(Illustration: Taking the day off. Happy Lunar New Year! Image source: Business Next.)
✳️ Coding is easy, Context is hard
I recently recorded an episode of Podcast “Digital Keywords” EP228 with James from Business Next, discussing a core theme I kept chewing on throughout my 2025 year-in-review.
AI has crossed the threshold in writing code. It is no longer a small assistant that auto-completes one line at a time. Take Claude Code for example: it reads through your entire project directory on its own, understands cross-file context and dependencies, plans how to coordinate changes, even dispatches sub-agents to handle different tasks in parallel, and then delivers an entire feature in one go. More importantly, it has memory. You write your project specs, goals, and coding style into a file, and every time it starts working, it reads that file first. No need to re-explain everything from scratch.
From Cline and RooCode, to Cursor, to Claude Code and Kiro, my team and I have walked the entire tool evolution path together, feeling the role shift at every step. Humans can no longer compete with AI on speed. The bottleneck has moved from “think fast, code fast, iterate fast” to “can you articulate your requirements, context, and intent clearly?” I joked with my team: we all need to start practicing how to communicate well. This is not just an observation. It is the firmest conclusion I reached after being pushed through all of 2025, and the central theme of this podcast episode.
✳️ From Cooking Yourself to Running the Kitchen
In the episode, I used a cooking analogy to break down the differences across three stages.
- Traditional coding is like cooking by yourself: grocery shopping, prepping ingredients, seasoning, washing dishes, all on you. Then GitHub Copilot came along, like having a helper who chops vegetables and preps ingredients. It guesses what you need next and gets it ready, but you are still the one driving the cooking process. It just helps a little on the side.
- Vibe Coding is like telling a skilled chef “give me a bowl of Taiwanese spicy beef noodle soup, not too oily.” What comes out might be close to what you want, or slightly off. You taste it, give feedback, and the chef adjusts. Back and forth, just like prompting AI in a conversation.
- Agentic Coding takes it to an entirely different level. You are the restaurant owner, telling the entire kitchen team “we are launching a new set menu this month.” Someone develops the menu, someone handles procurement and sources ingredients. They research what is in season, what cost structure makes sense. Someone does tastings and plating. They coordinate and divide the work among themselves. You only step in to make decisions at key checkpoints.
From cooking by yourself to running the entire kitchen, the role change is not an upgrade. It is a qualitative shift. Where to set the quality bar, which business model to choose: these are decisions humans cannot delegate. It is not about being better at writing code than AI, but about designing the system that gets things done right. Right now, dishes keep coming out of the kitchen so fast that we can barely keep up with tasting them.
✳️ Slow Down to Go Fast
What Vibe Coding produces looks like it works, but nobody knows why it works. It is like building a flashy sheet-metal house: looks impressive from the outside, but cannot survive a typhoon. Every modification breaks something somewhere else, and costs explode. That is the pit we fell into during the first half of last year.
The Spec-Driven Development approach requires defining your intent clearly in human language before writing a single line of code. Back in Q1 when we were using Cline, we had to explicitly write in the prompt “please don’t write code yet, discuss with me first.” By the time we moved to Kiro, that instruction was no longer necessary. Kiro formalized the spec-writing process, helping you break down requirements documents, design documents, and task lists layer by layer, so both humans and AI think things through before getting to work.
And writing specs is not just the engineer's job. Product managers and executives should all sit down together and write clear requirements. It is like drawing architectural plans before renovating a house. Better to go back and forth on paper than to tear down walls and re-route wiring after construction has started. The theme of my 2025 year-in-review was “Slow Down to Go Fast.” Deliberately investing time upfront to think things through makes the subsequent execution converge, quality stabilize, and output become predictable. The time spent writing specs is no less than the time spent coding, but every bit of that investment pays off with better results. It is not slow because you are idle. It is slow on purpose. AI will walk alongside us in this process. It is the most patient partner (right?
✳️ Start by Learning to Communicate Well
Whether you are an engineer, a product manager, an executive, or a marketer, now is the best time to start collaborating with AI Agents. Do not be afraid of the technology. Just start by communicating well and writing your requirements clearly. That is already a great first step. Coding is no longer the exclusive domain of engineers. Our own domain expertise is our superpower.
This podcast episode was our first attempt at a full one-hour conversation (three hours next time?). We covered the journey from Coding to Vibe Coding to Agentic Coding, how Claude Code’s sub-agents divide work, how Kiro’s intent-driven architecture works, the real-world story of helping a friend use AI to transform their ERP system, and how enterprises should pick the right problem rather than the right tool when adopting AI. Plenty of pitfalls we stepped in and lessons we figured out are all in there. Highly recommended. Tools will keep changing, but learning to communicate well never goes out of style. Let us practice communicating well together. (And remember to exercise too 😛