Google – Wrong Product.

Google needs to launch an agent.

• Google has launched its new model into an environment of uncertainty, as while it makes the grade on the benchmarks, there is still no answer to Muse or Dots leaving Google’s ecosystem unprotected should everyone suddenly decide that they want an agent.
• The details of Gemini 4 Argon have been made available, but the model itself is being phased in to ensure that it doesn’t misbehave and that any flaws or vulnerabilities that are uncovered can be patched before letting everyone get their hands on it.
• Google has had difficulty so far this year, as Gemini 3.5 (that was promised at Google i/o) never materialised and was shelved after it could not match the performance of its rivals.
• Google hopes that Gemini 4 will put this ignominy to bed.  
• The now familiar format of data and charts looks good, and Google can match GPT-6 Astra and Fable 5 on the standardised tests, but there is no data on token consumption or model size.
• It is important to remember that while all of these announcements look like scientific papers, they are in fact marketing press releases, as their materials are not peer-reviewed, nor are they published in scientific journals.
• Hence, one has to take everything they say with a grain of salt, as the benchmarks are likely to have been cherry-picked, and the fact that there is no answer to OpenAI’s efficiency claims suggests Google probably lags here also.
• Furthermore, there are rumblings from within Google that Argon does not fare as well as the benchmarks indicate when used in real-world coding tasks.
• This is the last thing that Google needs right now, as it is currently experiencing what I would refer to as a Meta moment, where a model goes wrong, exposing a flaw in the organisation which then needs to be rebuilt in the hope that better products will follow.
• With the launch of Muse, Meta is comfortably on the other side of this problem, and if Muse continues to be very popular, it won’t need to be cutting-edge, just useful, fun, as well as easy to use and set up.
• Google has clearly been very focused on Gemini 4, which, given the success of Muse and the launch of Dots, was arguably the wrong product.
• If agents become popular and Google has no product, then its flank is exposed, and OpenAI and Meta will be more than happy to take large chunks out of it.
• Hence, I am not convinced that Google’s AI woes are fully over, but I don’t think that this is the beginning of the decline of Google.
• Back in 2024, Google made a horrible mess of its AI products and the messaging around them, but it came back strongly.
• There is no reason to think that it won’t again, but if this period triggers a loss of faith in Google and the share price collapses, this could be a great opportunity.