How I use AI (Oct 2026): Agentic Coding and the New AI Workflow
As of October 2026, the landscape of AI-assisted work has shifted from simple autocomplete to a reliance on agentic workflows. For many, the primary takeaway is that while AI agents can now handle the bulk of routine coding tasks, they require significant human intervention to ensure architectural integrity, test validity, and project coherence.
The Shift to Agentic Coding
Coding is no longer a manual process of typing syntax; it has become a process of managing multiple AI agents. By utilizing tools like Claude Code, developers can delegate features, bug fixes, and test generation to autonomous agents. These agents operate in isolated Git worktrees, allowing for rapid prototyping and the ability to "fail fast" on new ideas. However, this shift introduces new friction: managing 5-6 concurrent agent conversations is mentally taxing and can lead to overstimulation. Furthermore, agents often produce "vibe-coded" output that lacks the precision and maintainability of hand-written code, necessitating a human to act as an architect and reviewer rather than a typist.
The Role of Human Oversight
Despite the capability of agents to generate functional code, human expertise remains critical for several reasons:
- Architectural Design: Agents excel at implementation but struggle with high-level system design. They often produce over-architected or unreadable solutions that require human refactoring.
- Validation Loops: Agentic pull requests are helpful but not yet reliable enough to bypass manual review. Developers must still validate tests and logic to prevent systemic issues.
- The "AI Fog": Relying too heavily on agents can lead to a loss of understanding of the underlying codebase. Successful projects often involve a hybrid approach where the human writes core tools and lets agents handle the surrounding boilerplate.
Research and Learning
Generative AI has become the default interface for search and learning. While traditional search engines now prioritize AI-generated summaries, these tools are best used for exploring well-known concepts rather than novel research. AI serves as an excellent tutor for drilling repetitive practice or explaining complex documents, but users must guard against the tendency of these models to "forget" their role as a teacher and revert to providing direct answers, which bypasses the user's need for deep, Type 2 thinking.
The State of Media and Creativity
AI's role in creative fields remains controversial and highly polarized. While AI music generation has fostered new, engaging genres like "Doompop," the commercialization of AI-generated art and video is widely viewed as a negative trend. The prevalence of "AI slop"—low-effort, mass-produced content designed purely for ad revenue—has soured many on the utility of image and video generation tools. The consensus among many practitioners is that these tools are most valuable when used as a lens for human creativity, rather than a replacement for it.
Critical Perspectives and Community Insights
Community discussion surrounding these developments highlights a divide between the perceived productivity gains and the loss of traditional craft. Key insights include:
"Managing 6 different conversations with 6 different robots about 6 different features is overstimulating & exhausting in a way that coding never was."
Critics also point out that the high cost of token usage—sometimes reaching thousands of dollars per month—is unsustainable and potentially inefficient compared to local infrastructure. There is also significant skepticism regarding the "solved" nature of coding; many argue that the need for deep code reading and understanding is more important than ever to catch the subtle, incorrect assumptions baked into AI-generated output. Ultimately, the most successful practitioners view AI as a tool to accelerate their own work, not as a substitute for the fundamental skills of engineering or creative expression.
Sources
Related
- Dispatch
- Dispatch
- Dispatch
- Dispatch
- Dispatch