#1 Re-learning to Code in the Age of AI
One month in, things are getting interesting...
A little over a month ago, after more than a decade, I decided to re-learn to code.
In that time, averaging about 15-20hrs/week, I built the following:
Sparks: A fun app for building deeper connections with friends & family.
DocFizz: A radically cheaper Docsend competitor for securely sharing PDFs, pitch decks, clients quotes, etc.
The coming soon page for Brooklyn Uprising (built with AI using v0).
PromptExperimenter: a free tool that I needed for running bulk tests of AI prompts.
ActionJournal (personal only, for now): a journaling tool morning pages and turning it into action items with AI nudges/reminders and (soon) the ability to actually complete tasks for me.
It’s pretty incredible to write it all down and consider the timeline.
Some observations
AI so good at writing code: If you haven’t written code with AI you just can’t understand. Whatever else you think AI is good at, it is much much better at writing code than almost anything else (transcription and translation might be exceptions where it’s better than coding).
A v1 in hours, not weeks: AI compresses the time from idea to prototype so much. For simple things you can just dictate a 6-paragraph prompt, let it rip with code generation, and get a working prototype in an hour or two. There’s absolutely no point in designing mockups any more. It’s just as fast to create and iterate on a working app.
But you can’t just apply speed factor that to the rest of the project. As you start to layer in the details and tweaks that customers expect from a working product, the leverage of AI starts to decrease a lot. You have to more specific in your commands, and keep its focus and context tighter. So the speed gains start to decrease the further you get from the starting line (though they probably level off still at 200-300% more efficient than pre-AI)
Human oversight 100% required: As good as it is at coding, AI can make mistakes. Big mistakes. Mistakes that range from radically misinterpreting the intended user actions to catastrophically deleting large swathes of code for no reason.
It’s more worth it than ever to learn to code: I can see how some might look at this and think that they shouldn’t bother to learn to code. It’s already pretty good at coding, won’t it quickly get so good that anything I learn now will be obsolete? I really don’t think so.
First, for the next several years there will be a huge advantage to those who know how to code with AI. Building personal software custom-tailored to your individual needs (like Action Journal is quickly becoming for me) is incredible. A one-person SaaS as a side hustle, internal tools to supercharge your team… these things are all just as valuable, but the cost in terms of time and bandwidth has just dropped by 10x.
Second, it’s fun! If you’re at all interested in building fun things that people on the internet can use, AI-assisted coding has lowered the time cost required to make that a reality. My first app in a decade, Sparks, is a great example. It’s just a fun little app that I wanted to exist in the world. It wouldn’t have been worth it to me to spend weeks building, but with AI assistance I built it in a day.
That’s it for today. Coming in the future updates:
The tool stack for speed-running a new B2B SaaS
An update on lease negotiations for Brooklyn Uprising
A sneak peek of the AI features I’m exploring with Action Journal