it is now claiming that it has 800m weekly active users and 2m developers, making it by far the leader in the space.
• New launches aim to consolidate and grow this advantage and include:
  • First, Apps SDK: which is the implementation of OpenAI’s strategy to ensure that users spend more time with OpenAI and ChatGPT than with the current digital ecosystems, making them less relevant.  
  • Apps SDK allows Apps to function inside ChatGPT and to be controlled by the agent, which, if the demos are a true reflection of reality (big if), will make them much more useful.
  • The Apps SDK is built on the rapidly standardising Model Context Protocol (MCP) (first launched by Anthropic), which is a single standard by which any agent can be connected to an App or service, allowing the agent to control it.
  • This comes in handy as it becomes much easier to book travel on Expedia involving complex conditions as opposed to having to spend a lot of time doing it oneself (as I often do).
  • This use case is to make the experience more like having a human knowing what the user wants and booking it on his or her behalf.
  • Launch partners are Booking.com, Canva, Zillow, Cousrera, Figma and Spotify and the demos for each of these point at significantly enhancing the utility of these apps.
  • If I can get ChatGPT to accurately book my often-complex travel trips without ending up in Timbuktu, I will use it and use it a lot.
  • At the moment, OpenAI’s coverage in terms of the services that it can offer compared to what users regularly do with their smartphones is very limited, and I can’t see the likes of Meta, Google or Apple using Apps SDK limiting the appeal of Apps in ChatGPT.
  • However, if many apps from other players start working well in ChatGPT, and users love it, it may just start to put them under the kind of pressure that could upend the whole smartphone digital ecosystem.
  • We are very far from this, but clearly, the intent of OpenAI is to become the place where users live their digital lives and steal value from the incumbents.
  • Second, AgentKit: which is OpenAI’s gambit to ensure that it is the prevalent supplier of AI tools to enterprises.
  • The process of creating and customising an agent for an enterprise is currently quite difficult and fragmented, and AgentKit is designed to make this process much easier.
  • The MIT study clearly demonstrated that just giving employees a ChatGPT prompt and expecting a large productivity bump is unrealistic most of the time, meaning that customisation and adjustments are necessary to make AI useful.
  • AgentKit is a set of tools that makes it much easier to create, deploy and fine-tune AI agents specific to enterprise use cases based on OpenAI’s foundation models.
  • By making it easy to create agents, OpenAI is hoping that developers choose to use GPT as their foundation rather than Gemini or anything else, which is how it aims to take control of the Enterprise AI ecosystem.
  • Third, Codex SDK: which is a tool designed for integrating the code-building capability of OpenAI (GPT-5 Codex) into software development workflows.
  • This is an attempt to push back against the dominance of Anthropic, which is widely recognised to be the best agent for writing code.
  • I think that Anthropic remains the best at writing code, but if OpenAI can be almost as good but much easier to implement and use, then it is likely to win support over Anthropic.
• The key to any digital ecosystem is scale and developer traction, and with these tools, OpenAI is hoping to translate its household brand into a real business.
• This is a rinse and repeat of what happened in the smartphone market, where the launch of tools to help developers write apps for iOS and Android and a store to distribute them was central to allowing Google, Tencent and Apple to become the dominant digital ecosystems today.
• With these launches, OpenAI is making an early move to ensure that GPT foundation models are the go-to place to create AI apps and services, which is what will allow it to become one of the gatekeepers to the market for AI services.
• This is how it could begin to justify the $500bn (and rising) valuation that has already been attributed to the company.
• The digital ecosystem has a newcomer, and the incumbents will need to step up to prevent the newcomer from undermining their entrenched positions.
• The fact that the ability to use apps inside ChatGPT will not be available in Europe is yet another sign that the European environment remains not fit for purpose when it comes to AI.
• The main culprit here is the AI Act, which is regulating an industry that does not yet exist and, in many ways, will prevent it from existing.
• At the Vivatech conference earlier this year, there was a big push by European CEOs to advance the cause of AI in Europe and sovereign AI in particular.
• Politicians also in attendance stated that the AI Act need to be toned down to make it easier for small companies and start-ups to comply because, at the moment, it makes impossible demands, demonstrating that those who wrote the act do not really understand how AI works (see here)
• I left the conference cautiously optimistic, but in the subsequent 4 months, virtually nothing has happened that I am aware of.
• The longer it takes for the EU to fix its regulatory environment, the further it will fall behind and the more that its start-ups and entrepreneurs will go elsewhere.
• The signs of this are everywhere, as Europe has one LLM company while the USA has many, and Europe remains nowhere when looking at the league tables of AI companies.
• There are plenty of places to go, and most of the Middle East was to be found camped out at the gates of Vivatech, enticing start-ups to move to the UAE or Saudi Arabia.
• In the past, these entreaties have fallen on deaf ears, but no longer, as the population of the UAE is growing and access to talent is now much less of a problem than it was 5 years ago.
• Unless something changes soon, Europe will not catch up in AI as it will be too far behind and its talent will have fled, meaning that the economic benefits will accrue elsewhere.
• The vibe at the Vivatech trade show was supposed to light a fire under the tail of AI in Europe, but 4 months later, my cautious optimism already looks like it was overly optimistic.
• OpenAI is moving to build the next digital empire while Europe fiddles as Rome, Paris, and Frankfurt are burning.