Agentic development is transforming software creation multiplying the productivity of developers. The change happens faster than most teams anticipated. What began as a shift to AI-assisted software development inside traditional IDEs for modest productivity gains, is now evolving into agentic engineering workflows writing code, generating tests, and code documentation. While code written by AI agents for open-source frameworks such as Qt is already of good quality, the technology is not ready for fully autonomous AI agents. Such gaps can be efficiently closed by dedicated agent skills and MCP tools while keeping the human in the loop as the director of workflows.
Software development is undergoing a fundamental shift from traditional IDE use and AI-assisted coding to fully agentic engineering. Open frameworks and ecosystems, including Qt, are well-positioned in this new reality, allowing AI models to learn them and making them a natural fit for AI-driven workflows. However, AI agents still fall short of handling all steps of the software creation workflow.
2024 was the year where coding assistants in IDEs had their breakthrough. The shift towards agentic development solutions such as Claude Code and Codex gained momentum during 2025. The centre of gravity is transitioning further towards agentic engineering in 2026. Accelerating this trend, Cursor, once the rising star of the next generation IDE, introduced recently their agentic development UX, which unsurprisingly feels like Claude Code. While traditional IDEs will have their benefits also during the current decade, the transition away from pure manual programming is unstoppable.
Frontier models have already been getting great in writing Qt-based UIs. Since Qt comes with a source-code-available approach, it has allowed Large Language Models to learn from a wide variety of pre-training material.
Claude, GPT, and Gemini models are achieving now coding performances on benchmarks, such as the QML100 benchmark for single-turn tasks, of over 75% and up to 86%. Fortunately to Qt, cross-platform frameworks which have a source-code-available approach have allowed frontier models to learn from a wide variety of pre-training material. The wide adoption of the Qt Framework and the resulting fine-tuning of frontier models over time through Reinforcement Learning from Human Feedback (RLHF) cements Qt’s availability for AI agents further.
Even when frontier models have become sufficiently good at writing Qt UI code for daily tasks, the shortcomings of agentic development solutions are still significant. The list of tasks that AI agents cannot do yet is still longer than the things they actually can do: Converting Figma designs to UI code, managing CMake project context, implementing complex UI controls, installing Qt libraries on the fly, creating Qt Test cases, or doing a deep analysis as the code review with a linter. Even frontier models fall short in completing these tasks purely based on their pre-training and fine-tuning.
Making traditional tools such as linters or code performance profilers available to AI agents through the Model Context Protocol (MCP) will close some of the gaps between human developers and AI agents.
Agent skills bring further domain expertise directly into the AI-powered development workflow. A skill is a portable, version-controlled package that gives an AI agent specialized knowledge and capabilities for a specific task — from avoiding common mistakes when writing code, creating unit tests, deep analysis of Qt C++ code, and producing API documentation. Rather than relying on pre-trained model knowledge alone, skills let UI developers compose an agentic workflow precisely tuned to their needs.
Each skill loads its resources on demand, so the agent's active context stays lean regardless of how many skills are installed. When a task is triggered — whether that's writing a QML binding, reviewing code, or generating structured code documentation— the AI agent selects the relevant skill automatically based on context, executes it using its built-in tools, and delivers a high-quality, consistent result.
Because agent skills follow an open standard, skills authored for development workflows are portable across any compliant agentic development solution, making them a durable investment for teams building cross-platform applications, especially embedded device UIs.
Image: Illustration of Qt's C++ Code Review skill
The role of the software developer is evolving from simple AI-assisted software development to setting direction and governing the work that AI agents carry out. At Qt, our intent is to actively enable this shift for cross-platform development, building toward end-to-end agentic workflows that values keeping the human in the loop.
Plug-in coding assistants for IDEs like GitHub Copilot, at least before they started to adopt also agentic behavior, delivered productivity gains in the single digits. Agentic development solutions, however, show the promise of multiplying software engineering velocity by automating significant parts of the application creation. We can see this already happening for web, desktop, and mobile app development.
While we at Qt believe that the human in the loop will always be crucial for professional software development, we intend to enable agentic engineering of cross-platform applications, starting by providing skills and MCP-enabled tools that then build up to workflows.
Figure: Simplified illustration of an agentic development workflow
We expect that the modern cross-platform developer will delegate many software engineering tasks, such as writing boiler plate code, creating test cases and code documentation to AI agents. The role of the software developer will shift, over time, towards making architecture and technology decisions, governing the workflow with development roadmaps and testing harnesses, and guiding AI agents towards the desired goal. The new bottleneck for the development workflow will be code reviews and quality assurance.
As more code and more test cases are being created by AI agents, code reviews, especially in regulated industries, will take new forms. Deep code analysis, hunting for verbose and inefficient code constructs, will help battling future technology debt. Code security analysis, for example with models such as Claude Mythos, will help embedded device code generated by AI agents to meet cybersecurity resilience requirements, such as EU Cybersecurity Resilience Act.
AI agents are writing Qt UI code already now, thanks to Qt C++ and QML training data being available already for decades. But this is only the beginning. Cross-platform development will see its biggest transformation ever in the coming years.
While the need for skilled software engineers and optimized IDEs will continue to exist for years to come, for-profit organizations will invest into automating their development workflow with agentic development solutions to stay competitive. Organizations will increasingly adopt and adjust AI agent skills and MCP-tools across their development teams while coaching their individual contributors to focus on innovative, complex, and responsibility-bearing tasks. The Qt Group will support our customers and our ecosystem in this journey.