The Accidental Orchestrator For over two decades, I’ve written about software development for practitioners, covering coding, architecture, project management, and team dynamics. In recent years, my focus has shifted to AI and its role in software engineering. Despite the growing interest in AI tools like Claude Code, Copilot, and Cursor, I’ve struggled to find a structured approach for experienced developers to integrate these tools effectively. While there are plenty of tips and hype, there’s little guidance on how to practice, teach, or improve agentic engineering—a discipline that combines AI agents with human expertise. The debate around AI in software development often splits into two extremes: one claiming AI will render developers obsolete, the other insisting it’s just another tool. Neither view is accurate. AI doesn’t replace human expertise; it raises the bar for what developers need to know. The gap between theoretical understanding and practical application is a major source of anxiety for engineers. Many know they should review AI-generated code, maintain architecture, write tests, and stay in control of the codebase, but applying these principles in practice remains challenging. This tension led me to experiment with agentic engineering by building a production system from scratch, with AI writing all the code. The goal was to test a structured approach to using AI tools while addressing the complexities of real-world development. I chose Monte Carlo simulations as the project, a decision rooted in my childhood fascination with the technique. My father, an epidemiologist, introduced me to the concept of using simulations to uncover patterns in chaotic data.#ai #monte_carlo_simulations #drunken_sailor_problem #agentic_engineering #software_development
GPT-5.4 Now Available in GitHub Copilot OpenAI’s latest agentic coding model, GPT-5.4, is now being rolled out in GitHub Copilot. Early testing of the model’s real-world capabilities in software development has shown it achieving higher success rates compared to previous versions. The model demonstrates improved logical reasoning and task execution, particularly in handling complex, multi-step processes that require the use of external tools. The update is available to GitHub Copilot users on Pro, Pro+, Business, and Enterprise plans. Users can select the GPT-5.4 model through the model picker within the Copilot interface. While the model is accessible across all supported versions, OpenAI recommends upgrading to the most recent version of each product to ensure optimal performance with prompting and model parameters. For Copilot Enterprise and Copilot Business plan administrators, the GPT-5.4 model must be enabled through the Copilot settings by applying the appropriate policy. Detailed information about the available models in GitHub Copilot can be found in the official documentation, which also provides guidance on getting started with the platform. Users are encouraged to join the GitHub Community to share feedback and insights about their experiences with GPT-5.4. The release marks a significant step in enhancing the capabilities of AI-driven coding tools, offering developers more efficient and reliable assistance in their workflows.#software_development #github_copilot #openai #gpt_5_4 #github_community
