If you write software for a living, you’ve probably felt it.

AI is making us faster.

A feature that used to take a day or two now takes an afternoon.

A bug that used to take half an hour to locate now takes seconds once you hand the logs to the model.

Spinning up a project from nothing used to take a week, but now, with the right tools, it runs in two or three days.

Here’s what nags at me, though:

If everyone becomes a 10x engineer, and demand for software doesn’t also jump 10x, then right now, today, don’t we only need a tenth of the people?

Well, it sounds extreme. But think it through, and the logic definitely holds.

Go read Hacker News, Reddit, the tech side of X, or maybe LinkedIn.

It’s wall-to-wall anxiety.

People are asking whether software is still worth getting into. People are sharing that they applied to forty companies and heard nothing back. People are all seriously talking about switching careers.

The numbers are also inclining towards that direction. Per layoffs.fyi, tech cut around 165,000 jobs in 2022, then a brutal 264,000 in 2023, about 153,000 in 2024, and roughly 124,000 in 2025. That’s more than 700,000 people since 2022.

The reasons behind those numbers I’ll get to. But whatever the cause, this is the current sentiment. A lot of engineers have quietly decided job security is a thing of the past.

In that mood, one idea keeps getting raised: if a job isn’t safe, build your own product.

Quick definition, in case indie SaaS is new to you.

One person, or a very small team, builds and runs a software product and charges a subscription. No funding and hiring. One person carrying a small business.

And plenty of people will tell you now is the best time ever to do it. AI tools let one person do what used to take a team. Cursor, Lovable, and Replit agent make you fast. In Y Combinator’s Winter 2025 batch, a quarter of the startups had codebases that were 95% AI-generated. A simple SaaS MVP that not long ago cost tens of thousands now runs for a few thousand, by many builders’ accounts. One person, one laptop, a few AI tools, an idea to launch.

I mean, this sounds great, right?

The brutal reality I see is different.

The big model companies ship features faster than any indie or small team can keep up with.

Every time OpenAI, Google, or Anthropic pushes an update, a batch of startups gets caught in the blast. When OpenAI shipped its Agent product, the running joke in these circles was: “There goes another batch.”

Here’s the situation. A team builds an AI debugging tool and grinds their way to tens of thousands of paying users. Then OpenAI (or Anthropic or Google) folds all these into ChatGPT Plus, same $20 a month, except now it also writes, searches, and generates images. A few months later, the startup shuts down.

This kept happening through 2024 and 2025. The logic behind it is simple: if your product is basically “a big model plus a custom prompt plus a nice UI,” you’re one platform update away from dead.

People who already have users and distribution may be able to ride it out.

But starting from zero now?

Unless you’ve got a real edge in some deep vertical: proprietary data, hard-won industry knowledge, a regulatory moat. Otherwise, someone can clone what you built in six weeks.

I used to daydream about building a SaaS myself (a few years ago). Honestly, I never got anywhere near product-market fit. Every time I had an idea, one search showed me five people who’d already built it. The furthest I ever got was collecting a few emails. Then nothing. I’m sure some of you reading this can resonate.

Looking back, quitting those was probably the right call.

Quick introduction, since we’re here.

I’m a full-time software engineer, working in Singapore, been writing code for years. I taught myself to program in 2019 and landed my first engineering job five months later. Still at it.

Besides the day-to-day, I’ve got a role on my team called AI Champion. Sounds cool. It mostly means helping the team figure out and adopt AI tools: the company hands me licenses to try things, I sit in the relevant meetings, gather people’s feedback and pass it up. Colleagues bring me their AI questions. I’ve run a sharing session. And I talk with our dev manager about how to build a culture where people actually use this stuff.

That role deserves its own piece someday. Not today.

Doing that role made me notice something. We are, measurably, spending less time writing code.

It used to be more than half my day. Now a lot of what we do is let the AI write first, then review, judge, adjust. The rest goes to talking: pinning down requirements with product, coordinating with other teams, meetings, arguing about architecture.

The work hasn’t shrunk. We’re still busy (like very busy). But the nature of it changed.

