What is low-code?

Low-code is a software development approach that lets people build applications with minimal hand coding. Instead of writing everything from scratch, the developer uses visual tools such as drag-and-drop interfaces, prebuilt components, workflows and templates.

Without a low-code platform, traditional development involves writing hundreds or thousands of lines of code to create even a simple app. With low-code, users with less formal coding ability can design with intuitive visual interfaces. Seasoned developers can benefit from these capabilities as well.

For example, some business applications require knowledge of a specific programming language, narrowing the selection of developers. By eliminating this bottleneck, low-code development platforms shorten the software development lifecycle (SDLC), enabling experienced engineers to accomplish more, faster.

The graphical user interface and drag-and-drop features of a low-code platform automate aspects of the development process, eliminating dependencies on traditional coding approaches. Low-code platforms democratize app development, particularly for citizen developers—that is, business users with little formal coding experience, such as business analysts or project managers.

Unlike no-code development platforms, which enable users to create applications and automate business processes without writing any code, low-code environments are used often used by professional developers to automate the generic aspects of coding. This redirects effort on the “last mile” of development.

Both approaches aim to abstract the complex aspects of coding by using visual interfaces and pre-configured templates, but they have their differences in terms of intended users, use cases and more.

The rise of AI-assisted coding has blurred the lines that previously defined these two categories. This shift represents the latest step in the long evolution of low-code stretching back to the 1980s.

In the 1980s, new tools allowed developers to create database-driven applications with significantly less code than traditional programming methods.

The 1990s saw the emergence of rapid application development (RAD), which focused on accelerating software delivery, in part through visual tools. Integrated development environments (IDEs) like Microsoft Visual Basic helped popularize drag-and-drop interfaces.

Visual designers, reusable controls and event-driven programming reduced development complexity and foreshadowed many of the capabilities found in modern low-code platforms.

The 2000s brought business process management (BPM) platforms to address the need for tools that could enable process automations without requiring extensive coding resources. Tools emerged to allow organizations to visually model and automate workflows.

The following decade introduced cloud services, which created ideal conditions for low-code platforms to flourish, leading to the rise of the “citizen developer” as a concept. Business users with limited coding skills needed to be able to build apps and workflows for their teams and low-code platforms filled the gap. Low-code platforms expanded beyond workflow automation to support customer-facing apps, mobile apps, API and data integration, analytics dashboards and more.

The 2020s have seen a new chapter in the evolution of low-code platforms. Previously, low-code vendors competed on a simple premise: software development is too difficult, so developers and business users need visual abstractions to avoid writing code. Software development was slow and skilled developers were scarce and expensive. Business users with limited coding ability could not participate in app creation. Platforms like Mendix, Microsoft Power Apps and Appian addressed this through smart visual abstractions that provided speed in exchange for control.

New AI-first coding systems have introduced a new proposition: What if generative AI can just write most or all of the code for you? Today’s agentic coding systems remove much of the need for visual abstractions. Now, you can make code do what you want by simply telling a model what to make.

For example, a developer can now describe an application in natural language: “Build a customer portal with authentication, subscription billing, email notifications and an admin dashboard.” An agentic system can generate much of the project structure, infrastructure configuration, database schema, APIs, tests and frontend components automatically.

Large language models (LLMs) can generate code, design workflows, create user interfaces and translate natural language prompts into software components. This has even enabled a practice known as “vibe coding,” the practice of prompting AI tools to generate code without even having to understand the fundamentals of programming, though this practice comes with serious risks, especially in the context of creating enterprise applications. Today’s organizations are now grappling with new forms of shadow IT, when employees use AI-assisted coding platforms without proper oversight and established processes.

AI agents can do much more than generate snippets of code. They can:

• Analyze entire repositories

• Understand application architecture

• Create multiple files simultaneously

• Refactor large systems

• Generate tests

• Debug issues

• Execute commands and tooling

• Iterate toward a goal autonomously

While many low-code vendors have integrated AI assistants that allow users to describe applications in plain human language, a more expansive shift is underway toward AI code generation platforms powered by agentic AI.

Agentic coding is more than AI-assisted autocomplete. These systems serve as collaborators with developers, with agents that can not only generate code but take autonomous actions across the SDLC. Such platforms increasingly include GitHub Copilot, OpenAI Codex, Anthropic’s Claude Code, Cursor and IBM Bob.

Some have declared the obsolescence of low-code as it was once known. A major consequence of agentic coding is the paradigm of “code-first low-code.” The result of an agentic coding project is not a low-code artifact but a conventional codebase. This means full ownership, custom code, reduced vendor lock-in and compatibility with modern tools.

Some organizations that would have chosen a low-code platform a few years ago may now instead use AI agents to develop software. In response to this, many low-code vendors have incorporated agentic AI into their products. The visual canvases these platforms are known for remain, but the manual coding experience is replaced with AI code generation.

In effect, low-code vendors are evolving their products from visual builders into orchestration environments for AI agents. The boundary between low-code and software engineering is fading, with agentic coding blurring the distinction. Instead of replacing code, agents replace much of the human expertise and effort required to produce code, an even more powerful abstraction. For these reasons, many think of agentic coding not as the next phase of low-code, but its successor.