Why I focus on Coda AI?
instead of calculations and complex business cases
I am an active Coda maker since early 2020. Before, I used AirTable to calculate business cases and follow up on plannings. Coda offers — from my point of view — a better alternative for AirTable and spreadsheets.
I don’t need speedy tables with 100K rows. Instead I want to write text and structure data. In the text I explain why we applied the assumptions and how to use controllers like sliders to go develop a feeling for a project. The extras Coda offers like adding images, icons and so forth are nice, but not essential to me. Still missing however is proper printing and permission management.
Even without these two basics, Coda was for a long time a no brainer compared to other applications, including spreadsheets.
As time passes by Coda added features, often improvements but not all. The recent ‘calculation’ seems not to encourage writing your own formulas. It puts guard rails where it should have focused on improving Coda coding skills. This new element arrived after I wrote most of this blog and long after I decided to shift my focus and relates to what follows.
It takes time to learn how to Coda
When companies invited me to work with them, mostly they assumed that my work could not be too difficult and that my main advantage is that I can set up something fast (and affordable) which they could understand and work with.
I noticed a few returning issues.
- The knowledge gap between me as an expert and the client as maker became too large and the solutions created became hard to maintain by the client. Accepting that in Coda everything is a list is one thing. Translating this concept and get it to work in your doc, is something else and not easy at all. It takes time and practice to learn how to Coda.
- Most clients assumed I would become an expert in their field and thus understanding their product and company logic as a condition for building smart docs. This is near to impossible, time consuming and expensive for me and the client. I cannot invoice my time properly and the time I need to understand their logic feels for the client as a waste of time and money. I am expected to understand and to know their business as they do and often even better.
- I met many gifted Excel wizards and the idea was that Coda could and would replace large parts of the spreadsheet logic. However it seems to me that Coda did not fully pursue the ambition to offer an alternative to spreadsheets. Although I see the benefit of Coda, for most Excel wizards Coda did not really provide an alternative and as far as it does, it requires quite some insight in the Coda formula language to get there. Learning how to Coda is harder than assumed.
General trend
In the community Nad summarized the evolution of Coda rather well:
Like me, he has no insight on the internal workings of Coda. But I believe it hit the nail on the head. To me it seems that bit by bit Coda moved away from docs as powerful as an app to the idea that Coda brings all your words, data, and teamwork into one powerful doc. This is not only a change of slogan, it relates directly to where the team invests in. For example much asked for features as proper permission management (2-way-sync) will only be supported under the condition that users share rights in the various docs, which is what most makers want to avoid.
[update Oct4, 2WS stage 4 will handle unbound 2WS, thus without shared permissions, more about it in this blog ]
The same goes for Print (to Pdf), we don’t see anything happen in this area while makers ask for it all the time, even to the extend that pack makers assume they can make money by ‘solving’ this problem Coda should fix properly in my opinion. Packs get very much the status of duct tape, what Coda does not offer natively, you can fix by adding a pack.
This is counter productive. I have a pack that permits advanced find & replace in any doc while respecting the mark up, thus headers, bullet litst, bold etc. Once Coda notices how much appreciated this is by enterprise clients, they can turn this — from a technical point of view — into a native function. Since I make my living with such a pack, they may be seriously hindered in doing so. What counts for me, counts for many. These packs in the end block the Coda potential. It is really a pity.
More focus on team collaboration, on shared content, does does not exclude any CRM logic, but it makes it harder. The same is true for shifting away from spreadsheets and using Coda instead. You remember the image below? Something like this still exists at the bottom of the Coda.io site, indeed at the bottom.
The main focus is on what you see below, the images comes from a doc comparing Coda with Notion:
Writeups and wikis are very much text oriented, trackers and databases, like workflows are likely a mix of numbers and text. Text is dominant in this approach, unlike in spreadsheets where numbers reign. Text is also the main focus of most Large Language Models.
