Will AI create or kill software engineering jobs? What the data says

By Sophie Novati

Let's get straight to it – is AI going to create or kill SWE jobs?

Ah, the question of the moment. As the founder of Formation, the median time it takes for any conversation to arrive at this question is probably about 5 minutes – faster if wine is involved.

The Internet loves to declare that software engineering is dead, and that we'll vibe code our way to a utopian future by 2026. Clearly, nuance doesn't get clicks.

But today, I'd like to invite you to take a breath with me and slow down: reality isn't binary, and neither is the future of software engineering. 

Let's unpack some of the most sensationalized headlines and trends floating around social media and explore a more balanced perspective on each topic. Buckle up.

Myth #1. The majority of code is already being written by AI

We’ve all heard it: “20% of our code is AI-written!” brags one CEO. Another fires back, “Oh yeah? Well, ours is 30%!” In the latest YC batch, a quarter of companies declared, "We're practically robots already – 95% AI-generated code!" 

But let's be honest. Counting percent of code written by AI is about as meaningful as measuring productivity in "lines of code", or, as Bill Gates famously said, "like measuring aircraft building progress by weight.”

The phrase “written by AI” covers everything from autocomplete to full-blown agent-mode vibe coding, but the latter is rare. In Jetbrain's Developer Ecosystem Survey, 77% of engineers using AI report saving only 4 hours or less per week (this is from 2024, so it's likely improved since then).

This is because most engineers still heavily guide the engineering process, sometimes pinpointing the exact file or function and giving clear instructions, which explains why two-thirds (67%) of AI-produced code is rejected by the overseeing engineer, as seen in this Github Copilot study conducted by ZoomInfo.

Additionally, we are still very much in the era of “humans are fully responsible”. Engineers are ultimately responsible for the code they check in. So, many AI suggestions still require deep review and often edits or rewrites by humans, which can sometimes take almost as much time and thought as writing it yourself.

So, while AI may sometimes sound like a genie-in-the-bottle, in practice, it is often just a helpful boost rather than a fully autonomous handoff, and so I always have one eyebrow raised whenever there's a claim that any codebases are largely autonomously written by AI.

Myth #2. Vibe coding is the future

Now, I don't actually disagree with the premise of this claim. AI is certainly going to become better at assisting engineers, and there will be more and more situations where full agent mode will be feasible and produce good results.

What I disagree with is the speed at which this change will occur. Yes, the idea of AI agents helping us write entire applications autonomously is exciting. And yes, small startups and scrappy teams are already experimenting with this. But if you’ve tried to apply the same methods to a 15-year-old enterprise codebase… you've experienced the limits. In Stack Overflow’s 2024 Developer Survey, only 3% of engineers said AI tools handle complex tasks well. 

Another problem is that the set of skills required to vibe code don't intersect much at all with the ones you need to debug if absolutely anything goes wrong. As an engineer, the statement "we're good as long as nothing breaks" is a scary one. If we truly never needed to debug, then vibe coding should just produce machine code directly without the hassle of the in between human readable code layer.

People needed to understand pointers long after higher-level language abstracted away the need to manage it for the majority of cases. But, until AI becomes reliable enough to never produce bugs, engineers will still need to break out of vibe mode to drop into the code frequently for the foreseeable future.

Myth #3 Junior software engineers are no longer needed

The shift towards AI has further exacerbated the trend of shifting towards more senior engineers. This started during the pandemic, when tech companies overhired, panicked, and then slashed roles. In the aftermath, job postings for junior engineers dropped by more than 50%, according to SignalFire's 2025 State of Tech Talent Report. 

At first blush, this makes sense. AI can handle the well-scoped, repetitive tasks junior engineers typically start with. But a lot of folks are missing another opportunity with AI: it doesn’t just replace junior dev work, it makes onboarding and training them way easier.

Many of us have been there – trying to mentor someone new, only to realize it’s faster to do it yourself. But now, we can have an AI mentor that’s infinitely patient, always available, and never says “I’ll get back to you.” That changes the game for rising talent.

With a growing shortage of experienced engineers on the horizon, we simply can’t afford to ignore the junior pipeline. AI gives us the perfect tool to nurture it, if we’re willing to rethink how we train and support early-career talent.

(At Formation, we’ve leaned into this. Pairing human mentorship with AI-guided practice has helped our Fellows land top-tier jobs with an average compensation lift of over $100K. Turns out, investing in talent pays off.)

Myth #4 You don't need to learn to code anymore

Nvidia’s CEO recently said kids don’t need to learn to code. Others like Mark Zuckerberg have echoed that AI will soon “replace mid-level engineers”. 

I get the sentiment. But, here's my take.

