ChatGPT Plugins — Automation for Everyone
OpenAI's Evolution from APIs to Consumer Platform
ChatGPT adds Plugins
OpenAI recently announced ChatGPT plugins which really got me thinking about what kind of company OpenAI is going to become and what it means for everyone else. Prior to the launch of plugins, OpenAI primarily operated as a standalone company that competed with other providers of AI models like Cohere, Anthropic, etc. As a result, most companies did not have to interact with OpenAI if they didn’t want to. OpenAI was not a strong force on the broader tech ecosystem in a way that companies like Apple, Google, and Microsoft are. But I believe that is starting to change with Plugins.
Plugins are tools designed specifically for language models with safety as a core principle, and help ChatGPT access up-to-date information, run computations, or use third-party services. — OpenAI
This is one of those things where examples are going to be the easiest way to understand. Here are a few that were part of the plugins launch:
Instacart — Their plugin lets you go from recipe to shopping cart in moments (link to video)
Wolfram Alpha — This one extends ChatGPT into areas where it typically has been less impressive (factual queries, visualization, etc).
Web Browsing — OpenAI themselves added a plugin that enables models to read the internet in real-time instead of relying solely on its quickly outdated training data. This solves the “My training data only goes up to 2021” problem that prevents it from taking on many useful queries.
Essentially, for all the things that ChatGPT can’t do — search for flights or hotels, book restaurant reservations, or order your groceries — it can address via first and third party plugins. Now, OpenAI’s powerful, generalized models can execute actions across other software applications all orchestrated through the easy-to-use ChatGPT interface.
Evolution of OpenAI
But this is more than just a cool new demo. This launch marks a meaningful shift in the OpenAI strategy and how the company interacts with the broader software ecosystem. It’s worth recapping how OpenAI has evolved thus far and how quickly things appear to be changing:
Act 0 — 2015-2018: The Research Lab
OpenAI originally set out towards AGI like a research institution, albeit one with $1B in backing from the extremely deep pockets of Silicon Valley. It was unclear exactly what this company was going to look like, but the goal was “to advance digital intelligence in the way that is most likely to benefit humanity as a whole”.
Act 1 — 2019-2020: API as a Service
In the next phase, OpenAI transitions to for-profit, takes a $1B investment from Microsoft, and finds its footing as a traditional software infrastructure business. The core product was a set of generalized models that customers could access via an API. It seemed that OpenAI would make its money being an outsourced R&D function for other businesses wanting to use AI. They were on the path to be the AWS of AI.
Act 2 — Nov 2022: Consumer Experiment
Now things get interesting. OpenAI launches ChatGPT as a free web-based product. When ChatGPT is released, it felt more like a large scale experiment than a concerted shift in strategy. That experiment exceeded everyone’s expectations when it hit 100m users in just two months with basically no marketing. At this point, it remained unclear what monetization would look like but at the very least, it captured consumer attention quickly.
Act 3 — Mar 2023: Platform??
That brings us to the present. OpenAI’s ChatGPT launches Plugins, a new way of interacting with the broader universe of software applications. ChatGPT is no longer just a standalone product, but a platform that interacts with other software applications.
Under the veil of a grand AGI vision, research origins, and initial nonprofit status, OpenAI was able to transition into what feels like the next Big Tech platform. So it’s only natural to start thinking about comparisons to companies like Google and Apple.
Plugins: Automation for Everyone
Ultimately, the reason plugins are so important is that they push ChatGPT towards being a platform that sits between end users and other software applications. Previously, OpenAI interacted with businesses via their API offering and separately end users via ChatGPT. The introduction of plugins puts them squarely between the two which is a very powerful place to build a business. Specifically, ChatGPT is made more useful to end users as more software plugins are made available which in turn makes it more compelling for businesses to integrate with ChatGPT.
Ben Thompson has written at length about Tech’s Two Philosophies and how it drives Big Tech incentives and decisions. He describes the two types of Big Tech business models: Aggregators (Google, Facebook) and Platforms (Apple, Microsoft).
“In Google’s view, computers help you get things done — and save you time — by doing things for you. Technology’s second philosophy, and it is orthogonal to the other: the expectation is not that the computer does your work for you, but rather that the computer enables you to do your work better and more efficiently.
While platforms need 3rd parties to make them useful and build their moat through the creation of ecosystems, aggregators attract end users by virtue of their inherent usefulness and, over time, leave suppliers no choice but to follow the aggregators’ dictates if they wish to reach end users.”
