Google is going to breach your "AI Moat"
So whats your business plan to mitigate that eventuality?
Google (and other hyper scale platforms) are going to breach whatever “moat” you think your AI startup has built.
It won’t happen overnight, but it will happen.
Here is a story of an AI service I use that Google is about to replace.
Ask Shagility
I have created two “AI Clones” as an experiment to see if there is any value in using this type of capability to augment my Agile Data Coaching service.
You can try them out here:
Agile Data Way of Working coach
https://agiledata.coach/agile-data-ways-of-working-coach/
This one has access to all my content, all my articles, podcasts, videos, templates and a few other things I haven’t published.
Information Product Canvas Coach
https://agiledata.coach/information-product-canvas-coach/
This one only has access to the Information Product Canvas content.
Invoke a McSpikey (experiment)
My experiment hypothesis for this was three fold:
Would people actually use an AI Clone?
How do you build an AI Clone?
Would I use the questions they ask of the AI Clone to prioritise the content I need to create?
As this was an experiment (McSpikey) I wanted to do it with the minimal amount of cost and friction as possible.
And so while Nigel and I could have built this into our AgileData Product, a faster time to answers was to use a SaaS product that enabled this experiment.
I ended up picking Delphi.ai
I was pretty impressed at what this product provided me with, the key things I have valued so far are:
I can give it access to all my content, it was multi modal;
It can scan my websites and harvest all the content, rather than me having to add articles one by one;
I can point it at my podcast and it will generate the transcript, removing the need to manually export and upload;
I can embed the clone UI into any of my websites;
I can have multiple instances of the “AI Clone” and share the same content across all of them;
There are a number of easy options for me to tweak the style of the “AI Clone”;
Its backend prompt engineering which I can’t see is pretty damn good at parroting my tone and style;
It shows me all questions that were asked and all the answers given;
People can register and I can then see their specific questions and manually respond directly to them via the AI Clone;
I can create a paywalled subscription if I want to and it will take care of all the billing and payment processes.
I pay a fixed monthly subscription, I don’t have to worry about LLM token “bill shock”.
There are a few other features I haven’t had time to experiment with yet, for example automated actions.
The next step
If the experiment ends up being successful (people use it and find value in it) then the plan would have been to bring the core features I needed into our AgileData product via Nigel engineering magic.
I can see how we can reuse the patterns of these AI Clones in multiple ways in our product that have potential value.
For us it would just be another set of features we use, when we need to use them.
Enter the Google Juggernaut
And then I saw this post from Google today:
Portraits: personalized AI coaching built alongside real experts
https://blog.google/technology/google-labs/portraits/
This smells a lot like the same thing Delphi (and other SaaS products) have built.
We are a Google Cloud only platform, we decided early on to go boots and all on GCP and leverage its native serverless capabilities to save us time and money, compared to the other data and cloud infrastructure options.
If we built this “AI Clone” capability into our product , we would use the same GCP services Google are using for Portrait.
So if Portrait comes out of the Google Labs as a GCP service or another Google product format we can use it would make sense for us to Lease over Build.
But what about “AI Clone” products like Delphi?
Where is their “moat”.
Its not the backend services that allow me to upload my content. Google can do that already, I use that feature everyday in Google NotebookLM.
Its not a data moat, its all my data, I can move it anywhere I want to.
Its not a LLM moat (I don’t think), i’m guessing they are using a combination of the standard LLM’s available and haven’t created their own bespoke Small Language Model.
It’s not a prompt engineering moat, my guess is the Google team can create the equivalent prompts to provide that level of personalised tone and style.
It might be a UI moat. Google are not the best when it comes to creating non technical UI’s, but then do I value that UI at USD $99 a month over what ever Google charges?
It might be a feature moat where google doesn’t do everything I find useful from Deplhi. But they will eventually get to that 80% stage where they have enough to make it less painful to move.
It’s not just the “AI Clone” products
Its not just the AI Clone products that have a moat that will be breached by hyper-scalers like Google, that is just the example that popped up into my feed today and triggered the need to think and type.
We have seen the LLM powered tools remove the moats for photo / image identification, how long will the “identify this plant” apps survive?
We have seen content / copy writing apps start to get replaced.
We have GenAI alternatives to manual image generation and stock art sites .
The juggernaut has just started to get rolling.
So whats your “moat”
It still takes a lot of thought, sweat and effort to build a start-up, even with the latest “vibe” coding tools.
What’s your moat that means the next Google Lab announcement won’t render your product dead in the water?