AI Arms Race Volume: 9

Will AI essentially replace all human customer support roles?

Despite the fundamental AI memory limitation that we described here, Brett Taylor thinks most customer support roles will be done by AI.

Sierra CEO Brett Taylor is betting that his company can beat the fundamental AI memory problem we described in our last update. Some of his goals sound achievable using known techniques and agents we have discussed (it’s essentially selective filling of the context window), but Sierra does not appear to have any new technology, so we remain skeptical about their bigger claims, eg: AI coaches that are better than good coaches.

https://www.cnbc.com/video/2025/11/06/sierra-ceo-brett-taylor-on-implementing-agentic-ai-were-making-agents-that-have-memory.html

Where is Google Gemini 3.0? It’s here!

There were rumours that Google had delayed Gemini 3.0 because it is not any better than recent releases from Anthropic, OpenAI or even the Chinese Open Source Kimi K2 from Moonshot AI. But no, it’s now here and it’s possibly the most intelligent AI model yet.

However, for many use cases it’s expensive and slow. This is not really surprising, we have been saying for a while that major gains in AI are unlikely anytime soon. Perhaps the most interesting part is the new impressive IDE (‘code editor’) called Antigravity that has been released with Gemini 3.0. It’s currently free, which is sure to worry Anthropic (Cursor) and Microsoft (CoPilot/VScode).

Has Big Short character Michael Burry bet against AI?

We have been saying for a while that AI LLMs are likely to become a commodity, great for companies like ours but not so great for many of the big AI investments, and potentially very bad for AI hyperscalers. There are a number of people on social media claiming famous short seller Michael Burry has bet big against NVDIA and Palantair. Has he really? The FT ran a story on this and didn’t make that conclusion:

https://www.ft.com/content/7fe1362b-d696-4334-86ef-607b80f1739f

Is Google doing far better than it has let on? What does this mean for OpenAI, NVIDIA and other big AI players?

Despite the cost issues of Gemini 3.0 we mention above, Google seem to be doing very well and now have perhaps the most intelligent AI model in the world. AI model inference costs are falling fast and the Google are still leaders in driving down costs. Some have called for the break-up of Google recently, and this has probably led to it not publicising how well it is now doing in the AI market. Google has vertical integration that OpenAI, NVIDIA and others can only dream of. For example it is the only major player with its own AI chips and does not need to rely on NVIDIA. 

Below is a chart from the Artificial Analysis 2024 review. Little has changed since that report was created and Google remains the only company with all 4 parts of the value chain covered.

It seems that legendary investor Warren Buffet has noticed this as well and quietly made a $4.3B bet on Google (“Alphabet”) making it his 10th largest holding.