Taeho Kim
Menu
HomeWritingNotesReadingAbout
KOEN

Background and Career

The author holds a computer science degree and has been doing .NET development since .NET first appeared 20 years ago. He has built backend systems, microservices, and cloud architecture at 3 Fortune 500 companies, led teams of multiple engineers, and successfully delivered projects generating $2-3 million per month in revenue and processing over 60,000 transactions per minute. He's not a FAANG engineer, but has made it to the final rounds at some of those interviews.

"I've been doing .NET development since .NET first came out, and my Reddit post history shows activity in C#, .NET, and programming subreddits going back 20 years."

Why Claude AI

Despite extensive experience across many projects, the author candidly admits that with age, absorbing massive business models or codebases at a new team isn't as easy as it used to be.

"My brain no longer absorbs information bombs the way it used to."

So he uses Claude AI to rapidly acquire structured information. Specifically, he asks Claude for three things:

  • Business terminology: Key concepts and terms used in the project or company
  • Technical terminology: Frequently used terms across the tech stack, frameworks, and libraries
  • Codebase patterns and conventions: Code structure, recurring design patterns, team coding style

This lets him "upload" the codebase into his mind before diving into work.

Actual Claude AI Workflow

The author uses Claude in the following way for each project:

1. Optimize Requests with the Lyra Prompt
  • Customizes the Lyra prompt (searchable on Google) to optimize every request sent to Claude
  • This was a major productivity turning point

"Using the Lyra prompt to optimize my requests was a huge turning point for me."

2. Rewrite Jira Tickets
  • When a new task comes in, rewrites the Jira ticket with Claude
  • This creates a clearer, more focused context to start working from
3. Task Chunking
  • Asks Claude to break the ticket into the smallest implementable units
  • Then feeds the first chunk into the prompt optimizer to begin work
4. Scoped Prompting
  • This is where the author says "the real magic happens"
  • Very strictly limits the scope of what Claude handles
    • For example: define only a specific interface, target only a specific method, or first request only red/green unit tests
  • The goal is to make each output small and digestible enough to read and evaluate in minutes

"I scope Claude's work very tightly. Sometimes just defining an interface, sometimes targeting a specific method. Other times I'll only ask for red/green unit tests first."

5. Iterative Development
  • Refines each chunk until it's good enough
  • Once a chunk is complete, moves to the next and repeats

How Claude AI Has Changed Work, and Advice

Thanks to this workflow, the author can organize thoughts and work systematically even in complex environments.

"Claude isn't just helping me code -- it's helping me organize, structure, and stay sharp without being overwhelmed by complexity."

Regarding the recent study showing 16 engineers were actually slower using Claude, the author advises trying his workflow first:

  • Understand the tool
  • Limit the scope
  • Maintain a consistent process
  • Find what works for you

This, he emphasizes, is the key to the AI era.

"Before using AI as a friendly pair programmer, read this workflow first. Understanding the tool, limiting scope, maintaining a consistent process, and finding what works for you are the keys to this AI kingdom."

Claude AI Workflow Summary Diagram

Below is a visual overview of the author's Claude AI workflow:

graph TD
    A[Start task] --> B[Optimize request with Lyra prompt]
    B --> C[Rewrite Jira ticket with Claude]
    C --> D[Break task into small units]
    D --> E[Send scoped request to Claude]
    E --> F[Iteratively develop each chunk]
    F --> G[Move to next chunk]
    G --> E

Key Concepts

  • How an experienced engineer uses AI
  • Structured information acquisition
  • Prompt optimization
  • Task chunking and scope limitation
  • Iterative development
  • Proper use of AI tools

This post demonstrates, with real examples, how an experienced developer can smartly leverage AI tools. It emphasizes that structuring information, breaking down tasks, and limiting scope are the keys, and that finding a consistent process and your own approach is essential for getting the most from AI.

Related writing

Harvest•Engineering Leadership•English

Marc Brooker on Incidents and AI Engineering

This summary captures Marc Brooker's lessons from thousands of postmortems and his view of how AI is changing software engineering. The main idea is that engineering judgment still compounds through writing, systems thinking, and careful problem selection.

Apr 14, 2026Read more
Harvest•Startup · Engineering Leadership•English

Keith Rabois on Team Building in the AI Era

This summary distills Keith Rabois's blunt advice on hiring, organizational design, customer feedback, and personal standards in the AI era. The recurring theme is that great companies are still built by exceptional people, not by process theater.

Apr 13, 2026Read more
Harvest•Engineering Leadership•English

The New Standard for Engineers in the AI Era

Gergely Orosz reflects on how AI is changing software engineering careers, from coding skill and management paths to the traits that still matter. His main view is that adaptability, judgment, and product sense are becoming even more important, not less.

Apr 12, 2026Read more