The Noise

With all the hype around using AI in software development over the past year, I felt it necessary to get into the thick of the action myself to see what the fuss was all about. There was just so much noise about “vibecoding”, “agents” and a whole slew of new buzzwords.

Between October 2025 and March 2026, I began tinkering with AI tools and saw how AI workflows operated in the wild in my consulting work. It was really intriguing how everyone was figuring things out as they went along. Everyone was essentially starting almost from scratch, learning the new CLI tools, how to review a ton more PRs, shipping code (and bugs) at breakneck speeds, etc.

There was definitely a new energy reverberating across industries, not just in tech. As a quiet attendee lurking in various AI enthusiast events, I saw technologists, business owners, marketers, developers, CEOs, CTOs and funemployed tinkerers chatting animatedly about the benefits of AI in their day-to-day work. The excitement was palpable as people envisioned a new future with their machine overlords. Most were trying out new tools (which seemed to be spawning non-stop daily), seeking advice on how to optimise their OpenClaw or finding ways to tokenmax.

But what struck me wasn’t just the excitement — it was how different the work itself started to feel.

A New Kind of Engineer

Now, I haven’t been coding for a while, given that I transitioned to becoming a product manager early in my career, but I’ve always been rather hands-on and technical so I had to give agentic tools a shot to see what the fuss was about.

I was definitely impressed and they wrote code at a speed that blew my mind. I found that I was “managing” and “directing” these agents to do my bidding, which was quite a different experience. It was also a different way of technical thinking that exercised a different part of my brain.

I had to now think in systems, architectures and use cases while the agents built. Not only that, it felt like a game, and, dare I say it, fun.

But speed and power came with trade-offs.

Speed, Slop, and Judgment

I began going down a rabbit hole of reading blog posts from developers I really respect who were also exploring the new world of AI and found their experiences illuminating. Code slop is a thing and we should definitely be critical of the tools we use, neither being too overzealous about them nor being utter luddites.

Tools are temporary but the abstraction is here to stay. True mastery of this new way of thinking and working will require us to slow down, so as to ensure quality.

If this is how the work itself is changing, then it raises a bigger question.

The One Person Engineering Team

With these principles in mind, I did feel that we are on the cusp of something cool. Most companies currently employ teams of engineers and are struggling to migrate them to this new paradigm. If I had the privilege of starting from scratch, how would I reimagine the organisational structure of the company of today, especially on the engineering front?

If frontier labs are pushing for the one person billion-dollar company, can we then assume that the one person engineering team also exists? This logic has to follow, right?

What I’m beginning to realise is that this isn’t about one person doing everything manually. It’s about one person orchestrating a system that can do the work of many.

The role of the engineer starts to shift:

  • from writing every line of code to defining intent clearly;
  • from implementing features to designing systems that can be extended and modified safely; and
  • from debugging code to debugging workflows, prompts and abstractions

In this model, agents become leverage. They generate code, tests and even entire features, but they require direction, constraints and taste. The bottleneck is no longer typing speed or even raw technical knowledge; it is judgment.

This also changes how we think about traditional team functions. QA, DevOps, even parts of product thinking start to collapse into a single feedback loop. Instead of handoffs between roles, you have tighter cycles of: specify → generate → review → refine.

The cost of coordination drops dramatically, and with it, the need for larger teams whose primary job was to manage that coordination.

Of course, this does not mean that teams disappear entirely. There are still limits scale, domain complexity and organisational context all matter. But for a large class of problems, especially in the early stages of building products or transforming operations, the default assumption that you need a full engineering team starts to feel outdated.

Thankfully, I’m not that brilliant an engineer and am more or less proficient in Ruby (and love it). I’m grateful that we already have the One Person Framework. The stack is boring and token efficient (I hear). So my engineering choices are more or less sorted.

A Real-World Experiment

I’m proud to say that the above principles and this boring stack helped me get my current role as the new CTO at Presto Drycleaners, a traditional, family-run drycleaning business that still runs on WhatsApp messages and paper records.

I’m their first technical hire and today is my first day of work. No legacy codebase, no inherited architecture decisions, no team to onboard. Just me, a Rails app, and a stack of dirty laundry.

The hypothesis I’m testing is simple: can one person, armed with the right tools and the right mindset, do what previously required an engineering team? Not in a startup with runway and engineering culture baked in, but in a real SME, one with non-technical stakeholders, tight budgets, and zero tolerance for downtime on the ironing machine.

I don’t know if it’ll work. The agents might produce slop I can’t catch. The abstractions might leak at the worst moment. I might find that some things genuinely still need more than one brain. But that’s what makes it an experiment worth running.

I’ll be documenting it here, the stack decisions, the mistakes, the moments of magic, and the moments of pain. If you’re curious about what AI-native engineering actually looks like in the wild, outside of a VC-funded tech company, follow along.