No-Code: Stuck at the Trailing Edge

The acceleration of software development driven by new possibilities with generative AI is evident everywhere we roam. Today, we have but a third of the coders needed to build all that is envisioned.

There’s nothing inherently wrong with #no-code. Still, if you’re all-in using codeless platforms or generative AI to build a profession or a business and you want to work on leading technologies, you might want to pump the brakes and dust off that Python or JavaScript tutorial.

This article is mostly about economic supply and demand for software developers and less about AI. However, it also aligns deeply with the advent of palatable AI(1).

No-code is anchored to a dance of SaaS disruption.

No-code platforms are unable to participate in the coming expansion of software development. I'll talk about that later.

Please support my work by becoming a free or paid subscriber.

There have been many waves of AI since my coding career started, which is an extension of my accounting career. My accounting career ended abruptly when I tried to automate the auditing profession in 1978. My CPA partners “outsourced” me when I demonstrated that billable hours could fall precipitously if auditors were armed with my HP calculator MAG-strips. I woke up the next day at ABC Television, started writing a lot of software, and … as they say … the rest of the story unfolded. No regrets.

Code became my life passion. I couldn’t learn it fast enough. Last month, I learned Rust. This month—how to avoid learning more Rust by learning Mojo.

Generative AI has allowed me —even at my 70+ age— to expand on my skills in real-time data, intelligent conversations at human speed, near-instant inferencing, and opportunities to perhaps predict answers before users finish writing their prompts or saying what they want.

My original penchant and passion for writing code have blended well with my often irrational addiction for counting milliseconds. I deeply thank Lance King for pulling me subtly away from early retirement to build the first real-time video-analytics platform transportation.

I have worked on many no-code systems, but mainly as a coder. My consulting engagements are more like that of a ringer in a softball game. If no-code were football, I would be the long-snapper.

Knowing about me is much like understanding one of the most misunderstood specialized roles in the NFL (and increasingly College football). The “long snapper” is a specialist whose craft is not really known by even the most ardent football fans.

And like the long-snapper, no one cares if there’s a tiny snippet of code in a vast set of functionality created without [additional](2) code. They only care if it fails, precisely the long-snapper's realm.

No-code platforms like Airtable and SmartSuite are doing well. Many of the automation services are also killing it. And tools like Operator and FillOut are gaining momentum. Code may have vanished in this relatively limited world, but the complexities of a no-code environment remain. DevOps matter. Data capture matters. Testing still matters, although few do it. All of the shortcomings of no-code platforms represent gaps that must be filled. Opportunities exist all around.

I believe no-code is losing ground as a percentage of application expansion.

The urgency to push generative AI solutions into production has reached a pace that has now absorbed the vast supply of development resources. I have no data to indicate this is the truthful trend line. I’ll bet there is some, however. And if you can gauge development focus by adjacent trends, AI projects will saturate the marketplace now and through mid-2025.

Get a FREE one-year subscription to Impertinent if you can produce some corroborating data in the comments.

Leave a comment

Here’s why no-code platforms are anchored to the trailing edge of solutions and why it struggles to play in early adoption curves such as generative AI:

• No-code is winning at the expense of enterprise SaaS losses
• No-code platforms are still weak; they can barely perform well enough to achieve usual and customary business requirements
• Architecturally, no-code platforms were designed to mimick existing business requirements
• Code Intelligence (more on that later)
• Code is writing parts of itself (in some contexts)
• The demand for hundreds of thousands of new applications has emerged because of the latest AI wave

No-code is ill-prepared to offer solution-builders access to the primitives required to build the next 100,000 applications that lean into AI. No-code systems won’t be ready to address these requirements until half of these future applications are crafted.

No-code may forever be hamstrung by the same principles that guide laggard adoption.

I’ve interviewed many no-code platform engineers and innovators. They all view the current AI wave with great skepticism and universally believe users aren’t ready for the additional costs associated with inferencing at scale. Many believe the LLMs aren’t ready for mainstream business use. Some feel that a lot more experience is necessary to throw down on a specific strategy. This is rational. But it’s not only about the fear of adopting AI too quickly. They fear anything that could unsettle a relatively stable environment where no-code solutions compete well with SaaS competitors at 1/5th to 1/10th the cost. This is their wheelhouse—a lane proven ideal for basic requirements in marketing, sales, HR, product management, task management, and all things central to the SaaS movement.

No-code platforms are tearing into SaaS apps like Salesforce, Asana, SAP, SugarCRM, HubSpot, Zendesk, and Sage. They’re slowly dismantling these stalwarts by a thousand cuts with the help of information glue factories like Make and Zapier. As viable as no-code platforms are, this sector cannot play well at the bleeding edge because it is — by definition — designed to disrupt the status quo at the trailing edge.

While there’s no debate that no-code competes (to a degree) with solutions built by software engineers, it does nothing to stem the need for all of the applications that were once considered unaffordable luxuries and may now be engineered for a future wave of more intelligent solutions. No-code platforms will have little impact on that which now needs to be developed and which is cost-effective and made practical by advancing AI sciences.

But, the no-code leaders are adding AI features, right?

Yes. Slowly. If you examine the use cases, it’s almost laughable. They’re integrating features to accentuate data - summations, keyword extractions, and content generation. These are narrow, limited, and incrementalized me-too functionality. They lack innovation, and most companies won’t use them. Many subscribers will reject these features based on cost and risk.

As such, software engineering with code will flourish through 2032 and likely well beyond.

