This post is a reflection of a comment from @Kuovonne. Thanks for the inspiration.
I don’t think AI is ready to create complex logic …
I agree, and I think we need to be careful about the expectations of AI as code creators. However, you used a few predicates (++ for doing so). It caused me to reflect on this comment with the need to clarify.
Low-Code
I don’t think AI is ready to create complex logic …
Most novice coders don’t need complex logic; they want usual and customary logic. The type of logic that is in abundance and represented on the public Interwebs in spades. This is the type of logic generative AI is very familiar with and can replicate with good success. If also presented with learner shots - e.g., in the context of a copilot, the output accuracy climbs from ~80% to ~98%.
The predicates I use here are “usual and customary”. And even if you are a more advanced low-coder, you benefit greatly by getting the usual and customary elements of the logic out of the way quickly so you can focus on more advanced logic that may be required.
Generative AI for low-coding activities has value TODAY for EVERYONE but not for every context.
Clear AI Instructions
[Novice low-coders] … who struggle to clearly articulate their needs.
Again, an excellent choice of predicates. This is a deeply intertwined aspect of the ability to convey your desires, a task that is not unlike conveying technical requirements. Some of the best engineers cannot describe in words what they want. They might have a better chance than a novice low-coder, but the vast indictment of AI by expert coders shows they struggle composing prompts much the same as novice coders do. In fact, I think they struggle more than novices.
AI and Domain Experts (aka novice low-coders)
This persona is actually better equipped to use AI to create solutions. They understand the business requirements, and they are typically accustomed to using words in their work far more than software engineers. I predict Replit’s vision of a billion coders will be spot on, and largely for this reason.
The merger of low-code, low-ops, and AI is undeniable; it’s as if they were always intended to form this triumvirate alchemy designed to overcome the shortage of software engineers who invariably struggle to match technical requirements to business requirements.
AI plus domain expertise plus CaaS (code as a service) is the future.
Low-Ops
One of the big advantages of low-code in Airtable and Make (for example) is the streamlined and largely hidden ability to deploy your code and run it. This is advantageous for novice and expert low-coders. You need not worry about where your code is deployed or how it is hosted. It just runs. Coda’s Pack Studio is a similar approach, and many SaaS platforms are quickly realizing there’s something beyond the omnipresent model - IaC (Infrastructure as Code).
This relatively new approach is spreading quickly. Replit has a big deal with Google and a hundred million in top-tier funding to improve it. Microsoft is betting heavily. Amazon is partly at pace with early investments in Lambda. Val Town actually has a marketing slogan that articulates the benefit of IaC -
Val Town asks the question we should have asked 10 years ago.
Why can’t code run anywhere and everywhere in a social network?
Val Town is a social network where your code runs, can be shared privately or openly, and can be subclassed, forked, and deployed in seconds. Like Replit, it will eventually integrate AI with a low-code copilot. But with its API, I am already doing that with Coda AI serving as the low-code development copilot. Val Town serves as the lo-ops facilitator.
This demonstrates the agility and capability of IaC platforms - they can live anywhere, run at any time, they can be scheduled, and be integrated into almost every platform.
If Airtable (and other low-code/low-ops platforms) don’t facilitate IaC, a competitor will.
Airtable and IaC
It’s already possible for an Airtable solution to manifest a low-code/low-ops solution leaning on an IaC provider. It’s not ideal because Airtable doesn’t support a pluggable editor suitable for software development work. But there are ways to overcome this limitation - they’re just not pretty.
Conclusion
No-code was both the beginning and the end of no-code. That surge proved code was necessary. Low-code has proven to be extremely important and valuable. Low-ops have also proved itself as a key requirement. AI doesn’t change a thing about these two realizations except to squeeze out the remaining elements of complexity which will ultimately lead to low-effort-code.
The path ahead for low-coders is very bright.