Software Engineering Isn't Dead: The Bright Future of Developers

Lovable lasts three years: code superabundance and permissionless creation

“It’s over. It’s so over.”

As I yet again scrolled through my X feed to the tune of the vibecoders’ horde announcing the demise of tech as we know it and chanting the opportunity of making “$10k/month overnight without writing a single line of code - and buy my course to learn how to”, I proverbially screamed at my screen: ENOUGH!

Software engineering isn't dead; it's evolving.

In today’s startup ecosystem - part hype cycle, part existential crisis, part meme economy, part WTF is Trump doing - products like Lovable, Bolt, Replit, Cursor or Windsurf are populating the GenAI landscape and grab attention because of the insane levels of ARR they have been able to reach in a very short time (Lovable went from 0 to $10M ARR in 2 months). And yes, that is very impressive.

However, behind those storylines - I think it’s the deeper, strategic shift that deserves our full attention: what GenAI means for coding and the tech ecosystem in general.

Windsurfin’: a world of permissionless creation and permissionless distribution

If everybody had Notion across the USA, then everybody be vibecoding like Californ.ai / You’d see ‘em prompting their ideas and cloning repos too: a pushy, pushy software though. Windsurfin’ USA

Discarded draft, Beach Boys

GenAI has democratized software creation. Today, anyone equipped with a few prompts, ChatGPT and Lovable/Bolt can conjure up functioning software within minutes. To be clear: this is a profoundly positive shift, and the natural evolution of something that started decades ago with the advent of the Internet and has matured through years of Google, Wikipedia, YouTube, Coursera and TikTok: facilitated access to creation. To borrow an idea that’s dear to my friend Hugo Amsellem, it’s permissionless creation.

In the past decade, launching a fully functional business with an e-commerce website in a day through Wix, Shopify and Stripe became the new normal. These days, building an app or a SaaS can be done in under a day through Gen AI tools. This is the new normal.

A revolution? Maybe not, arguably the continuation of the same permissionless creation movement, nibbling on what seemed to be not gatekept, but at least only accessible to those who went through the knowledge and learning process of becoming a coder.

What GenAI allows - or seems to allow (more on that later) - is for the knowledge barrier to disappear.

Yet, this new abundance in permissionless creation inevitably demands equally permissionless distribution. Paradoxically, as barriers to software creation vanish, reaching and captivating an audience or clients becomes increasingly expensive. Customer Acquisition Costs (CAC) surge as noise multiplies and competition intensifies.

More competitors buy more ads targeting the same people in the same niche. More creators push more content with even less equipment and time spent producing. Hell, some of them aren’t even real people: they’re AI generated avatars mimicking content with the same positioning that went viral on TikTok (see Arcads, for instance). Even if you’re doing community-led growth, open-source, DevRel through X/LinkedIn and manage a lower CAC, channels are saturated.

We’ve entered an era akin to streaming platforms: endless content is produced, but discovering quality is harder than ever - and producing quality becomes an entirely new process.

So sure, you can expect to see hobbyists flourish and perhaps even marvel at a solo SaaS founder hitting a billion-dollar ARR. Yet these intriguing outliers won't define the industry's future or reshape markets at scale. Mass adoption requires capital, scale, distribution strategies - and yes, I’ll say it: proprietary intellectual property.

Enduring Moats: Proprietary Tech is King

I used to teach business strategy and tech at SciencesPo Paris. Invariably, I’d start introducing the students to the wonderful power and leverage of tech companies through an analysis of the production process:

In the industrial era, ideas need → capital to start → production. Products then need → distribution which is concentrated in a few retail stores to reach → clients.

With the Internet, ideas can be → produced for a few bucks online (provided you know how) and → distributed directly to → an audience through social media, newsletters, etc. Since production is vastly available and cheap to learn ; and distribution can be done directly, capital is no longer a barrier to reach an audience: it only serves as an echo chamber to scale. As a result, more ideas reach potential users, so it’s important to validate market-fit early before even producing or seeking capital.

Yes, yes … we’ve heard that story. Does that change with GenAI? No, it gets even more true as even the knowledge barrier falters.

However, because of the superabundance of permissionless distribution methods (influencers, newsletters, TikTok, LinkedIn - whatever you’re selling there’s a channel to reach your clients) acquisition becomes costly, and capital becomes crucial to scale. And it’s usually a VC’s role to fuel innovative tech ideas with capital.

Sure, you can also create a profitable operation and reinvest your profits in acquisition. In a world of superabundance though, what are the odds that a bigger, meaner, better-funded competitor reaches the market first with the same angle and captures the opportunity?

And what if you can’t consider that the very product that you’ve built is a moat because it was simply prompted into existence, which is replicable by anyone?

Here’s the rub: software built with GenAI tools, despite their disruptive potential, face a significant hurdle when it comes to venture investment: proprietary defensibility.

Serious investors generally don’t finance codebases easily replicable by another engineer with the same generative prompt. Proprietary technology, defensible data sets, and unique approaches are more crucial than ever to secure capital. Yes, cheap money has funded flimsy moats in bull cycles (and boy, is AI a hype theme!), yes some AI-wrappers have managed to raise funds but durable returns are likely to concentrate in proprietary data or infra.

Also, I’m talking about tech companies: software, AI, etc. – a marketing-first play, even fueled by tech, is different in essence. A funny example is OnlyFans ($6.6bn revenue in 2023, paying $1.3bn in dividends to its founder): while the site relies on tech that scales, the real core of the business is in how it partners with creators and helps them to monetize, not the tech itself. Back to our main theme.