Just a few years ago - 2017, as I recall - experts proclaimed the #no-code revolution would eat the software industry, leaving it in a pile of rubble in just five years. Seven years later, they are still wrong.

PREDICTION: AI Will Trigger an Explosion of New Software Developers

Read the full story...

In that article, I cited four key reasons this growth curve was likely. I was right then. I’m more convinced now. I’ve seen some applications that generative AI will make possible and practical. Mind-blowing solutions are at the doorstep, and none are built of or for no-code platforms.

Future applications will drop jaws in the code intelligence sector. They are practical because of generative AI.

Companies like SourceGraph, and Phase Change will raise engineering bars on two fronts with AI. I'm penciling in Phase Change adjacent to SourceGraph for now, but with a notable difference in trajectory, objectives, and revenue model.

SourceGraph is known for Cody, a blazingly fast platform that allows developers to rapidly search, write, and understand code by bringing insights from their entire codebase into the editor. They focus on competing with Github Copilot, Ghostwriter, and others who play in the veneer of code development. They do this by building a knowledge graph of your code and applying AI to give developers generative productivity.

Phase Change is a different beast. Don’t freak out when you see the reference to COBOL. This technology is being developed for many languages, but I understand why they’re cutting their teeth on what seems to be a dead language more than 800 billion lines in production use every day. If they can make sense of almost a trillion lines of COBOL(3), the founders and developers will be very wealthy.

Among the many vectors that Phase Change pitches, this one resonates most with me.

Code comprehension takes longer than changing code. Developers spend 74% of their time understanding code--not writing code. COBOL Colleague understands your code at machine precision & speed. Get your answer in seconds and not hours or days. Spend more time on making code changes that matter to the business & less on researching legacy code.

A good friend and former partner (MyST Technology Partners), currently at Phase Change, mentioned in passing …

“LLMs are not a magic replacement for actual engineering/design thinking … many people are under the impression that LLMs can do the thinking for them. Consultants are encouraging the LLM-is-a-quick-solution-to-all-our-engineering-problems mindset.” — F. Andy Seidl (4)

I think it's more nuanced (and subliminally enticing) from the consultant's perspective. In the shadow of generative AI, they believe it’s possible to serve as an imposter in a deeply skilled environment such as software engineering. Many no-codists exhibit similar sentiments. Their questions in the no-code forums about coding issues tell a disturbing tale that typically starts with attempts to write code with ChatGPT. Some have made massive mistakes. Most can’t get their shit to work.

Like many consultants, they are empowered with additional degrees of unearned “engineering confidence” based on prompting prowess. They're delusional, of course, but they're also capable of leaning into this new realm of bullshitery with some degree of success. Indeed, they've expanded the BS meter to the point where even a logarithmic visual cannot adequately convey how tall our boots need to be.

Wrapping Up

No-code is doing well. It will continue to do well. It is not the harbinger everyone imagined concerning the elimination of code. The irony is that every feature request by a no-codeist is, itself, built with code. Sometimes, lots of additional code. We are poised at a time when AI research has created a galaxy-sized funnel of new things that must now be codified. As such, we need to prepare for a galaxy-sized expansion of code, not a compression, as AI hypesters and prompt clowns would have us believe.

AI-induced Examples

Advances that employ inline AI-generated semantic pinch and zoom features require software engineering. Who do you think is going to develop stuff like this? And what technologies will be required for real-time inferencing?

By the time Airtable implements features like this, many years will have passed.

Another example involves my research into Biasity(5), the ability to rapidly guide conversational compass headings to increase user success while lowering inferencing costs and increasing generative AI accuracy. This also requires real-time, event-driven inferencing and new engineering approaches that blend embeddings with large and small action models, and intelligent caching. We assume that users must be required to prompt and LLMs must be required to respond. The software serves as the buffer to accelerate productivity. Prompts as we know them will soon be extinct.

So Much To Build by So Few Developers

The cone of software opportunity is extremely large and growing for software developers. No-codeists and their cherished platforms cannot take us there or help us address the mountain of new software development opportunities that lay ahead.

The first phase of generative AI was hysteria. The second phase elevated the use of chatbots to new delusional heights. The third phase is software engineering, which harnesses the value of these balls of language data and vectors for both developers and the applications they develop.

No-codeists have tried to become relevant in the first two generative AI phases. Mostly, they have failed because no-code platforms are landlocked to a disruptive wave of their own.

While advances in no-code require a crap-ton more code with each new feature, advances in all software solutions require greater code intelligence, smarter inference-based code writing, and more capabilities to craft solutions that have never been considered anything short of impossible until recently. We have a lot of work to do.

Thank you for reading Impertinent. This post is public, so feel free to share it.

(1) Palatable AI refers to generative AI and a whole lot more. AI is now more palatable because it has made new applications practical and affordable that were previously not thought possible.

(2) All no-code features are built with code. No-code is a euphemism for “no additional code required”. Similarly, we misunderstand NoSQL which means Not Only SQL.

(3) An increasingly difficult challenge as old farts in my generation keep retiring.

(4) F. Andy Seidl - the one person who transformed me into a quasi-competent software engineer. I am grateful to Andy for my depth of testing skills, documentation, and requirements management techniques to this person. Sure, I’ve nurtured this knowledge for another 20+ years, but my modern engineering and architecture fuse was lit in 2002, 20+ years after I created LapLink.

(5) A philosophical premise that generative AI systems may benefit from intentional biases.