My AI Coding Workflow 2026: This is how I AM CODING right now!

AI-Assisted Coding Workflow & Tooling

Key Concepts:

• Git: Version control system for tracking changes in code.
• BetterStack: Tool for rapidly scaffolding new projects with pre-configured stacks.
• Shadcn/ui: UI component library for streamlining design and maintaining consistency.
• GLM Coding Plan: Cost-effective Large Language Model (LLM) for coding tasks.
• Kilo Code: VS Code extension providing AI-powered pair programming.
• Verdant: AI agent platform for task delegation and code generation, emphasizing workspace isolation.
• Conductor: AI agent utilizing GLM for smaller, focused coding tasks.
• Sonnet: (Mentioned for comparison) AI coding agent, considered inferior to Conductor.
• Graphite: Git management tool with a strong PR review UI (now acquired by Cursor).
• Gemini Code Assist: Free, automatic code review tool integrated with GitHub.
• Grapile: AI-powered code review tool, offering faster and potentially superior review quality.
• PR (Pull Request): A request to merge code changes from a branch into the main branch.

Project Setup & Initial Stack

The speaker emphasizes a streamlined project setup process, leveraging automation to minimize manual configuration. They begin with Git for robust version control, a fundamental practice often overlooked by those new to coding.

To accelerate project initialization, they utilize BetterStack. This tool allows for the creation of custom technology stacks, enabling developers to quickly establish a project with desired components like Next.js for the frontend, Polar for payments, Convex for the backend, and Bun as the runtime environment. The benefit is eliminating unnecessary dependencies and optimizing for performance. A command can be copied from BetterStack to initiate the project.

For consistent visual design, the speaker employs shadcn/ui, a UI component library. This ensures a streamlined and cohesive aesthetic across the project.

Core Coding Agents & Task Delegation

The speaker currently relies on three primary AI coding agents, each serving a distinct purpose.

1. Kilo Code (Pair Programmer): Used directly within VS Code for small refactors, code understanding, and general assistance. The GLM coding plan is the preferred model due to its cost-effectiveness and model quality. Kilo Code is described as “super fast, snappy, and works to the point.” The speaker also utilizes APIs for other models through Kilo Code, recharging their account as needed.

2. Verdant (Task Delegator): This agent is used for delegating larger, more complex tasks. A key feature of Verdant is its workspace/branching system, which isolates tasks and prevents damage to the main repository. The speaker subscribes to Verdant’s $200 plan, praising its pricing, un-nerfed models, and UI/memory efficiency. Verdant allows for running “hundreds of tasks seamlessly” with continuous notifications.

3. Conductor (Focused Tasks): Utilizing the GLM coding plan, Conductor is reserved for smaller to medium-sized changes that don’t require extensive file modifications. This strategy conserves Verdant credits for more demanding tasks. Conductor is described as capable of handling complex tasks reliably, but caution is advised when providing it with documentation for large libraries. The speaker recommends Conductor over Sonnet, citing its superior performance at a lower cost.

Code Review & Quality Assurance

Once code is generated, the speaker initiates a Pull Request (PR) to their repository. Code review is a critical step, and the speaker employs a two-pronged approach:

1. Gemini Code Assist: A fully automated and free code review tool that provides line-by-line annotations. The speaker notes the surprisingly high quality of the reviews.

2. Grapile: An AI-powered code review tool that the speaker is increasingly favoring due to its speed and improved review quality. A current limitation is the need to directly mention Grapile within GitHub, as it doesn’t function correctly through Graphite.

The speaker manages their Git workflow using Graphite, which was recently acquired by Cursor. While concerned about the acquisition, they believe Graphite will continue to function. Graphite’s UI is particularly valued for its stacked merges, allowing for simultaneous merging of multiple PRs and efficient conflict resolution.

Workflow Philosophy & Tooling Considerations

The speaker intentionally avoids more complex tools like Oh My Open Code or Ralph Loops, preferring a more manageable and cost-effective workflow. They prioritize staying “in the loop” during the coding process, regularly checking agent output for code quality and accuracy.

They express enthusiasm for the advancements in Gen AI coding tools, specifically highlighting Opus 4.5 and GLM4.7, and the improving UIs of platforms like Verdant and Conductor.

Synthesis/Conclusion:

The speaker’s workflow demonstrates a pragmatic approach to integrating AI into the software development lifecycle. By strategically combining specialized AI agents with established tools like Git and VS Code, they aim to automate repetitive tasks, delegate complex coding assignments, and maintain a high level of code quality. The emphasis on cost-effectiveness, workspace isolation, and continuous monitoring underscores a focus on practical implementation and control, rather than complete automation. The key takeaway is a layered approach, leveraging the strengths of different AI tools for specific stages of the development process, while retaining human oversight and quality assurance.