For several years, I was one of Power Apps' strongest advocates. I collected every single Power Platform certification Microsoft offered, from the fundamental badges all the way through to architect-level credentials. I helped teams build internal tools and championed the idea that low-code was the fastest way to deliver applications in environments where IT teams were stretched thin.
And at the time, that made perfect sense.
A lot of that belief came from my own background. I was always comfortable with the back end. Databases, APIs, integration layers. But front-end development was never something I felt confident in or had the capacity to learn deeply. Frameworks evolved quickly. UI design was its own discipline.
Power Apps solved that problem for me. It let me focus on the logic without getting lost in the visual side.
But low-code did not become popular by accident. For many organisations it solved a real bottleneck. Teams needed internal tools quickly, and development capacity was limited. Low-code allowed analysts, operations teams and domain experts to build working applications without waiting months for engineering resources.
In many environments that was transformational.
What has changed is not the business need. What has changed is the cost of writing software.
The ecosystem shifted. AI coding assistants changed the equation completely.
The early tools were helpful but inconsistent. Then I tried Claude, and everything moved forward dramatically. I could generate front-end components simply by describing them. I could request layout refinements, responsive behaviour, state logic or API integrations, and Claude produced clean working React or TypeScript code.
Then magicpatterns took that a step further by letting me describe entire screens and workflows. It produced full wireframes with working mock data that I could export and use immediately.
And then came Kiro, the agentic IDE from AWS.
Kiro is not a code generator. It behaves more like a staff-level engineer reviewing everything you build. It refactors projects, restructures folders, enforces patterns, catches logical inconsistencies and helps clean up technical debt before it appears.
Instead of simply producing code, Kiro maintains code quality as the project evolves. It felt less like using a tool and more like having a senior engineer quietly reviewing everything in the background.
On Kiro, the agentic IDE from AWSThe real shift is not that AI writes code. It is that software architecture, once a bottleneck, is becoming conversational.
This combination changed how I build software.
Earlier this year, I built a complete public-facing application in less than two days. The front end ran on Azure Static Web Apps. The API layer was an Azure Function. The back end was a fully indexed relational Azure SQL database with stored procedures and proper OLTP design.
The architecture itself was straightforward. What changed was the speed.
How it actually worked: ChatGPT helped me think through the architecture and generate detailed prompts, often through voice chat while walking between meetings. magicpatterns generated the working UI wireframe and components. Claude wrote the API layer. Claude and Gemini helped produce the SQL schema and stored procedures. And Kiro reviewed and improved the entire repository afterwards, restructuring the project so it looked like something built by a team rather than a single developer working quickly.
The speed difference was significant. Tasks that previously took days inside Power Apps now take a fraction of the time with AI-assisted full-stack development. Instead of being confined to a visual editor, I can shape the application exactly the way I want. I can version the code. I can scale it without friction.
Full control
There are no restrictions. I can change the UI, restructure the logic, rewrite the API layer, redesign the data model or implement security exactly how I want it. Nothing is locked behind a formula bar or a predefined widget.
The entire internet becomes your knowledge base
Most Power Apps resources focus on narrow patterns or isolated formulas. Full-stack development has more than a decade of open-source knowledge behind it. React components, SQL patterns, Python workflows and infrastructure patterns are all widely documented. The answers are everywhere.
Code is portable
React, Node, .NET and Python can run almost anywhere. Azure, AWS, containers, serverless platforms, static hosting and mobile frameworks. Low-code applications are usually tied to the ecosystem that created them.
Versioning and DevOps are natural
Git was built for code. Pull requests, CI pipelines, testing, environments and proper release management all fit naturally into the workflow. Low-code platforms attempt to retrofit this process, but it rarely feels native.
AI fills the skill gaps
You do not need to be a senior engineer to start building. AI can generate scaffolding, components and integration logic that previously required deep expertise. The developer's role shifts from writing every line to describing systems, validating outputs and refining architecture. In many ways, this was always the promise of low-code. AI simply delivers it through a different path.
AI understands code better than low-code platforms
Modern AI models were trained on enormous amounts of publicly available code. Open-source repositories, documentation, Stack Overflow discussions, tutorials and technical blogs form a massive shared knowledge base spanning decades of software development. Frameworks like React, Python, SQL and Node are deeply represented in that training data. Low-code platforms do not have the same footprint. Their logic often lives inside proprietary editors, visual designers or closed ecosystems that are not publicly visible. As a result, AI can generate, refactor and debug traditional code far more effectively than it can reason about formula-based or visual low-code systems.
The irony: The same open ecosystem that made full-stack development powerful for developers is now what makes it powerful for AI as well.
Of course, giving full-stack capabilities to non-developers introduces risk. But it is important to remember that low-code was never an ungoverned playground. Power Platform administrators still set boundaries, applied security policies and controlled environments. Without that governance, poorly designed low-code apps could break things just as easily.
Full-stack development is no different.
Senior developers can review pull requests. Cloud engineers can control permissions and infrastructure. Security policies can still govern deployment pipelines.
AI assistance also introduces a new discipline developers will need to learn: verification. Generated code can look correct while hiding subtle flaws. The role of the engineer increasingly shifts from writing every line to validating architecture, security, edge cases and operational behaviour.
In enterprise environments this shift can actually be easier to manage. When infrastructure, CI pipelines and cloud resources are already standardised, AI-assisted development removes much of the remaining friction between an idea and a working system.
The guardrails remain. But the speed improves dramatically.
What makes this shift particularly powerful inside organisations is that the people who understand the operational problem are often not professional software engineers. With AI-assisted development, those domain experts can move much closer to building the solution themselves while still operating inside engineering guardrails.
Will Low-Code Disappear?
Not immediately.
Low-code still excels in certain scenarios. Internal forms, simple workflows, departmental tools and applications tightly integrated with Microsoft 365 can often be built faster inside Power Platform than from scratch.
For organisations already deeply invested in the ecosystem, it remains a practical tool.
But the strategic picture is shifting. Low-code was built on the assumption that coding is difficult and slow. AI is rapidly erasing that assumption.
The Likely Future: Low-Code Becomes the "Excel of App Building"
Like Excel, low-code will always have a place. It will remain the convenience layer. The rapid prototyping environment. The place where ideas begin and simple solutions live.
But full-stack development, accelerated and guided by AI, is becoming accessible, fast and intuitive in a way it never was before.
Once organisations experience that shift, the limitations of low-code platforms become much harder to justify for systems that need scale, flexibility or long-term maintainability.
The real change is not that AI replaces developers. It is that the distance between an idea and a working system is collapsing. And when that distance shrinks, the people closest to the problem suddenly have the power to build the solution.