How to build AI RAG Agents that use large Knowledge Bases in Coda
without needing any Packs, APIs, or programming code.
Just uses normal Coda Pages, Tables, Formulas, and Coda AI Columns.
(a small version of Claude Skills - suitable for small businesses and departmental processes.)
Thanks to @Nina_Ledid for hosting this podcast.
Respect
Max
GENIUS !
Amazing idea and implementation @Agile_Dynamics !!!
More timely than ever, @Agile_Dynamics Agile_Dynamics! As always, you identified a massive pain point on your clients’ side and built an easy-to-understand solution for their citizen developers in Coda! Outstanding!
Thanks for sharing your brain with us, Max!
I feel privileged to share a community with you, Max!
So much I learn every time you share knowledge with the community.
Thank you! 
Just the other day I was wondering how to avoid all packs and build a simple tools that extracts the exact piece of information from documents/pages in Coda. This gives me a good idea. Next level is how to do that with table data with added complexity of data in various languages 
Very impressive, Max!
Thank you so much for sharing. You are a true master!
From a practical standpoint, this makes a lot of sense. I like the idea that you can magically blend everything you know through a product or a canvas you already know. I can’t think of a better way to transition to agentic systems that understand with great precision.
However, … taking an agent from 70 to 80% accuracy or competence (we’ll call it competence) to 90% using this approach is certainly possible. I haven’t measured how well this approach actually improves accuracy in generative AI. What I have measured is that using an ontology (a topic map) can get us to two nines of accuracy. Three nines is still out of the question unless we use symbolic AI, and that’s another big ball of worms we can’t practically utilize.
I think the industry is moving towards the idea that agents will be based on skills. Basically, skills become a higher-level form of an agent. We think of sub-agents and skills; they all blend together quite nicely in terms of descriptions.
However, in terms of actual implementation details, it gets pretty dicey until you introduce something like a knowledge graph. A knowledge graph needs to be implemented, not with Coda. I suppose you can simulate it, but I’m unsure if you can implement it effectively in Coda. Performance takes a big hit too.
Until I can actually build a superhuman agent, I should just shut up. LOL