Build Ai Agents in Coda - Training - Beta Test Cohort

I want regular business folks to be able to build their own AI Agents in Coda.

It’s over 18 months since I first published my techniques for building AI Agentic Workflows in Coda (without needing any packs or programming) - just using pure Coda formulas and AI columns. (How I build AI AGENTS in Coda - #6 by Agile_Dynamics)

Since then, I have deployed a significant number of these autonomous agentic solutions for my clients, and the techniques have been refined and improved. And… Coda has vastly improved its AI capabilities as well.

Now I have the ability to add “Knowledge Based Agents” (RAG) to the process, again without the need for packs, programming, or external databases. (Coda AI RAG Knowledge Graph without Packs or APIs)

I am on a “mission” for 2026.

I want to empower regular business managers (with no IT skills or programming know-how) to become the AI leaders for their businesses - creating and deploying advanced autonomous AI Agentic Workflows using nothing but plain old Coda.

So many managers I have worked with over the decades have the unique combination of skills needed to achieve this - they know their business’s operating procedures and they have lots of experience using spreadsheets to model them and automate them using tables and formulas. They just need some guidance on how to build AI Agents with those skills.

I have managed to do this with several managers in my client organisations. I have delivered a series of training workshops on this topic to my biggest client - but due to NDA agreements, I cannot share the recordings of those sessions here, alas.

So instead, I have set up a ‘training hub’ to host my videos, tutorials, and workshops. It also provides a place where we can gather and share ideas, experiences, questions, and solutions about rolling out “Business-Operations AI” using no-code platforms like Coda.

(The techniques I am promoting also work on Notion, Airtable, Power Apps, and good old Excel.)

The topics I am covering are:
1. Building AI Agentic Workflows using only formulas and native Coda AI features
2. Using normal Documents to build Knowledge Bases for use by these AI Agents
3. Designing efficient Business Data Structures as Relational Tables in Coda
4. Techniques in BPM (Business Process Management) for automating SOPs in Coda
5. Build Superhuman AI Agents using Coda Formulas without needing JavaScript skills

Before publishing this more widely, I am asking Coda makers from this forum to join a 6-week “beta-test phase” of this, so I can get feedback on what works and what needs to be improved.

This program will always be free - it’s a passionate mission of mine - not a business.
(I make my money doing coaching and consulting.)

If you share my mission and would like to participate, please sign up via the link below (yes, it’s a Coda form)…

AGENTIX-AI Beta Program

You will get an email invitation to the hub. This is open to all Coda makers regardless of Coda skill level or experience with AI.

Many thanks in advance. I look forward to some interesting exchanges.

Respect
Max

7 Likes

What a vision for empowering business experts!

That’s what’s sorely needed in the Agentic Ai space, less “it will somehow magically work” and more “it will work deterministically, because we’re grounding our agents in editable knowledge and establishing the guardrails for governed actions”.

Signing up right now!

Can’t wait to learn what you’ve already unlocked so successfully with your corporate clients! Many thanks for this generous offer.

3 Likes

I love the passion, but I cannot avoid thinking about the [two] elephants in the room.

  1. The pace of Coda AI has all but seized up. I don’t think there’s been a meaningful release since the original team (David) built it. It has been stagnant, dead in the water for at least two years now. If there’s a profound ability to compete with agentic systems bound by the Packs SDK, it will be too little, too late.
  2. Workspaces Studio (and whatever Microsoft’s equivalent is). These tools are poised to address non-technical workers who “…just need some guidance on how to build AI Agents with those skills.”.

Am I missing something?

BTW, I’m pro-Coda, I’m not pro-Coda-agents perhaps because I’m in the dark. I can glean from the Packs Agents SDK, though, and it suggests:

The SDK allows you to build the capabilities (Skills and Tools) that the platform’s existing AI uses to plan and reason. You are effectively creating the “limbs” and “manuals” for a central brain provided by Superhuman Go, rather than building a standalone brain from scratch. You define specific functions (e.g., postToSlack, queryDatabase, resizeImage) that the agent can invoke.

1 Like

Bill,

I feel strongly that the elephant you refer to is not in my room for the following reasons:

A: This approach I am using is not dependent on Coda in particular.

  • It will work on any no-code platform that has the following attributes:
    1. use of tables with relations to store the business data
    2. ability to execute business logic with declarative formulas
    3. ability to represent knowledge as rich-text documents accessible by those formulas
    4. built-in AI inferences where the prompt can be built using formulas

So Coda, Notion, Airtable, Power Apps, and even Excel can be used for this.

