The Year I Started Writing Code, Again.
Sometime this year, I realised that it’s still easy to get into a flow state if I’m writing code. This was surprising because the last time I wrote code for something meaningful was ~5 years ago.
This year was out of my comfort zone to start with.
In January, I transitioned into a business-centric role leading a new vertical, to the extent that I stopped hanging out with tech and product folks. By design, I had the maximum distance from tech teams. I even set my base on a different floor.
Overall, I did a number of things that were a first for me. In the end, it turned out to be a journey filled with growth and improved clarity on what I want to do more of over time.
For example
- First time in my work life, I started owning direct revenue numbers. (and delivered that with a stellar month-on-month growth for the entire year)
- Built a nearly 20-member team that started with me. (While I’ve done a lot of hiring earlier, it was more cross-functional)
- Even role-played a sales-person at times (3-4 out of 10, a long way to go)
My background
Now, on the topic of programming or code, I come from a computer science and software engineering background, so out of modesty, you can assume I have written some code.
I will say, it’s what I did for all my college life.
Sometimes on my own while lurking on GitHub , and at times with talented folks as part of a stellar campus group .
If you use the Go programming language, there must be some 70-80 lines in it, written by me when I was a student.
At work, I started as a full-stack engineer, but then moved into a product role. In recent years, I’ve used this background to my advantage in different scenarios.
For example
- To check if people have thought things through, or if they are double-speaking and just scratching the surface (as any product-person should)
- To befriend the IT head of a large customer, and expedite the due diligence (by subtly showing that I can take notes in Vim and know all the options for the curl command)
When code felt like magic
When writing this article, I realised that I still remember some of the instances when I wrote code for something meaningful.
First Day at Work
On my first day at work, I was asked if I had ever built a Web SDK. While the answer at that time was “no,” it took me a few days to get to a “yes.” We used that till our pivot from chat to voice.
Beating Google
One fine night, we decided to crowdsource our efforts to collect Indian speech data, instead of manually collecting it.
After 2-3 days of combined efforts, we had an Android app that was scaled to pan-India across 10 languages, and became the foundation that helped us build a better Indic Speech-to-Text engine than Google (circa 2017).
Building a Bot
We used to have an if-else statement-based conversational-bot (circa 2016), and “context” was not something it would respect.
I remember designing and building an orchestrator system with a friend, with a Finite-State Machine (FSM) at its core. We still have some legacy bots that are running on that system.
And, many other such instances like this, but let’s stop with the self-patting.
Falling out of love
After that, I stopped writing code. I vaguely remember opening my terminal.
It felt daunting to even make changes to my website after I lost the local/dev setup, and I wasn’t motivated enough to set it up on the newer systems.
That was another reason I kept adding to a list of articles/blogs I wanted to write, or projects I wanted to make, but never found the time, or should I say, never had the intrinsic motivation to act on that list.
In late 2023, I remember flirting with no-code while teaching “software development” to my cousin brother during his semester break.
We spun up a FlutterFlow instance and pushed through to build SoundSafe, a semi-successful side-project, but overall a failed attempt at creating a safety companion for women.
At best, we were featured in the recap video created by FlutterFlow for their FFDC Hackathon.
It’s safe to say that my love affair with programming was fading, and it was slowly becoming a vestigial skill.
A late-night revival
Around July, my team faced a technical issue regarding a large client. We had no other option than to wait for the relevant technical bandwidth to get assigned from a floor above us.
The clock was ticking, and delay would mean less revenue for the month.
So, I rolled up my sleeves, sat with the folks who were closest to the problem, and started asking ChatGPT to write the appropriate code in Python. (Codex was still in research preview; I had not used Claude Code or Cursor)
After a lot of copy-pasting between ChatGPT and the terminal running the scripts, we were able to make it work.
But there was one more positive impact from that late-night session.
Many team members started using ChatGPT for Python code instead of using it for a text-in/text-out mechanism. Some of them even picked up other tools and started building dashboards or single-page applications instead of Excel or Google Sheets.
Enter Claude Code
All of that motivated me enough, and one fine weekend in August, I decided it was time to get myself familiar with Claude Code. Everyone on HN or X/Twitter was talking about it.
People had already gone to MCPs, and I wasn’t even done with Claude Code yet; not good.
I wasn’t aiming for anything specific and had zero expectations; I just wanted to experience it.
After using it for an hour, I was amazed by its simplicity and capability.
Tweet
One of the key observations was how little I had to intervene when compared to the number of lines written by it. If it were a real developer writing the same amount of [working] code, it would need at least two meetings and one tea break.
Overall, it felt like test-driving a fancy hyped car where the drive goes super smooth, and you ask them to let you drive for 5 more kilometres (or miles, whatever).
I was ready to book the car by the end of it.
What it means now
Over the years, I’ve always nodded to the “you are not your job” preach, but haven’t done it justice due to the demanding needs of working at a startup.
At the same time, I strongly believe that experimenting with life, be it a side-project or a new habit or skill, increases the odds of your success at any venture.
Now, it feels like we have more time at hand, because our earlier norm of the time delta between thinking and execution was too high. Shipping faster used to be a team or company-level moat, which is now democratised if you use these tools in the right way.
I’m happy that I was able to strike off an idea I’ve had on my “Ideas” note for a few years, Plug That In , and many other folks have found that helpful. I hope to do that more often.
—
It’s now clear that you need a distinctive taste or quality bar of your own. If not, AI will only amplify your bad taste, and you might just converge to the mean. (How not to use AI at work? )
There is a lot of talk about getting your work done in one shot. I personally want to remain cautious and develop stronger opinions before I get a lifetime pass to that train.
For personal software experiments, I tend to have extended Q&A sessions with Claude-Code before granting it full file-edit access, but I’m on the lookout for any best practices that help me maintain the quality and attention to detail.
—
After working on AI products for knowledge workers, I’ve learned that you add diminishing returns to their daily lives if your product takes up more space than what they’re used to. (Call centre agents already see 3 screens; if you ask them to look at a 4th screen, there will be blood)
Kudos to its simplicity, Claude-Code has already been abstracted to a minimal level, which was familiar to most of the developers, a terminal, which is closest to other tools they need for their craft.
We’ve been hearing this-is-the-year-of-agents, but getting tools like Claude-Code to a similar abstraction level for non-programmers will be the dawn of a real year-of-agents.]
Overall, everything is proceeding as I have foreseen; it feels good.