Decimal Comma instead of point

Years ago, I asked Coda for an option to use a comma instead of a dot in numbers. Large parts of the word write numbers like so: 1.000,00
It is still not implemented, right? Or am I overlooking something?
I mean, really? Do I still have to use Google when I work with numbers? Come on Coda…


Hey @Erik_van_Schaaik, could you expand on where you’re seeing this? A screenshot of some examples might help! In a number column I can write large numbers comma separated, so I want to make sure I’m following the use case you’re running into here.

@Jasmine_B the user is referring to number formats like this that are used commonly throughout the non-American world:

234 725,45

(American equivalent: 234,725.45)

There has been a long-standing, large chorus of users asking for international number and date support. Even though I use American number formats and can tolerate American date formats (even if they’re not officially correct in my country), I agree that it’s high time Coda implemented this.


I completely agree with what you said, @Nick_HE. Latin-based languages like Spanish, French, and Portuguese have different conventions when it comes to numbers and dates such as switching commas and dots.

Moreover, for dates, we put the day before the month. Even in the UK, the day comes before the month. Currently, there are no options like " 31 January, 2023", " 31 Jan, 2023" or " 31 Jan".

Custom formats may be the best solution for it (similar to Compose), as we have in Excel or Google Sheets.

CleanShot 2023-09-27 at 13.27.29

It would be great to see Coda paying more attention to markets outside the US.


Thanks Nick, for making that clear. :slight_smile:

I don’t have any inside knowledge of Coda’s plans here (in spite of the badge on my avatar), but here’s my armchair midgame analysis:

Coda wasn’t built with i18n in mind. Even in the early days, I remember struggling with time zones because it became apparent that nobody on the team had tried using Coda outside of California :sweat_smile: (events would show up on wrong dates in calendar view if you were East of Pacific Standard Time).

My hunch is that, in that startup-y rush to get an MVP to market, i18n was parked as a “we’ll add it later” concern - which is totally reasonable! But I suspect that some architectural choices were made early on without consideration of what eventual internationalization would require, and now it’s very difficult to bolt it on after the fact.

Now, perhaps I’m wrong (after all, Coda is full of startlingly good architectural decisions that still serve it well in use cases far beyond what was originally envisioned). Maybe Coda is successfully selling plenty of European enterprise subscriptions, and internationalization is just not big enough a pain point to warrant the resources.

If you agree that Coda’s America-centricity could use some foreign flair, please let these cowboys know by voting for all of these feature requests:


I think the answer lay partly in generative AI. AI-defined software needn’t be concerned with localization of content OR application functions, displays, or computations. (in my view)

Is it possible Coda AI can mitigate this issue while we wait for localization to catch up? Granted, this must be performed as text, but I suspect there are ways to wrap the render points with AI without too much difficulty.

CleanShot 2023-09-28 at 16.56.51@2x

The problem with AI is that it’s not precise. It can work for most, but you don’t know if or why it decided not to change one row or put it differently.
I don’t see any real-case usage of AI in Coda, although I use it a lot outside Coda.


I tested it with a thousand numbers and it was flawless. Do you have examples where it fails?

This is exactly the opposite of my experience. I have not found any practical reason to use generative AI in an isolated context where my own data does not exist. Consider this example which is highly productive, but almost impossible to achieve in ChatGPT…

1 Like

Nobody will ever disburse a sum generated by AI. And nobody in their right mind will create a complex budget with the help of AI. Numbers are always determined by an intelligent human being, not by AI.
This is probably whole new topic, but since it is brought up: Personally, I would much rather be able to write in a cel ‘=100/2’ resulting in 50, then write a sentence asking AI to divide a hundred by two, assuming it is smart enough to do the math.

1 Like

I tested a few times. Last, I asked to check a column with the name of a product and select the category equivalent. The name has the category for most of the options. So I gave some instructions: if the product name includes the word “Premium”, for example, “Product Premium X”, set the category as “Premium”; if “Product Base X”, the category “Base”. From 200 rows, it filled correctly around 80%, leaving some of them - that could perfectly match the given example, and 2 or 3, it chose the category Base for a Premium product - which was in the name of the product. Ah: also invented a category “X”, which was not listed.

That was my last try. I have tried it before and had the same level of inconsistencies. I don’t trust.

I’m not saying it’s useless, but it doesn’t work for me. I don’t need to resume or evaluate texts, evaluate mood or anything in this sense. It’s all about reducing steps in the processes.

If AI could help me to build a more complicated formula, I would prefer it. Because I will be sure, that calculation will not fail.

On the other hand, I have been asking on ChatGPT 4 about how to build formulas in Coda to achieve specific results, giving names of tables, columns and named formulas. It helped me to build formulas I would not know how to do without trying many times until I achieved them. I learned much about the formulas and logic in this process, using less now.

I feel this is getting a bit off-topic. Sorry, but it’s an exciting topic!


In this case, the sum is already determined. We are simply formatting the value, not computing the value.

And I don’t recall suggesting this.

When you make broad claims like this, it is best to consider adding some predicates. For example, AI has been producing trillions of numbers for the past thirty years in millions of applications and services. The salad you ate yesterday arrived at your table based on AI computations. The MRI a loved one received produced images based on AI using numbers that are far more critical than your budget values.

These numbers are not created by generative AI, the new class of AI we are using in Coda AI. In fact, generative AI has no capacity to perform math because it is designed to tell you the next word given a collection of words, specifically tokens. While the LLM knows what the number 22 means and the number 10 – it has no concept of the number 1022 or 2210 because these values resolve to two tokens - 10 and 22. This is why generative AI is incapable of computations.

And that is flawed thinking as well. Since the dawn of micro-computing, math computations have contained the infamous Java floating point bug. 100/2 may actually be 50.0000000031.

What do you think generative AI would make of this? :wink:

So, this is a different problem you are trying to address than the original question, right? If so, I can’t really comment on why you are not seeing more deterministic outcomes unless you show the exact prompts and sample data you are subjecting to that prompt.

In almost every case that I’ve worked on where a developer concludes generative AI is untrustworthy, the issue lay in the prompts, not in the ability of the LLM to produce deterministic results.

By the way, I liked the article, and I did similar tests with Bard+Gmail, and it looked impressive. It was a little inaccurate, but we can see the potential of what’s coming next.

1 Like

You’re probably right, but it means that unless you decide to master prompts, you will probably make mistakes without having any idea, since AI uses “natural” language. A wrong formula gives me an error, and a wrong prompt does not.

Yep - if you decide it’s not worth the trouble, then you stay with formulas. But if you’re not confident in formulas, you need to write code.

I have seen issues as well. It often loses track of the fact that I asked it to use my Gmail archive. You have to remind it often. This is probably a bug and there are likely many since it’s in beta.

This topic was automatically closed 90 days after the last reply. New replies are no longer allowed.