Coda AI — the future of the doc
AI is since the public launch late 2022 mainly known and applied in creative contexts. Write me an email, give me three reasons to, summarize in 5 lines and max 110 words etc. This is the entry level, no specific skills are required.
The next level is getting specific outcomes based on proper prompting. The development of this skill presumes practice. You test, you review, improve, adapt etc. In this context I proposed a prompt table that keeps track of your prompts and enables you to learn. More about this prompt table below:
The third level is where my work is mostly situated. You have a framework of texts, likely contracts and you have data sets you want to integrate in these text. We use AI to get the job done instead of the classical templating tools. AI will do it faster and allows smart modifications of the type you are unlikely to obtain using ForEach() Format() or any other native Coda function.
To make full use of AI in a Coda context :
- You need to understand how to Coda. I recently wrote two blogs about some essentials: blog 1 and blog 2. Both prepare you for the job on how to define your data architecture, which is crucial in an AI context. You need to feed your AI blocks or columns with data.
- You need to understand how to prompt. I use three levels : 1) Roles and Goals, 2) Ask a taks, 3) Format. Examples are key, certainly when you want to have an outcome that follows a specific pattern.
- You need to be willing to spend serious money. This line I added on Nov 8 when I had read the community contribution of David. AI usage is expensive. Coda will add credit packages you can purchase, but I am not sure it is smart to run your tests in Coda, when you cannot keep track of your learnings (prompts) intelligently and you have no control of the LLM used — Bring Your Own Key , BYOK — is required in my preferred set up.
Although the Coda realm still generates new challenges, even for experienced makers, fundamentally the coding logic required to deal with structured data, is not going to change. I can promise you that it will pay off to understand how to Coda.
Before my AirTable experience, I knew nothing about any code related logic, I simply started learning. What makes AI different is it’s unpredictability and that it is very easy to start with. It generates new content with a single click and is great to generate unstructured data (free texts). It is easy to start with because you can ask almost everything in simple natural language, no training required. Prompting only becomes relevant as of a certain level. Over time the quality of your prompting skills will become less relevant as the AI will suggest prompts to help you out.
I believe that AI will change the internet as mobile Internet once did. In the beginning the focus was for many people on what did not work and why you did not need it. Today we cannot imagine to live without internet on our mobile devices, but it took time. Expensive mobile phones in the early days got replaced by affordable well performing devices today. In terms of pricing, I expect the same with AI, but not without pain and struggle. We all got used to phone subscriptions, every individual, while people in the 90ths had one landline for the whole family. That got replaced by a subscription per family member. We accepted the relative high prices for connectivity. Something alike for AI, first it passes via one account, then you BYOK arrives and we all pay more on an individual basis, but we have something related to us and only us.
AI will offer serious opportunities for many verticals and I believe we have to be early with building and expanding our skillset. Besides Coda needs to step away from the easy use cases and create light house examples by allowing advanced AI usage inside the documents to inspire makers of all sorts.
There is a lot of AI stuff unexperienced users can do without any serious training. That will motivate many makers to make use of it on a daily basis. However in a Coda context the real added value comes from blending structured with unstructured data.
Besides calculations I’ll focus on contracts and content generation with AI. A new world is waiting for us to be explored. We are pioneers.
My name is Christiaan and blog about Coda. Since the summer of 2023 mainly about how to Coda with AI to support organisations dealing with texts and templates. My blogs are for beginners and experienced users. The central theme is that in Coda everything is a list.
I hope you enjoyed this article. If you have questions feel free to reach out. Though this article is for free, my work (including advice) won’t be, but there is always room for a chat to see what can be done. You find my (for free) contributions to the Coda Community and on Twitter.
Coda comes with a set of building blocks ー like pages for infinite depth, tables that talk to each other, and buttons that take action inside or outside your doc ーso anyone can make a doc as powerful as an app (source).
Not to forget: the Coda Community provides great insights for free once you add a sample doc.
