The tide is here
The profound feeling of willing digital products into existence
Imagine what it would feel like to will a digital product into existence. Not having to search an app store or hire a team. Being able to remix and modify what mattered to you. Wholly your own.
The tide is here.
Low-code/no-code products have long offered this promise — but have failed to widely democratize access to that feeling.
They tend to be expensive and targeted at businesses: OutSystems, Mendix, Zoho Creator, Microsoft PowerPlatform, Retool, Bubble are all modern examples of Microsoft Access1 from yesteryear.
They excel at relieving burden from product development teams and IT in an effort to help non-developer functions speed up their workflows.
But they also abstract. Often too much. They have their own flavors, templates and models to capture builder intent — through custom, proprietary wysiwyg UI editors.
They tend to struggle with complexity, lock you into their ecosystem, have limitations related to security, are difficult to scale and fail on the premise of being “no-code” for those that actually do bill themselves that way.
No doubt theres a large market for what they offer.
But when you use these platforms the feeling is like learning how to ride a bike with training wheels you can never take off.
Think about them: the waves of change and alteration, endlessly breaking. — IX. 28
There is no shortage of skepticism about the impact Generative AI will have on the field of software development. This week Gary Marcus published this substack claiming that it wouldn’t (or more precisely, hasn’t yet) 10x computer programming2. Some excerpts of evidence to support:
A 26% improvement in tasks completed is a far cry from 1000% (10x)
An increase in code churn vs. reuse (defying the DRY principle) leading to downward pressure on quality
But what if we’re asking the wrong questions about the wrong demographic?
Most, if not all, studies and surveys3 focus on organizations employing “career programmers” of various stripes by analyzing change sets in repos. Something I’m quite familiar with.
But stepping back what precisely is a developer? Does it differ from a computer programmer, coder or software engineer?
Broadly speaking they’re the same thing. They translate ideas into code a machine can execute. Differences emerge regarding scope, mindset and of course how pedantic you’d like to be.
LLM focused empathy
New tools lead to new paradigms that require change. And those changes can be profound.
My argument: Most don’t fully appreciate HOW to best interact with LLMs yet. Professional developers and novices alike
It’s not hard to understand why. If you’ve historically focused on implementation (i.e. writing the code) you may have less skill in comprehensive planning. It could have been “outsourced” for you.
But current generation products are exceptionally good at implementation given the right plans.
Example: Look no further than Jason Zhou on Twitter overview his Cursor, Claude, o1 and v0.dev workflow.
The video is 42 minutes of gold. And re-impresses how specifying precise plans with LLM focused empathy gets incredible results. But make no mistake: it’s neither easy or simple.
What we should be measuring
I’m vastly more interested in how new LLM powered IDEs and services unlock access to those who have been on the edges.
The folks in and around the implementation of software development. Product people designers, data scientists, testers.
The folks that know enough to be dangerous and can learn quickly what they don’t.
We need to start measuring the impact of generative AI on them.
And more broadly speaking those outside of this orbit that have great ideas but have been fearful or otherwise unable to start.
And not just their “productivity” but that feeling.
Because being able to go from 0 to 1 with far fewer constraints is a profoundly satisfying feeling.
And with English, tenacity and focus you absolutely can.
The tide is here.
Released in 1991, MS Access unsurprisingly lives on. But in a permanently frozen state. Its last update was 2021.
Often a naysayer against the current VC-industrial-AI-hype complex, from what I gather he believes (to his credit) in a more hybrid approach recognizing the limits of the deep learning models powering today’s SOTA.
Another example: the StackOverflow Dev Survey.