Make humans analog again
AI isn’t replacing humans - it’s pushing us to be more human. Here’s how agents can unlock creativity, movement, and better ways of working.
Oh, how the turns have tabled!
You’ve seen gloomy narratives in the tech media about AI making everything worse for humans. But I believe it has the power to unleash our humanity, not diminish it.
It requires us to flip our habits, change our culture, and reverse our heuristics. We’ll be rewarded with improved physical health, better communication, and boundless creativity. These are the biggest reversals we're experiencing in this new era:
Software is not a desk job
This is my favorite reversal - I don’t need to sit at a desk anymore! I can code while on the move, and I LOVE to move. I built the majority of bay.dance by texting/speaking to my agent in English - initially on replit, and now using a combo of Claude Code and Happy. I contributed way more features to it this way than I ever did sitting at a desk. And it's used by hundreds of real dancers!
Software is now a couch job, walk job, park job, train job, bus job, studio job, toilet job, maybe even a desk job - it’s wherever you want it to be.
Make humans analog again
At my job, I drew a flowchart in my notebook and sent a photo of it to Claude Code. It digitized my hand-drawing as an excalidraw diagram that I could tweak and plop straight into my slide deck!
Excited by the possibilities, I next drew an architectural block diagram for refactoring out existing spaghetti code into a separate library, and fed that to Claude. It created a stack of 5 high-quality commits realizing the design, complete with unit tests, telemetry, and descriptive commit messages! This is not a throwaway demo, like most vibe-coded examples you see on the web/social media; this is real code running in real prod for a real 1M+ user base (sadly, I can't share this diagram/code because it belongs to my company).
We can move and create IRL with our hands and voices, just like we evolved to.
The todo-list is just a do-list
Take the “to” out of “todo” and the “back” out of “backlog”. The agent can just get started on your idea right away! Why let your idea wither and die in a todo list when an agent could implement it in the background and ping you when it’s done? We can speak ideas into existence now.
Btw, I am so excited by this idea that I started building a prototype of do-list. Stay tuned!
The “proper” way is the fastest way
“Be scrappy” used to be the best way to move fast. Startups were advised to implement it as messy as possible, just make it work and get it out there ASAP.
I still see engineers in my company lean on this old heuristic, hacking in spaghetti code and copy-pasta because “there isn’t enough time to do it properly”…only to get tangled and stuck 1 month later in their own tech debt.
The truth is that text is cheap now, so refactoring, docs and testing are cheap too! Quality is crucial for AI to receive self-correcting feedback as agents swarm all over your codebase. Give agents a framework, documentation, and test driven development (TDD) - these “overhead” processes will actually enable them to go faster, not slower.
Everyone is a manager now
If you read articles about The Great Flattening currently taking place in tech companies, you’d be forgiven for thinking that the job of management is toast. But the flattening isn’t killing management, it’s just spreading out those responsibilities to every engineer (and AI agent) in the company.
Think about the skills you need as a software engineer now - writing prompts, delegating to agents, orchestrating multiple agents, managing context, and unblocking your agent and yourself. These verbs are all managerial skills! And they are transferable from managing agents to managing humans.
I’ve found that my skills of human communication, coordination, and management have grown immensely in the past 6 months, despite always (and still) being an engineer. I think most of this growth came from having a safe space to practice these skills with my nonjudgmental agents first.
AI isn’t
There’s a fantastic novel named Ra by qntm (a dope pseudonym for an author), in which humanity finds out that “magic” is real. But in this world, the most popular phrase among mages is “magic isn’t”, signifying that magic isn’t actually magical. They take a scientific approach to understanding magic, breaking it down and expanding upon it from first principles, just like our scientists do with physics.
In a similar vein, I propose the phrase “AI isn’t,” because LLMs ≠ intelligence. Real power in this era will belong to those who understand the limits of LLMs - who see the jagged edges, fill the gaps with human intelligence, and recognize that this sufficiently advanced technology is distinguishable from magic.
Thank you to these wonderful humans for reading drafts and providing feedback: Dani, Maya, Mehika, Bhavit, and Mohsin.