Interestingly ChatGPT Plugins share characteristics of both philosophies depending on what you’re trying to do. For certain tasks, ChatGPT is fully doing the work and in other cases, ChatGPT is a way to enable you to use some 3rd party application more effectively. The most relevant examples I can think of are Google and Apple.
Comparisons to Google Search
The comparisons to Google began almost immediately after ChatGPT initially launched. Headlines highlighted Google’s internal code red as people turned to ChatGPT instead of Google for a portion of queries.
While Google and ChatGPT have some overlap in the types of queries they address, they are very different from a business perspective. Google’s mission is to “organize the world's information and make it universally accessible and useful”. As such, it primarily aggregates and indexes the world’s webpages. In doing so, Google Search is able to monetize by pushing this search traffic (demand) where they want (advertisers). There isn’t really a similar analogy for ChatGPT — it’s not aggregating existing content but rather generating new content. As a result, the ads model is not likely the end destination for OpenAI.
However, I think the comparison to Google is far more relevant now with plugins specifically. Just as Google commoditized the Internet’s web pages, ChatGPT may be able to do that for a set of Internet businesses by providing an easily programmable automation layer.
For example, a well-defined task like booking a dinner reservation is likely one that I could entirely orchestrate through ChatGPT without even logging into OpenTable, Resy, etc. It could be that just as websites had to reorient their websites to rank highly on Google, there could be a world where companies design applications such that they can be easily programmable through ChatGPT.
Comparisons to the iPhone and App Store
There are also similarities between ChatGPT + Plugins and iPhone + App Store. Both the iPhone and ChatGPT provide some general functionality that draws in consumers, but are extended in value by 3rd party applications.
To elaborate on this analogy, Apple's iPhone's success was not due to it being the first mobile phone, but rather because it excelled in essential capabilities such as camera quality, security features, and an integrated software/hardware experience. Similarly, OpenAI's ChatGPT has captured consumer interest not because it was the first chatbot but because it is has significantly better performance than previous version while being widely accessible.
If we look to Apple as a trailblazer here, we saw that they introduced the iPhone App Store shortly after the iPhone, allowing developers to build 3rd party applications that could be easily downloaded by mobile users. This allowed Apple to extend the value of their phone because a host of new services like Facebook, Spotify, Fruit Ninja were available. Similarly, ChatGPT launched plugins as a way to extend the value of its generalized model into very specific applications that others have built.
Apple gives its 3rd party applications the convenience of reaching consumers via mobile. Similarly, ChatGPT Plugins give 3rd party applications a natural language automation layer via the ChatGPT interface.
This ultimately turns OpenAI’s original commercialization strategy on its head. OpenAI’s focus on APIs were essentially a way to embed OpenAI’s AI into other software application. But now, ChatGPT Plugins are a way to extend your software applications into OpenAI’s user interface.
Implications
I’ll caveat that I don’t think OpenAI has quite sealed their fate on what type of company it will be, but it is clear that it will matter to the ecosystem. And while it’s easy to make comparisons to existing products, ChatGPT is a unique product that likely warrants a unique business model.
There is a lot of stuff happening in AI that you probably don’t need to keep up with, but I think this is worth following. What OpenAI does with plugins could become a significant piece of the tech ecosystem and we haven’t seen a new entrant here in a while.
If you’re building a software application, it’s of course worth understanding how the past few years of advancements in AI can augment or disrupt your business. High growth software companies will be some of the biggest winners of this AI inflection point as they have the DNA to build product quickly with the existing distribution and data necessary to build valuable AI products. The losers may be companies that are simply glue code sitting between applications which could soon be OpenAI’s domain.
At the same time, I think software companies will need to understand how to coexist and evolve to not be commoditized or replaced by something like ChatGPT. Apple ultimately built many products that competed directly with 3rd party applications like Mail, Safari, Apple Music, and so on. And Google ended up making certain businesses overly dependent on them for customer acquisition.
The key question will be how much value do you provide over a generalized language model? Those that don’t provide enough value to build a real end-user relationship are likely to be commoditized by ChatGPT or plugins they build.
While ChatGPT is largely a consumer tool, I’m also seeing similar chat-based design patterns leveraged in the enterprise software world by a new crop of startups. I suspect similar dynamics may play out in domains like business intelligence, sales, software development, etc. I’ll be closely watching how OpenAI’s business evolves and what it could mean for the rest of the AI and software ecosystem.
Nandu | @nanduanilal