Which got me stuck on a question: when AI can reach into every corner of building software, what are we here for?

You could say solving problems still needs people. Someone has to understand the business, define what’s needed, make the calls, coordinate everyone. AI can’t do that.

Right. True.

Here’s a take that might not be the popular one.

In 2021 and 2022, tech went on a hiring bender. The pandemic accelerated everything digital, and every company decided it needed more engineers. Meta went from about 58,600 employees at the end of 2020 to a peak of roughly 86,500 two years later. Almost every big shop was scrambling to hire.

Then the bubble popped. Mass layoffs started in the back half of 2022, and 2023 was the worst of it.

A lot of companies mentioned AI in their layoff notices: AI made us more efficient, so we need fewer people. Honestly, I think AI mostly got handed the blame. The real reason was they’d overhired, bet too optimistically, and then interest rates climbed, and they had to pull back. AI just gave them a nicer line to use.

Excuse or not, the outcome’s the same. An engineer plus AI, at today’s level of demand, means you don’t need as many hands.

Demand might come back. Historically, every jump in productivity eventually creates new demand and new jobs. AI might go the same way; once people adjust to the new normal, new needs show up on their own.

But AI is also improving fast. So whether demand catches up or supply keeps running ahead, nobody knows yet.

My conclusion is that software engineering isn’t going away, but the job will look very different and will probably need fewer people than it does now.

At this point, you might expect some career-pivot framework.

Nope.

My thinking is simple. I decided to start learning robotics from scratch.

Why robotics? A few reasons.

I’ve followed the field for a long time, just never sat down and learned it from the ground up. It was more the watch-the-news, scroll-the-videos level of interest. This time I want to get my hands dirty.

I also think embodied intelligence is the next big problem after AI. However strong today’s AI gets, it still lives inside a screen. Getting AI to step into the physical world and deal with real environments is far harder than writing code. We’ve got plenty of AI that can write software. AI that can reliably assemble parts in a factory, or fold your laundry at home? Nowhere close.

And the field might be near an inflection point, the kind AI hit after GPT-3, when the whole thing suddenly sped up. At GTC 2026, NVIDIA came right out and said the big bang of physical AI had begun.

We can obviously see how the money flows, too. In 2025, more than $34 billion of private capital went into robotics companies, more than double the year before. Goldman Sachs has raised its humanoid-robot market forecast to $38 billion by 2035. One forecast has China’s humanoid market alone compounding at 47% a year through 2035. Money doesn’t flow like that for no reason, right?

The last reason is more personal.

I really want to see smart home robots become normal in my lifetime. The general-purpose kind that can do chores, help care for the elderly, and handle the daily grind. Watching it happen won’t be enough for me. I want to help build it. I hope I will have one when I am old and single (crying in happiness).

In 2019, I taught myself to code and had a software job five months later.

Back then, I knew nothing. Couldn’t tell you the difference between Java and JavaScript. I watched tutorials, built projects, interviewed, got rejected, interviewed again, and somehow ended up here.

It’s 2026. I’ve decided to do it again.

Right now I’m working through Ian Juby’s Robotics, Electronics and Electrical course on the company’s Udemy Business, starting from the basics of electricity. Yes — a guy who’s written code for years, back to Ohm’s law.

Progress is slow. But that “I know nothing” feeling? Last time I felt it was 2019, first time I opened a code editor. I’ve kind of missed it.

This is Robot Economy.

What I’m doing here is simple: figure out the business of robots, and take it apart as I learn it. A software engineer, learning the hardware side from zero, thinking out loud in public. Companies, technology, the money, the talent, where all of this is going. That’s the beat. I read both English and Chinese sources, which helps, because a lot of the real action in robots right now is happening in China, and the English coverage of it tends to run thin.

I don’t come to this as an authority. I come to it as someone doing the homework in front of you, willing to say out loud when I don’t get something yet. If that sounds like your kind of thing — watching an engineer teach himself an industry, calling what’s real versus what’s just a good demo — do come along.

That’s it. See you next week, hopefully?