Why I started blogging
tldr: writing skill will only become more important when AI writes more code.
Why Writing Matters More Than Ever for Engineers
This is my first public blog. I'm Harrison Tin, co-founder of Palmier. I left my big tech job in 2024 to work on my own startup with my co-founder. Over this past year, I've written thousands of lines of code — much of it generated by AI. This experience has fundamentally changed how I think about what skills matter for software engineers as AI becomes more capable.
The surprising answer? Writing.
The Art of Talking to LLM
Coding agents are everywhere now — we've even built one ourselves. Tools like Cursor and Claude Code have become essential parts of my daily workflow. These agents are essentially API calls to language models, and their effectiveness depends entirely on the context you provide.
Most companies use the best models on the market (today that's gpt-5 and sonnet-4/opus-4.1) rather than training their own models, due to cost and expertise constraints. This means the only way we can control output quality is through input quality. You hear buzzwords like "context engineering" (managing context windows during agent loops) and "prompt engineering" (writing your goals precisely) — and prompt engineering is just good writing.
Every Conversation Starts from Zero
Here's the crucial insight: talking to an LLM is like meeting a cracked but completely uninformed intern every single time. They have no memory of previous conversations (some could say they do after context engineering), no understanding of your preferences, and no context about your goals.
This means you need to be explicit about everything. Those mental shortcuts and assumptions that work with human teammates? Useless here. You need to spell out your requirements, constraints, and expectations in detail. It's like conducting a leetcode interview where they always ask you to think out loud — except instead of speaking, you're writing prompts.
The clearer you can articulate your vision in writing, the better the LLM performs at delivering exactly what you want.
Code Reviews in the Age of AI
When my co-founder and I started building Palmier, we both came from backend infrastructure roles — he from LinkedIn's infrastructure team, me from Microsoft's AKS team. Frontend was completely foreign territory.
Cursor became our lifeline, generating massive amounts of TypeScript code that we barely understood. But this created a new problem: code reviews became brutal. Each PR contained 1000+ lines of auto-generated code, and neither of us could easily grasp what the other was trying to accomplish.
We experimented with AI-powered PR bots (eventually building our own), but discovered something interesting: the best approach combined AI-generated descriptions with human-written context. AI excels at describing what changed (and also generating mermaid diagrams, so good!), but humans are still better at explaining why — the bigger picture, the design decisions, the trade-offs considered.
The more thoughtfully you can write about your code changes, the easier it becomes for reviewers to understand the code.
English Is Becoming the Universal Programming Language
The future of software development is already here, and it speaks English. Tools like Lovable, Cursor, and Claude Code are making programming accessible to anyone who can write clearly.
My wife, who's a nurse with zero technical background, recently built a web app using Lovable that calculates cardiac output for patients. It's now used by her entire unit. She accomplished this using nothing but detailed English descriptions.
Here's what struck me: she's actually better at "prompt engineering" than many developers I know. She told me she writes extremely detailed descriptions to Lovable to avoid wasting credits. Sound familiar? That's exactly what we do when working with AI coding tools — except we also read the code afterwards.
Why I'm Learning to Write
This realization got me excited about the future of software development, but it also made me recognize that my writing needs work. English isn't my native language, and I've always focused more on technical skills than communication.
But if writing is becoming the primary interface for software creation, then improving my writing isn't just nice-to-have — it's essential.
Ironically, I used Claude Code to help rewrite this very blog post, applying the same prompt engineering principles I've been discussing. The meta-irony isn't lost on me: I'm using AI to improve my writing about why writing matters in an AI world.
The skills that matter for software engineers are changing rapidly. While technical expertise will always be valuable, the ability to clearly communicate ideas — whether to humans or machines—is becoming the differentiating factor. Those who can write well will be the ones who can harness AI most effectively.
Disclaimer: I don't believe software engineers will be replaced by AI. In fact, building great software with AI requires several human skills: Higher-level technical skills, actual human reasoning, etc.