Two A's: AI and Automation

Airtable Questions
1

PREDICTION

Any day now, I expect Airtable will announce a bold new feature that will serve to reinvent itself in the no-code sector. In so doing, it will expand the #no/low-code mantra to include - wait for it…

#no/low-thinking

Gartner will herald this as the most remarkable big-tech move yet. Databases not just for all, but for people who are not only ignorant about information systems, but also who generally have no interest in learning anything or exerting any effort to understand how data should be collected, organized, or utilized.

Sorry to come off as sarcastic so early in the day, but there will be some truth to this looming announcement. There might also be some benefits as well.

  • It will allow you to describe a computation and it will design the formula.
  • It will allow you to describe an automation process, and it will build it.
  • It will allow you to describe a table, and it will do its best to build what you envision from your words.
  • It will allow you to describe a solution, and it will build the base, the tables, and the automations that are most likely to achieve your objectives.

Is this good? Where might it go rogue?

2

That’s not no/low thinking. You still have to describe what you want. You just get to use English as the language to describe what you want instead of learning a different language. The AI then translates you English-code-description into a lower-level language and takes care of pesky details for you.

When I write JavaScript, I am using a higher level language that lets me focus on the thinking. Then the computer compiles the JavaScript into a lower level language that can actually be executed. The complied takes care of pesky details like memory allocation, garbage collection, data typing, and a bunch of other stuff that I don’t know or care about.

No-code adds another layer to the same process. Instead of typing out computer code, the no-code tool lets the developer describe what he wants, without having to deal with pesky details like spelling and grammar. Then the computer compiled that into lower and lower level languages until it reaches something that can actually be executed. The no-code developer still has to think; he just doesn’t have to think about things like tedious stuff like spelling, grammar, memory allocation, handshakes, etc.

Knowing what you want takes thinking. Having AI translate what you want into code that the computer can execute will enable more people than ever to think because they will be able to actually implement their ideas.

I say that no-code tools leads to an explosion of thinking, and the addition of natural-language tools will also lead to an increase in thinking. Will there be bad/incomplete/faulty thinking? Sure, because there is so much thinking.

Back before typewriters, very few people were novelists. Writing prose required a bunch of skills that had nothing to do with coming up with a good story. Typewriter and word processors have enabled more and more people to try their hands at being writers. There are even more bad novelists now than ever before, and even more good novelists too. The good ones have to think better than the bad ones, not type faster or have better handwriting.

If people want to brand this as no/low thinking, that’s just a marketing ploy.

If you think that describing what you want doesn’t take thinking, you are probably just really good at thinking.

3

How do you think this is going to go? Who amongst us is likely able to craft what we want from a collection of words?

4

I have no idea. Lots of people, once the tools are widespread enough to use.

When a toddler learns to think, it is an amazing process, in part because you know that babies aren’t capable of much advanced thinking. Thus having the cogs in the brain visibly engage at all is a minor miracle.

Watching a teenager learn to think is similarly amazing but also really frustrating. The pre-teen was capable of thinking, but is now slammed with so much more to process that what previously worked is no longer sufficient. It might appear that the teenager isn’t thinking when the teen is really working harder than ever, just sometimes without great effect.

For me, learning to code was like going from a baby to a toddler, and now looking at AI, I feel like a teenager.

1 Like
5

Wait! I thought you said recently [somewhere] you would not likely use AI for development because it was too imprecise. Did something change that caused you to feel more chummy with silicon?

6

I’ve stated that so far my attempts to use AI for development have been unsatisfactory. Thus far, it has been easier for me to write the code I want without using AI. However, that doesn’t mean that I won’t use AI for development in the future.

AI tools (and my current ability to use them) just aren’t at a place right now that efficiently produce results better than what I can produce without AI, at least for the limited things that I do. That’s what makes it the teenager stage for me. I know that more is possible, and I’ve dabbled with it a little, but I’m still learning and making mistakes, and have not yet fully embraced it as normal operating procedure.

7

Me: Design an Airtable schema for CRM with tables for companies, contacts, opportunities and tasks

ChatGPT: Sure, here is an Airtable schema for CRM with tables for companies, contacts, opportunities, and tasks:

Table 1: Companies

Field Name Field Type Description
Company ID AutoNumber Unique identifier for each company
Company Name Short Text Name of the company
Description Long Text Brief description of the company
Industry Single Select Industry the company belongs to
Phone Phone Number Company’s phone number
Email Email Company’s email address
Address Address Company’s physical address
Contacts Linked Records Links to contacts associated with the company
Opportunities Linked Records Links to opportunities associated with the company
Tasks Linked Records Links to tasks associated with the company

Table 2: Contacts

Field Name Field Type Description
Contact ID AutoNumber Unique identifier for each contact
First Name Short Text Contact’s first name
Last Name Short Text Contact’s last name
Company Linked Record Links to the company the contact is associated with
Title Short Text Contact’s job title
Phone Phone Number Contact’s phone number
Email Email Contact’s email address
Opportunities Linked Records Links to opportunities associated with the contact
Tasks Linked Records Links to tasks associated with the contact

Table 3: Opportunities

Field Name Field Type Description
Opportunity ID AutoNumber Unique identifier for each opportunity
Opportunity Name Short Text Name of the opportunity
Company Linked Record Links to the company associated with the opportunity
Contact Linked Record Links to the contact associated with the opportunity
Stage Single Select Current stage of the opportunity
Value Currency Estimated value of the opportunity
Close Date Date Estimated date of closing the opportunity
Tasks Linked Records Links to tasks associated with the opportunity

Table 4: Tasks

Field Name Field Type Description
Task ID AutoNumber Unique identifier for each task
Task Name Short Text Name of the task
Company Linked Record Links to the company associated with the task
Contact Linked Record Links to the contact associated with the task
Opportunity Linked Record Links to the opportunity associated with the task
Due Date Date Date by which the task is due
Priority Single Select Priority level of the task
Status Single Select Current status of the task
Description Long Text Detailed description of the task

This Airtable scheme should allow you to effectively manage your CRM data, and easily link between companies, contacts, opportunities, and tasks.

Not perfect (too many linked fields, not enough lookup fields), but this is a better structure than I’ve seen on a lot of bases.

8

Yep - GPT is able to layout a reasonable approach. The feature they will announce will probably actually implement that.

9

I’m guessing in the end it will come down to money, unless these tools somehow become ‘free’. How much will it cost to keep using your words to iterate, iterate, iterate… until you get it right.

10

Costs for AI inferencing will fall quickly. Some speculate it will be approach free by 2025. A few have said the cost of measuring inferencing will soon exceed the benefit of charging users for it.

The cost of LLMs is sunk; it’s in the past. Using LLMs is very cheap now. And beyond GPT-4 (and its cousins) it’s as large as we’re likely to see for a very long time. The cost of inferencing will continue to plummet. The big money will be earned with productivity increases.