Where The Pundits Are Getting It Wrong – We Still Need Talented Software Engineers

There’s a narrative gaining steam in the AI discourse that goes something like this: coding assistants are getting so good that we won’t need software engineers anymore. Some version of “AI writes code now, so why would you hire humans to do it?”

This take is wrong. Not slightly wrong — fundamentally wrong. And it’s wrong because it confuses three very different things.

Writing Code, Building Products, and Building Companies

Coding assistants write code. That’s a real and valuable capability. They can generate functions, debug errors, scaffold entire applications, and produce boilerplate at superhuman speed. No argument.

But writing code is not the same as building a software product. And building a software product is not the same as building a software company.

Here’s a simple example. A coding assistant can generate a perfectly functional login page in 30 seconds. But it can’t tell you whether your product should have a login page, or whether a passwordless flow would reduce friction and improve activation rates by 15%. It can’t look at your user data and realize that the reason people are churning isn’t a code problem — it’s that the onboarding sequence makes no sense. It can’t decide to kill a feature that engineering spent three months building because the market signal says it doesn’t matter.

Here’s another. A coding assistant can build you a beautifully structured API. But it can’t decide whether that API should be open or closed based on your partnership strategy. It can’t weigh the trade-off between shipping a feature that one enterprise customer is screaming for versus building platform infrastructure that will serve 50 customers in two years. It can’t sense that the architectural decision you’re about to make will create technical debt that cripples your team 18 months from now.

Those decisions require judgment. Taste. Business context. Experience. Pattern recognition built over years of shipping real products to real users and watching what happens. AI doesn’t have any of that. Not yet. Maybe not ever — at least not in the way that matters.

Zero Business Sense

Kamil Banc wrote a great piece on this exact topic — the role of judgment and architecture in AI-assisted business decisions. The core insight is that AI is exceptional at execution within constraints but terrible at setting the constraints. It can optimize within a framework. It can’t tell you whether the framework is the right one.

Coding assistants have zero business sense. They don’t understand your market, your customers, your competitive position, your team dynamics, or the messy reality of shipping software in an organization where engineering, product, design, sales, and support all have to work together. They produce code. They don’t produce strategy.

The Productivity Gap Tells the Story

I’ve written in the past about the productivity gains we’re seeing at Markup AI from AI coding assistants. The numbers are real and dramatic — but they’re not evenly distributed.

Our junior and mid-level developers are getting 25-50% more productive. That’s significant. But our top-performing, deeply experienced full-stack engineers? 5-10x more productive.

That gap is the whole story. The 5-10x number only happens when you start with an exceptionally talented and deeply experienced engineer who knows what to build, why to build it, how to architect it for scale, and when to tell the AI it’s heading in the wrong direction. That person uses the coding assistant as a force multiplier for their judgment and experience. The assistant handles the mechanical work while the engineer focuses on the decisions that actually matter.

A mediocre engineer with a great coding assistant produces more mediocre code, faster. A brilliant engineer with a great coding assistant produces great software at a pace that was previously impossible. The tool amplifies what’s already there.

Fewer Engineers, But Not TOO Fewer Engineers

So yes — the world will have fewer software engineers. The economics are clear. When your best people are 5-10x more productive, you simply don’t need as many humans to produce the same output.

But the pundits who think we’re heading toward a world with 10% of today’s engineers are getting it wrong. What we’re heading toward is a world where the kind of engineer who thrives looks different. Less “person who can write a clean for-loop” and more “person who can architect a system, understand the business context, exercise judgment about trade-offs, and use AI as a tool to execute at extraordinary speed.”The demand for that person — the engineer with taste, judgment, and deep experience — is going to go up, not down. They’re the ones who will build the products and companies that matter. The coding assistant will write their code for them. But it will never replace the thing that makes them irreplaceable. And the only way they’re going to get qualified is by starting as a junior developer, get trained, and learn the ins and outs, which doesn’t work if you decide to outsource all your entry-level engineering work to a coding assistant.