Notes from writing software in 2025

A year full of lessons and learnings

2025 just ended and I want to talk about some things I learned last year as a software developer.

And some things I’m carrying forward into 2026.

I hope some of this is useful to you too. So, let's start.

1. Writing code became… optional?

At some point in 2025 I realized something weird:

I almost never write code anymore.

Most of my code is written by AI agents now. Cursor agents. Claude Code. Whatever tool is good that week.

And no, I’m not saying I became a “prompt engineer” or a “10x engineer”.

I’m saying: the bottleneck moved.

Before 2025, my bottleneck (and maybe yours too) was obvious:

  • writing code took time
  • building a product took time
  • shipping took time

But in 2025… it changed so drastically that I started feeling like:

I can build almost anything I want to build.

And when you get that kind of power, the question changes.

It’s no longer: “can I build it?”

It’s: “should I build it?”

2. “Knowing what to build” became the real skill

Here’s the trap.

When MVPs become cheap, it becomes insanely easy to waste your life hopping between projects like a hyperactive squirrel.

One week it’s:

“bro what if I build this…”

Next week it’s:

“wait wait that idea is better…”

And then:

“okay I’ll just ship both.”

And suddenly you’re 6 months older with 9 half-built products and 0 belief in any of them.

So 2025 taught me this:

With AI, execution is cheap. Belief is expensive.

You need to first pick something you actually care about. Something you’ll stick with even after the initial dopamine wears off.

For me, the “good” ideas were not world-changing. They were just… meaningful to me.

Some examples from my year:

  • Lumi: a to-do / day-planner type app I genuinely use.
  • StocksBrew: an AI agent that tracks stock news and gives summaries before market open.
  • Instafy: a tool I didn’t need daily, but I saw a small market moment and went for it.

And one of these gave me my first ever internet dollar.

That sounds tiny (and it is) but it changed something in my brain.

It proved that shipping small things consistently is not “random side projects”.

It’s reps.

3. Your personal brand is now non-negotiable

I’ve been saying this for a while and 2025 just screamed it in my face:

It’s more important than ever to build a personal brand.

Because now… everyone can ship.

Cursor exists. Agents exist. Cloud whatever exists. 18 year olds are launching products every weekend.

And you’ll see some launch and think:

“bruh that idea isn’t even revolutionary. I could’ve built that.”

Yes. You could have.

But here’s the part most devs ignore:

Distribution is a feature.

If you don’t have a list of people who trust you / like your work / want to see you win… your product launches will feel like shouting into the void.

In 2025 I leaned harder into posting. YouTube, Twitter, LinkedIn, my website — all of it.

I crossed 2,000 YouTube subscribers last year. Thank you btw.

And then something funny happened.

One of my experiments I posted on my site, I tweeted about it, it popped off…

…and then Google reposted it.

Butterfly effect is real. Small things snowball.

And suddenly I started getting collaboration opportunities from people I thought would never even notice me.

Not because I became smarter overnight.

But because I showed up consistently.

So in 2026, I’m investing even more into my brand. Not in a cringe way.

In a “I want more doors to open” way.

If you’re not posting your work online, please start. Even if it’s messy. Even if it’s small. Even if it’s “just a side project”.

Stop waiting. Start building. Start posting.

4. AI helped me go deep (not just ship fast)

This was a big one for me.

Before 2025, I wanted to do deeper stuff:

  • research papers
  • understanding LLMs
  • building a transformer
  • fine-tuning models
  • messing with vision
  • doing weird experiments

But if I’m being honest…

I never had the guts.

Or I always felt like I needed hand-holding. Or I’d get stuck on some concept and then my motivation would die.

In 2025, AI became the hand-holding.

And because of that, I actually did the things I kept postponing for years.

I built my own transformer model. I fine-tuned a model on a custom dataset. I fine-tuned NanoChat. I added vision to NanoChat (and that got a lot of attention online).

And I also started reading more papers.

Not because papers became easier.

But because now I could sit with ChatGPT/Claude and ask:

  • “what does this term mean?”
  • “explain this figure like I’m 5”
  • “okay now explain it like I’m not 5”
  • “what should I Google next?”

So yeah, AI didn’t just make me faster.

It made me braver.

And I want to keep doing that in 2026: Use AI to ship, yes.

But also use it to learn aggressively.

okay so what am I carrying into 2026?

If I had to compress 2025 into a few rules:

  • Build with belief: ideas are cheap now. pick something you actually care about.
  • Brand is leverage: ship in public. post. collect trust over time.
  • Go deep with AI: use it as your tutor, not just your coder.

That’s it.

Now your turn:

What did you learn in 2025? What are you carrying into 2026?

I’d genuinely love to know.

Chalo, bye!