B: I’ve been using Coda Ai daily in my projects and seen significant improvements

  • Unfortunately, Coda has not been good at announcing or promoting those improvements
    1. The underlying LLM model being used has been upgraded (ok, belatedly)
    2. The context window size available in AI Blocks and AI Columns has massively grown.
    3. The performance of the AI columns has improved enormously also.
    4. The Superhuman AI Agent pack building SDK is a major step forward.

For my clientele (who are not looking for leading-edge AI, just easy-to-use AI), these are more than sufficient to produce really impressive ROI for their time and energy investments.

C: This MISSION of mine is ‘across the chasm’ and well inside the Mainstream segment

  • This work is not for the Bleeding-edge, Early-Adaptors of the AI Tsunami
  • But it is aimed at the Mainstream Business-Ops world where reliability and stability are valued
    1. Breaking tasks down into small, single-purpose inferences does not need the latest LLMs
    2. We use formulas to build prompts, parse responses, and orchestrate workflows.
    3. So we find GPT-4 class LLMs are fast, cheap, and good enough for the job
    4. We specifically do NOT want thinking models that make their own plans
    5. We need simple, straightforward, single-task inferences that are deterministic and reliable

I follow your work with great interest, Bill, and I appreciate the depth and breadth of your experience and worldview. But I am avoiding the leading edge side of the ‘chasm’ and dumbing down the approach to deliver reliable, easy-to-use tools to busy managers who know SOPs and spreadsheet formulas extremely well, and want to just use that to harness AI in dull, boring, repeatable, work-a-day automations.

Like you I wish Coda was advancing its AI capabilities faster and more openly.
But that is (IMHO) for another thread in this forum.

I want to focus on passing on what I have learned, using Coda (and Notion etc) as it is, to those who are interested in it.

Respect
Max

3 Likes

SLC (simple, lovable, complete). Good stuff, and a good strategy.

This is really good to hear. I hope they open it up before we lose interest.

3 Likes

What’s the expected time commitment during those 6 weeks? I want to make sure it’s something I can feasibly commit to before signing up. Thanks!

@AJM
I plan a 1 hour workshop each week.
Followup calls with individuals if they want them.

Max

Thanks for this amazingly generous offer. Signing up right now, even though I share @AJM’s concerns - we’ll see how it goes.

I love your approach on so many levels, but I just wanted to pinpoint one. I’m no expert, but my understanding is that if we want AI to be at least a tiny bit sustainable from an ecological point of view, using lightweight models to perform very specific tasks is the only way to go. Seeing people use cutting edge models to perform “magic” on stupid and “expensive” tasks always depresses me. (Real life example whenever a new model is unveiled → “oh look, I’ve used AI to generate a short clip of myself drinking from a forbidden bottle of expensive whisky on a plane, see how clever and funny I am lol”). But what you describe here looks top notch to me, and I can’t wait to know more about it. Once again, MASSIVE respect!

1 Like

Welcome aboard @Bertrand_Bouteiller, you are most welcome!
Over 30 makers have signed-up and over 20 are participating already.

As this is a beta test of the whole process, I am keen to adjust the time commitment based on students feedback. Let’s see how that goes.

As for concerns about environmental impact: I also cringe when I see people burning mega tokens on frivolous things.

My own urge to minimize tokens is driven not by cost concerns, but by the fact that too many tokens crammed into the small and precious context window of an LLM is the main cause of errors and hallucinations during inferences.

Hence my “Precision AI” framework’s motto - use as much AI as you need - but as little as possible. Use formulas for compute and predicate logic - keep AI for knowledge, judgment, and linguistic operations.

I am almost ready for our first workshop next week. Meantime the first module of the course is available on the hub - Beta Course

And the beta program is still open for anyone interested (its free) here - Coda Sign-up Form

So, if you agree with my mission for 2026:

Help non-IT business managers who know spreadsheet & no-code formulas to become the AI Ninjas for their business in 2026 (and survive the ‘AI Tsunami’ that’s coming).

Then please come and join us.

:red_circle:➤𝖒𝖆𝖝

2 Likes

FIRST WORKSHOP ON THURSDAY 18 DEC

The first live workshop will occur on Thursday, 18th December 2025.

See the CALENDAR tab in the Skool group below for the details.

It will also be recorded and posted in the classroom later.
But this is a live interactive session, so it’s best if you can attend in person.
You can have your camera on or off - that’s up to you.
And I plan to answer any and all questions that arise.

This session covers the very basics of making AI Agents in Coda
(without programming, packs or external APIs).

  1. A quick demo of a simple Agentic Workflow in action to illustrate the concepts
  2. A quick overview of the techniques used to build the demo.
  3. A step by step live build of the same workflow live, so you see how its done
  4. Q&A session to deal with all your questions (smart and dumb questions)

If you have not yet joined our group, you can sign-up using the Coda form here..

:red_circle:➤𝖒𝖆𝖝

3 Likes