Notes from 6 months of AI transformation
That is another “wow” moment since the ChatGPT days, like from zero to one
After a year of bringing AI into our technology team, we started expanding Agentic AI to every department in the company at the beginning of this year.
This post describes some of my thoughts from watching the changes over the first six months.
Another “wow” moment
Business people, especially the non-IT ones, are the most excited about AI in the company — more passionate than even the IT guys and technical people.
I think it’s common sense. That is another “wow” moment since the ChatGPT days, like from zero to one.
They work 10 hours at work and spend nights and weekends on AI. They send their app at 3 am and have a lot of questions about skills and MCP when they meet me in the morning.
We expected they would love it, but never imagined they would love it this much.
Motivation is the most important thing for change.
IT’s mission is to build innovation platform and data connectors to leverage how business works with AI.
Everyone can code
Across all our Agentic AI topics, coding is the one people picked up and used most effectively.
Nearly 50 apps were built in the first month. They are not simple landing pages but real business applications.
Agentic coding (or vibe coding, as it was called) proves that software will rely more on people who deeply understand the business domain.
I see business people in my company build very complete, well-finished web applications for their job. If a scrum team built that application, I’m sure they would miss some edge cases, but that doesn’t happen when the business builds it.
Today they build their daily work applications, but maybe by the end of this year, they will sit by a developer and do almost all the prompting to AI to code their business applications — the space of software engineers.
We will not need UI review round trips, no need to showcase back and forth.
If that happens, I think software delivery time will go extremely fast.
Coding Agent eating the world
“All in one packaged” agent platforms like Claude Desktop or Kiro IDE are so strong, and stronger day by day.
Last year, we built Python agent applications to loop, handle errors and send prompts to the LLM API, but that is not needed now. By now, all we need is to talk to them.
In my company, more than 80% of our people have an AI platform account to use as a personal assistant.
There is an analyst who uses CLI agent for his research job. He built a dynamic workflow that looks elegant and is much more powerful than I expected.
Prompt engineering matters less and less. What we need now is skills — not because the agent is not smart enough, but because we have “hidden things” still in our heads, and skills are the way to guide AI to follow them.
These thoughts push us to go further with AI transformation, starting with the technology team.
First is the mindset change. Engineers are comfortable with AI coding as a tool. Now they need to treat it as an autonomous agent.
The bottleneck now is people. So the next step is to reduce the human-in-the-loop role of each engineer in their own tasks.