My Real AI Coding Workflow (build anything)

Tech With Tim ·2026-06-24 ·1 min read

Summary written by us from the video's transcript. The video, and everything in it, is Tech With Tim's work.

Learn how to set up an AI‑assisted coding workflow—from idea and tool selection through environment configuration, planning, and debugging—using Cursor, Claude, Whisper, and ImageKit to build a web app that creates short video clips from long videos.

Takeaways

Spend dedicated time on project definition and tool research before writing code.
Leverage Cursor’s agent skills and MCP servers to integrate services like ImageKit without manual API handling.
Use spoken prompts (Whisper Flow) to quickly convey complex requirements to Claude within Cursor.
Automate Git commits via Cursor rules to maintain version control without extra commands.
Iteratively debug AI‑generated pipelines by feeding error messages back into the model for corrections.

Define the project and research tools

Start by clearly stating the goal: an AI shorts tool that converts a 16:9 video into multiple vertical clips with captions and titles. Spend time researching existing solutions, tech stacks, and relevant APIs before coding to avoid poor results later.

Select the core components

Identify three essential services: Whisper (from Grok) for audio transcription, Claude (Sonnet model) for selecting clip moments and generating text, and ImageKit for video transformations, captioning, and hosting. Choose Next.js as the web framework to tie everything together.

Configure Cursor with agents and MCP servers

Install Cursor, open a new project folder, and set the model (Opus 8) with a large context window. Add agent skills and MCP server definitions for ImageKit (and optionally GitHub) via a terminal command, then authenticate using your ImageKit private key in the MCP settings. This lets Cursor invoke tool‑specific actions automatically.

Create a high‑level plan inside Cursor

Use Whisper Flow to dictate a prompt that outlines the app’s purpose, stack, and required APIs. Ask Claude to generate an architecture document and a step‑by‑step implementation roadmap, storing it as markdown in the project for reference throughout development.

Set up version control and automation rules

Tell Cursor to create a GitHub repository and enforce a rule that commits are made after major changes. The rule file is stored under a rules folder so future code generation automatically includes commit steps, keeping history clean.

Implement, test, and debug iteratively

Run the generated code, upload a video via ImageKit, and observe failures (e.g., transcription errors). Provide error messages to Claude, clarify that only audio should be sent to Whisper, and adjust the pipeline using ImageKit’s URL transformation parameters to extract audio. Repeat debugging cycles until transcripts and clip selections appear correctly.