Coda AI Agentic Workflow Framework (Beta Release)

Agile Dynamics is a boutique consultancy that has been automating SOPs for Fintech, Aerospace Engineering, and Clinical Research CROs for decades.

We champion the “Citizen Developer” movement that empowers business process experts who are NOT developers to build SOP automations for themselves and their teams. To break free from dependence on the Corporate IT Department. :pirate_flag:

Coda has become our #1 choice of no-code workflow automation platform. Watch this video to see WHY we chose Coda; https://youtu.be/PF_DBeSUmUk

We were asked by a client to look into ways that Citizen Developers (aka Makers) could harness AI in the automation of SOPs. What has become the Buzz-Word de jour; Agentic AI Workflows

  • We built Custom GPTs in OpenAI but they could not be invoked outside the ChatGPT platform.
  • We expeirimented with MS Autogen and Google Vertex - with good results
  • We found Langchain to be the most flexible and extensive framework

But these platforms had a serious problem for Citezen Developers;

  • either they needed extensive use of Python to define all the business logic
  • or they tried to avoid using Python with rather lame, very limited, ‘no-code’ interfaces

We found ourselves saying,

  • if only they offered the kind of no-code capabilities we know and love in Coda!

And then BOOM!, a thought suddenly struck me…

Why NOT harness Coda as the platform to build Agentic AI Workflows?

Well, obviously! there are a few problems with that!

  1. Coda AI is a bit ‘behind the times’ and does not use the latest LLMs
  2. so the Context Window is limited (around 3,000 words we reckon)
  3. you cant specify which LLM you want to use
  4. Coda AI does not implement RAG - no way to ‘upload’ gigabytes of knowledge files & query them
  5. Coda does not have objects like Agents or Tools or Chains or Memory

But we experimented with the concept and found we could overcome these limitations.

  • most use-cases (simple prompt/response) are well-served by the default Coda AI
  • Coda AI is expected to be greatly improved in upcomming releases (we hope)
  • we use Packs in those cases where we need other LLMs (Perplexity & Gemini)
  • RAG and RAGNOR capabilities can also be added using Packs
  • we used the following ‘design pattern’ to replicate the Langchain objects in Coda;
    • Object Classes = Coda Tables
    • Instances = Rows in Tables
    • Attributes = Table Columns
    • Functions = Formula Columns and/or Coda AI Columns
    • Methods = Action Button Columns

Using this approach, we have sucessfully built Agentic AI Workflows (aka Chains) in Coda where we use the Coda Formula Language (CFL) to replace Python code.

So today we begin a BETA Program to test this framework. And we invite makers to join the program by sending me a PM (private message) here on this forum.

We are looking for makers who…

  • already know CFL and Coda-making quite well
  • have some familiarity with OOP
  • have good knowledge of DB design (3NF)
  • understand the methods of AI Agents and RAG (or are willing to learn)
  • have use-cases in mind for using AI Agent Workflows for beta testing

You dont need to know Python or pack making - the packs are already built.

We call it the AGENTIX FRAMEWORK - because it more than a PACK, more than a TEMPLATE, and more than a simple maker technique. Its a ‘methodology’ for designing and building complex workflows using Coda Tables and Formulas (and a few Packs) .

There is a Pack for the framework (but you need to register for the Beta Program) to be able to use it correctly.

We have included our Knowledge-Graph UX Design technique in the framework, which helps makers design clean database models of business objects, and the UX elements to implement them.

Respect
Max

6 Likes