Another busy week (and I’m not even going to talk about Stargate)! Here are four items I found relevant and interesting:
- There is no moat, part 1: OpenAI’s o1 model came out in December, and it was “smarter” than anything we’ve seen. It mostly used what is now called TTC: Test Time Compute (which means using a lot of inference before generating a final answer, like Chain of Thought, etc.). Now Google has their version, Gemini Flash Thinking at a much lower price point, and even more surprising, a Chinese firm called DeepSeek has released an open source model with similar capability!
- There is no moat, part 2: Anthropic was the first to market with an agent, allowing Claude to control a computer in October. This week OpenAI announced Operator, a ChatGPT agent that performs tasks in the browser.
- A big week for Perplexity. They released the Perplexity Assistant on Android…and claim it’s a transition from an “answer engine” to a “natively integrated assistant that can call other apps and perform basic tasks for you.” Kudos to them for calling it an Assistant instead of jumping into the over-hyped Agent label! More interesting for enterprise RAG, they released a generative search API (named Sonar) so companies can build Perplexity into apps. Of course, the source is only the internet (which isn’t where the value is for enterprises), we can expect them to add custom sources in the future.
- Finally, LlamaIndex introduced AgentWorkflow, a system to build and orchestrate AI Agent Systems.
My take on why does it matter, particularly for generative AI in the workplace
There is no moat! These companies have invested billions of dollars into training advanced LLMs. Normally that kind of investment is a “moat” – a protection from competition. It appears however that in this space, the moat really just is the investment itself. The technology and the data and the techniques are (reasonably) available to others as well, so as soon as one company achieves a new milestone, they will be matched or beaten by another. It will be very interesting to watch this play out and see how these companies make money. Will they have a path to profitability?
