Digital Landfills

Do yourself a favour, don't turn around.

A key part of my childhood was the annual summer trip to Rajasthan. I was born and raised in Bangalore, but my parents made sure to preserve the connection to our roots. The fact that temperatures routinely crossed 45°C was irrelevant. Wedding, funerals and summers were all consecrated in Rajasthan.

This was pre-Air Deccan India, and air travel was prohibitively expensive. Especially when one did headcounts by the dozen. That meant a 2000km cross country train journey with multiple transfers and many, many hours spent on railway platforms, surrounded by mountains of our own luggage.

We’d carry home-cooked food to last the journey. Vendors made sure chai and “kol-drinks” were free flowing at all hours of the day. Uncles, known and unknown, shared insider information about which stations had the best hot snacks¹.

But the central edifice and source of sustenance in this epic journey was our Milton water can.

Notice the tasteful beige. Source.

Listen up, kids. When the train would stop at random railway stations, we’d lug this thing to the communal taps, and refill it. With water of unknown provenance. And that was our only supply. What happened if we ran out, or didn’t plan our refills to last the night? When all else failed, we could count on the kindness of strangers.

Unthinkable today, isn’t it? But that’s a whole other essay.

---

Much has been made of Claude Code, vibe coding and the Saaspocalypse². I’m not going to rehash the whole thing, so feel free to listen to my rants on a dedicated podcast episode here³. TL;DW: AI is here, SaaS is in mortal danger and Bangalore real estate is screwed. But not just yet.

Today, I want to look past the immediate manic-depressive swerves of the market. What might such a sea change actually look like?

In the current paradigm, if you have a niche but repetitive digital task with a finite number of variations, you can commission (or build) a custom software tool for it⁴. This mechanism has several constraints. Building working, error-free code is expensive. It gets costlier and more entangled as you try to cover every edge case.

This sunk cost is justified because of the near infinite reusability of code. Even a marginal saving (in time, money or both) per use has excellent return on investment because of the sheer number of times you can reuse the tool. Those pennies add up.

The flip side is that it doesn’t make much sense to build code that isn’t used often. The returns simply don’t justify the investment. In many cases, it’s cheaper to do it by hand, using a patchwork of tools.

That’s what Claude Code (and vibe-coding in general) changes.

If building a software tool is nearly free and needs no training, why not build a niche, custom tool that only you need, and even then, not so often? In fact, why not build a tool even if you only need it once? Vibe coding requires no programming knowledge, no outside help. No decades of experience in the nuances of efficient resource utilisation⁵, or the principles of software design.

That, I believe, is the natural and inevitable direction in which we’re headed: disposable, non-recyclable software. Like the bottled water that replaced our trusty Milton, it will have far reaching implications. Let’s try and map out a few.

Right off the bat, more people using software for more tasks has always led to immense productivity gains. That spreadsheet task you spend 2 hours on every week/month/quarter, now takes all of 3 clicks and a quick and simple python script⁶.

Second, if this tool is so powerful in the hands of an amateur like me, imagine what a truly qualified and experienced engineer can achieve. Everyone’s talking about IT layoffs, but I can’t wait to see the tsunami of startups that will be unleashed, doing incredible things with tiny teams and the power of AI.

Now, for the flip side.

Most AI advances so far have been followed immediately by the proliferation of slop. I don’t think this wave will be any different. Like the billions of discarded water bottles in landfills, most new tools written with AI will be junked far too soon to justify their energy footprint. In fact, most tools available off the shelf too will land somewhere on the slop spectrum, with averages moving towards more sloppiness over time.

For the engineers, AI coding represents a genuine chance of burnout- as they try and do too much, too fast and all by themselves. Especially when more and more of their job becomes policing and checking for logical/framework level errors and edge cases.

For the rest of us, once the novelty factor wears off, realisation will set in that we haven’t really changed the world or become 100x more efficient. All we’ve built are a bunch of inefficient, incomprehensible, unscalable one-off tools that give us the illusion of productivity gains without allowing for institutional learning or compounding of knowledge. And that’s if our projects actually make it out of production hell.

For a layperson, vibe-coding is dangerous because it enforces none of the guardrails of good programming: Documentation, clear problem definition, version control, data security, resource utilisation, and on and on. Worst of all, while the LLM is constantly getting better at execution, the user has no such incentives, getting dumber if anything.

Like in life, in software development too one can take shortcuts, choosing short term gains that have hidden long term costs. This is called technical debt. As vibe-coding adoption increases, the technical debt being accrued will increase exponentially.

Further, vibe-coding discourages collaboration and collective innovation. It takes from the environment (open-source) but gives nothing back. Not to mention the energy/water costs and carbon footprint of AI.

Can we prevent this impending digital garbage avalanche?

No amount of my singing the virtues of Milton or the dangers of using plastic water bottles is going to change habits or people’s minds. The transition was one way, and it’s going to be the same here.

Even if there is a resurgence of people carrying their own hydration flasks, the public infrastructure to ensure potable water at a random railway station at 3 AM is broken. As is our collective trust in drinking unlabelled, unsealed water from an unknown source.

No, once again, I think The Simpsons called it⁷.

We’ll just pack up and move town once we’re done here.

1. I wonder if they’ve moved on to stock tips now.

2. Terrible name.

3. And while you’re there, consider subscribing?

4. I’m deliberately leaving out Open-Source tools from this discussion, since discoverability, lack of support and lack of customisation remain large hurdles for lay users.

5. We all have access to huge amounts of storage and computing power on our personal devices anyway.

6. Whatever task that made you think of, go now and try the free or base tier of any LLM to automate it. You will be blown away. I know I was.

7. The PSA being referenced.