I’m not saying parts of the codebase cannot be prompted: the same way some of it can rely on an open-source library, there’s no reason not to leverage this technology when it can be useful.

What I am saying is that ambitious tech projects need proprietary tech supervised by developers. Developers that can be augmented by AI, but definitely real humans that know how to read, write, build, comment, structure and correct code.

Tech teams: more crucial than ever

Let’s be crystal clear: the magic behind Generative AI isn’t in the black box - it’s in the humans who guide it. Knowledge remains the ultimate gatekeeper.

Without developers who understand guardrails, architectures, and legacy quirks, tomorrow’s systems will resemble spaghetti code drenched in AI-generated sauce impossible to untangle in 20 or 30 years. As nations and enterprises increasingly rely on software for defense, healthcare, and critical infrastructure, the demand for engineers with deep, vetted expertise is more important than ever, especially to build, guide and review those AI-generated systems.

This is why rock-star engineers are vaulting into roles that look more like hedge-fund gigs: AGI at OpenAI, rocket code at SpaceX, self-driving stacks at Tesla. Frontier engineers have a critical role to play in our collective future.

And yet, as recent data indicates, software developer job postings are down to 70% of what they were in 2020, and much much lower than peak demand in 2022. Indeed, we’re in a recession period and difficult access to capital + hiring freezes distort the numbers, but still: SWE are less in demand.

However, I see something interesting in it: it’s not software engineering that’s dying, it’s a certain sort of software engineer. What used to take entire squads of developers now gets shipped by a lone “product builder” wielding Copilot in a matter of days; sometimes hours. True tech companies have an output revenue per employee that’s higher than ever.

In 2025, the “middle-class engineer” is no longer a thing: the center has collapsed, the summit is reserved for a hyper-elite few armed with PhD or frontier knowledge, and everyone else is scrambling to be the AI-augmented one-(wo)man-show that can outpace a team. Is that so? Do we just write-off 90% of SWE jobs and call it a day? There’s more to it.

My friend Gilles Barbier has dubbed the mounting sense of “AI fatigue”: a weariness born not only of constant updates, endless model versions, and the ever-present fear of falling behind - but also from the perpetual need to review and clean up after an always productive, fast-shipping code tools spanning line upon line of code. Far from being plug-and-play, AI introduces layers of complexity - astute debugging, security concerns, compliance wrinkles, latency trade-offs - that demand seasoned navigators at the helm. This is not a bug; it’s a feature: complexity that can only be tamed by thoughtful, strategic minds.

Consequently, training and leading engineering teams will be more vital than ever. As AI systems get increasingly autonomous and code functions, then features, then entire products, our engineers will still need to “learn the craft” somehow. You can’t become an elite, frontier engineer overnight. You can’t become a senior developer able to supervise AI overnight. There’s no leapfrogging in human knowledge. And if we don’t want to create insane amounts of legacy, dependency and technical debt down the line, we still need software engineers that go through the learning process.

However yes, the developer journey is likely to be different. Entry-level “prompt coders” may pop up. Code curators, AI collaborators/trainers/reviewers, QA, AI-specific reliability, security or devops jobs will also probably be on the rise. However, the bulk of AI-driven innovation will hinge on those advanced engineers who can architect robust pipelines, version control data, and embed security by design. Also the talent geography can be impacted: remote AI collaboration tools can shift hiring to cheaper regions; as “middle-class engineers” may thrive in new markets rather than vanish. The software engineer of the future will be less focused on manual coding and more on system design, AI collaboration, integration, and oversight. New roles will emerge and career advancement will depend on the progressive mastering of AI-augmented development, strategic thinking, and the ability to orchestrate complex systems where human creativity and judgment remain essential.

It’s a bright future: one where the value of a dev increasingly lies not in their output, knowledge of many languages or tidiness, but in knowing how to identify the right problems to solve, knowing how to frame it correctly and leveraging an AI tool to shape a solution that’s robust, efficient, maintainable, scalable, understandable by other humans - and above all that does the job well.

And the coding itself? It will probably become a hobby for some - much like driving will be a hobby for circuits once countries start banning human drivers behind the wheel for safety reasons. Or it will be regarded as a curiosity from the past, like those accountants that used to calculate a P&L manually before Excel arrived (“oh wow, you did those ledgers BY HAND?! My God, I couldn’t do that”).

In short, commodity surface code is automatable ; systems thinking, architecture, and socio-technical stewardship are not.

From Enablement to Empowerment

Remember when Shopify emerged and empowered virtually anyone to become an online merchant? That seismic shift opened entrepreneurship to masses who never previously imagined running a business. Similarly, the GenAI coding revolution will open doors to an entirely new breed of software engineer-entrepreneurs: vibecoders, product builders, solopreneurs, creators.

Yet, just as Shopify didn't render e-commerce giants irrelevant, traditional software companies will persist. Handling, integrating, and modernizing tech infrastructure will remain essential: we have many more problems that software hasn’t tackled yet. Hell, “AI” is software itself.

Software development (in the sense, the development of software) is a cornerstone of future progress. Far from fading into irrelevance, software engineers will be more critical than ever to bridge the old and the new.

Software engineering is not dying: it's undergoing a metamorphosis. GenAI will democratize creativity and spark unprecedented opportunities, but the true industry shapers ; the tech companies that will tackle large-scale complex problems will master distribution, access to capital, proprietary IP, strategic enablement, and nuanced technical management. If you’re building one of those, we at OPRTRS invest in the founders that create those initiatives.

Far from the gloom of obsolescence, software engineers will find themselves central to a landscape richer, more complex, and more promising than ever. SWE's future isn't bleak - it's vibrant, sophisticated, and crucially, more human than we might realize.