Google Gemini – Rear Guard Action

Project Astra is all about self-preservation.

• Google’s latest models show incremental improvements in a sign that the scaling effect is slowing down but Google is making an early move towards AI agents which I see as a strategy to defend its dominance of the Android platform.
• Google has launched Gemini 2.0 which, by the observable data, offers an incremental improvement in performance over Gemini 1.5 in yet another sign that the scaling effect is slowing down.
• This is why Google has shifted the focus to “agentic AI” which in plain English means a voice bot that can also complete tasks on the user’s behalf.
• This is nothing new as this was promised more than 6 years ago, but due to the AI’s inability to understand and converse in natural language at the time meant that the proposition fell flat on its face.
• This time around things are very different as the invention of the transformer architecture by Google in 2017 and LLMs by OpenAI in 2020 have finally allowed this hurdle to be overcome.
• LLMs may hallucinate and get things wrong but they are very good at understanding and conversing in natural language meaning that all the promises made in 2017 and 2018 may now finally be kept.
• Google is not alone in working on AI agents as Anthropic, Open AI and others have also demonstrated such use cases, but Google has an advantage in that its ecosystem is already being used by billions of users.
• Google has introduced three new prototype agents that are in the very early stages of general availability.
  • First, Project Astra: which was launched at Google I/O some months ago but has now been refreshed by using Gemini 2.0 as its foundation.
  • This is an agent for Android smartphones that can answer questions as well as interpret what the agent can “see” through the smartphone’s camera.
  • Gemini 2.0 brings more languages, more fluent conversation with lower latency as well as the ability to use Google Search, Lens and Maps on the user’s behalf.
  • If AI agents are going to take off on smartphones, then this is a must win battle for Google.
  • If users prefer to use GPT or Claude on their smartphones, then Google is in real trouble because it will have lost its direct access to the user.
  • Instead, it will have become a support act in the digital ecosystem and its ability to monetise traffic on smartphones will take a large hit.
  • I am far from sold on the idea that we will all be talking to our smartphones all the time as the touch-system is already perfectly good enough for many users, but I am not ruling out the possibility.
  • If it does happen, then Google, Tencent and Apple (weak at AI) will have a big challenge to resolve.
  • Second, Project Mariner: which is PC or Chromebook agent which can execute tasks at the user’s request in the Chrome browser.
  • This agent is most like a personal assistant that will go and complete tasks on the user’s behalf but, at the moment, Mariner is limited to the Chrome tab that the user has open.
  • This limitation highlights that putting agents on devices is much more complicated than it looks.
  • This is because a lot of messy software plumbing is needed to connect the agents to all of the apps and services so that they can perform all of the necessary tasks.
  • This is yet to be made widely available and is still in early testing.
  • Third, specific use case agents: which are targeted at developers to help them write code or help players when they are playing games.
  • The gaming agents under development can understand what is going on in the game and offer the player tips and tricks as well as answer questions on game strategy.
• While the possibilities offered by AI agents are an opportunity for some, they represent a huge risk for the incumbents such as Google, Apple and Tencent.
• This is because these three companies dominate the digital ecosystem on smartphones today and their relationship with users is central to their ability to generate revenue.
• If a 3^rd party is successful in inserting an agent in between their ecosystems and the user, then there is a strong likelihood that their business models will be disrupted.
• This is why from Google’s perspective, Gemini has to be the agent of choice on Android that is accessing and using apps on the user’s device.
• Uptake of agents is by no means guaranteed but it represents a dislocation point that could upend the entire status-quo on smartphones.
• Hence, Google’s Project Astra and Project Mariner are defensive products that must win in their categories if Google is to continue dominating the Android ecosystem.
• Whatever the outcome, this will take a long time to manifest and so I don’t see any disruption to Google’s financial performance in the short or medium term, but it is a long-term risk.
• Hence, I remain fairly indifferent to Google where its valuation in the market is a reasonable reflection of its outlook and the risks that it faces.
• There is more excitement to be had elsewhere.