Note: No software is defensible in terms of intellectual property. The models themselves are a prime example -> it took OpenAI 8 years and $1B to build a model, which Mistral AI built in 3 months and $20M.
For companies like Team-GPT the defensibility comes from several places:
We have a working product, battle tested by 35K people, being actively used by thousands.
We already have some big names under our belt, which are a source of credibility and case study material.
We are not theoretically thinking about an AI product. We have product inside thousands of businesses already. We also know what comes next.
All AI models are worthless without data.
Foundational models, such as OpenAI, Anthropic, Gemini, etc. are equally good (bad) because they’ve been trained on publicly available data. Mistral got there fast, too. LLaMa 3 allows companies like ours to be there, too.
The point of this AI wave is to get your hands on the enterprise data – something OpenAI doesn’t have access to. Once you do, you can really make magic happen – for this enterprise.
If you are inside EY, working with EY data, you have a MOAT over OpenAI.
More than 10M chats have been created on Team-GPT so far.
All of this is proprietary data which we currently have our hands on.
While OpenAI has more data than us on these, we have ‘cross-model’ data (between OpenAI, Anthropic, Mistral, when do they prefer what, etc.)
We are also introducing novel interfaces (work modes) like: Pages. This allows our users to interact with LLMs in completely novel ways, with which we accumulate even more proprietary data.
While now we are integrating a lot of models, very soon we will be hosting our own LLaMa 3 and similar, owning a larger bit of the data and keeping it within our own ecosystem.