Low-Code Is Dead

I've spent the last decade advising large enterprises on digital transformation. I've sat in rooms where low-code platforms were presented as the silver bullet — the thing that would finally close the gap between the business and IT. I've seen the slide decks, the vendor demos, the pilot projects. And I've watched the promises quietly evaporate.

Low-code had a good run. But it's over. AI didn't just compete with it — AI rendered the entire value proposition obsolete.

What low-code actually promised

The sales pitch was elegant: give your business analysts, operations teams, and "citizen developers" a visual, drag-and-drop environment to build apps without writing code. Free up your developers. Move faster. Reduce the backlog.

For a narrow band of use cases — automating a form-based workflow, building a simple approval process — it worked. Tools like OutSystems, Mendix, and later Microsoft Power Apps delivered real value. The market responded accordingly, reaching multi-billion dollar valuations on the back of genuine enterprise adoption.

But the ceiling was always low. And the floor was surprisingly high.

The dirty secret of low-code: The moment your use case grew in complexity, you needed developers anyway. You just had developers wrestling with a proprietary abstraction layer on top of the actual problem — which made everything slower, not faster.

Where the cracks always showed

Every low-code platform created the same failure pattern. A business analyst would build something in a weekend. It worked brilliantly for three months. Then the business requirement changed, the integration needed to go deeper, and the analyst hit the platform's invisible wall. IT was called in. They stared at the visual model, the auto-generated code, the vendor lock-in, and sighed.

The platforms responded to this by adding more features — more connectors, more components, more configuration options. Which made them more complex. Which required more training. Which eroded the "no-code, anyone can do it" narrative. The fastest implementations were always the simplest ones, which were also the ones that needed the least sophistication to begin with.

And then AI arrived.

"The gap between intent and implementation — the gap that low-code existed to fill — has collapsed."

What AI actually changed

The shift didn't happen because AI is better at dragging boxes around a canvas. It happened because AI fundamentally changed what it means to express intent to a computer. A business analyst who previously needed a low-code platform to translate their mental model into a working app can now describe what they want in plain language — and get working code back in seconds.

Not a configured template. Not a drag-and-drop approximation. Actual code. Testable, modifiable, deployable code. Code that doesn't require a vendor's runtime environment, doesn't carry proprietary data model lock-in, and doesn't hit an invisible ceiling when requirements evolve.

The gap between intent and implementation — the gap that low-code existed to fill — has collapsed. Not by making programming visual, but by making language itself the interface.

The numbers tell the story

GitHub Copilot launched in 2021. By late 2023, it was completing over 30% of code in repositories where it was enabled. By mid-2025, the latest generation of AI coding assistants — Claude, GPT-4o, Gemini — were regularly producing production-quality code across complex, multi-file tasks with minimal correction. The velocity delta between a developer with AI assistance and a citizen developer with a low-code platform has flipped decisively.

Meanwhile, Power Apps — Microsoft's flagship low-code offering — has quietly been repositioned. Microsoft Copilot is the story now. The low-code UI isn't going away, but it's no longer the lead. It's a configuration surface for AI-generated logic.

The enterprise reaction

I'm seeing this play out in client conversations right now. The enterprises that spent significant budget on OutSystems or Mendix licenses are asking harder questions at renewal time. Not because the tools are bad, but because the calculus has changed. A well-prompted AI assistant with a developer in the loop can produce a comparable output faster, with less vendor dependency, at lower total cost.

The citizen developer use case — the one that justified the premium pricing — is being absorbed into tools like Microsoft 365 Copilot, where AI-generated Power Automate flows and basic apps are an embedded capability, not a separate platform purchase. The standalone low-code vendor proposition is getting squeezed from both sides: AI commoditises the simple end, and the complex end still requires real developers.

None of this happens overnight. Enterprises move slowly — existing low-code contracts run for years, embedded workflows don't get ripped out without a business case, and AI adoption inside large organisations is still uneven at best. "Strategically terminal" is the more precise diagnosis; the body just hasn't caught up with the prognosis yet.

What I'm telling clients: Existing low-code investments don't need to be abandoned overnight. But new low-code platform commitments require a very clear answer to the question: why not AI-assisted development instead? The answer is narrowing.

What survives

Not everything dies. A few things will persist and evolve.

Workflow automation at the edges. Simple, human-readable automations — "when this email arrives, create this record" — will continue to live in low-code interfaces. But they'll be increasingly AI-generated rather than human-configured. The platform becomes the execution environment, not the authoring environment.

Regulated industries with specific compliance requirements. Where auditability, vendor certification, and explicit IT governance matter, known low-code platforms with validated deployment tracks retain an advantage. This is a real but shrinking moat.

Embedded AI tooling within existing platforms. SAP Build, Salesforce Flow, Microsoft Power Platform — these survive because they're not standalone low-code plays. They're integration surfaces within platforms organisations are already committed to. They'll absorb AI capabilities and continue to serve configuration and extension use cases.

What doesn't survive is the standalone, premium-priced, citizen-developer-as-hero narrative. That story required a world where expressing software intent was hard. AI made it easier. The market will follow.

The harder question

Low-code's death also exposes something uncomfortable: the digital skills gap it was supposed to solve still exists. We didn't close it — we just found a tool that made it less visible for a while. AI-assisted development lowers the bar again, but it doesn't eliminate the need for people who understand systems, data models, security, and architecture.

The citizen developer of 2026 isn't someone who learned to use a visual workflow builder. It's someone who can describe a problem precisely enough for an AI to solve it, evaluate the output critically, and connect it to the right systems. That's a different skill set — and a harder one to teach at scale.

Low-code promised democratisation through abstraction. AI delivers democratisation through dialogue. The result is more powerful, more flexible, and more honest about what it is. The abstraction layer is gone. What's left is the thing itself.