Coding isn’t about syntax. It’s about structuring logic, thinking in systems, and solving problems in a way that scales. As Microsoft’s CPO Anna Chennapragada said on Lenny's Podcast: AI is just a new, higher-level abstraction layer, just as high-level programming languages were to Assembly. And just like before, we’ll need to "drop down to the code layer" and debug under the hood for years to come.

In addition to making debugging possible, learning to code is also an important foundational skill on top of which to build higher-level engineering skills.  Think about math as a primitive skill. Very few people learn math to get a job as a mathematician. Math is usually a component skill serving as a building block to other professional skills like accounting, architecture, or data analysis, and skipping basic arithmetic makes it very hard to have a solid footing to develop those higher-level skills.

Even when engineering requires minimal writing of code as part of the role, the ability to read and understand code will still be very relevant even in fully agentic workflows. Lacking the ability to read code would be like trying to be an architect without understanding materials science or structural engineering – you might be able to sketch something that looks like a building, but it may collapse under its own weight in real-world conditions. As an engineer, you are responsible for what you check in. This means guiding the AI, verifying its outputs, and iterating on your instructions as needed to produce the end result.

So yes, learn to code! Build that mental muscle, sharpen your logical reasoning, and deepen your problem solving abilities. As Bill Gates once famously said, "Learning to write programs stretches your mind, and helps you think better, creates a way of thinking about things that I think is helpful in all domains."

Myth #5 AI making engineers more efficient means we need fewer engineers in the future

AI takes away the tedious portions of the engineering role, allowing engineers to focus on the higher-level work and therefore output more than they have ever been able to before. So, fewer engineers will be needed in the future, right? Well, that's one way to look at it. But, just because we get more productive in a field does not necessarily mean lower demand in that field.

Did you know that Notre Dame took 182 years to complete? Using modern technology, materials, techniques, and planning is estimated to take only ~10 years (a dramatic 95% improvement). Yet, the percentage of people "working in construction" has dramatically increased, not decreased, almost 100-fold in this time period due to the dramatic shift in the global population towards urban areas.

This trend has also played out in the field of engineering as well. While high-level programming languages like C or Java was arguably an order of magnitude efficiency gain on Assembly, the number of software engineers tripled following the rise of high-level languages according to data from the U.S. Labor Bureau, due to the massive shift in the world that has seen almost every industry being transformed by software and the internet over the last couple of decades.

Whether a profession grows or shrinks is more related to the general trends and shifts across the world and less about how efficient we are able to make that role. As I see it, the world is on an unstoppable momentum to have technology and software permeate bigger parts of more people's lives over time, meaning we need more people, not fewer, to be "software engineers."

That doesn't mean that software engineering as a role doesn't change. It will change, both in the types of work involved and the diversity in the types of roles available. But that is similar to the way that a construction worker today does very different work from a construction worker in the year 1200.

Myth #6 All jobs will disappear

Ok, we saved the best for last: the doomsday headline that's the foundation of all AI fear spirals: “Eventually, there will be no jobs at all.”

Let’s entertain that thought. Imagine a world where every need — food, shelter, health care — is met by hyper-efficient systems. What, then, does a human do?

First of all, we've been here before. At the start of the 20th century, about 38% of the U.S. labor force toiled on farms. Why? Because they needed to in order to produce the food we needed to literally survive. Today, that 38% has shrunk to a mere 1.2% of the labor market, and yet we're not all unemployed. We’ve filled that space with entirely new categories of work — many of which were unimaginable to our ancestors.

For a long time now, jobs have not been solely about survival. No one needs to become a painter, a pastry chef, or a professional gamer. And yet these roles exist (and deeply enrich our lives). The more our survival needs are met by automation, the more we'll see our job market climbing higher in Maslow’s hierarchy: fewer survival-critical tasks, more roles centered on curiosity, connection, and purpose. 

For engineers, designers, educators — and yes, even pastry chefs and pro-gamers — that’s an invitation to create richer and more fulfilling work, not to fear it.

So… is AI going to kill software engineering?

Not quite. But it is going to change radically, in ways that will challenge our assumptions, stretch our systems, and ultimately redefine what it means to be an engineer.

The engineers of tomorrow won’t just write code. They’ll orchestrate systems, mentor machines, and design at higher levels of abstraction. The best engineers won’t be the ones who resist change. They’ll be the ones who learn faster because of it.

So buckle up, experiment boldly, and remember: humans still write the prompts.

P.S. Want the cheat sheet? Check out our latest blog on how to use AI-boom to level up your career.  And, if you're looking for hands-on mentorship to incorporate AI into your day-to-day workflow, enroll in our upcoming Ship With AI course!