40 to 1,368: what a year of ai-assisted building taught me

my github contributions went from 40 in 2024 to 1,368 in 2025. that's 3,320% growth.

2025: 1,368 Contributions

2025 felt like earning white and blue belts in building. it was about developing the muscle of idea → code → launch, regardless of outcome.

the catalyst: ai coding tools

i'm not an engineer - i'm a pm at an ev mobility startup. i wanted to build things without waiting for sprint cycles.

i started using cursor around september 2024. my first project was trying to clone mailbrew and make some money - login flows, dashboards, content streams, the works. didn't work out, but i was hooked.

cursor was my main driver until around june 2025, then i switched to claude code and haven't looked back. claude code is what i enjoy using the most - it just fits how i think.

the progression

looking back at my repos, the pattern is clear:

early 2025 - learning projects

jan 2025

agency-match

a directory site at agencymatch.pro. fully automated scraping and templating. actually launched it, but seo is a slow compounding game - not what a first-timer wants.

mar 2025

ugc ai studio

tried copying reelfarm's automated tiktok content generation.

mid 2025 - shipped products

dec 2024 → jul 2025

smartjira

9 months from first commit to chrome store. got rejected multiple times, abandoned it, came back months later with a simpler approach and better models. small but growing - ~4 daily active users, 100% organic from chrome store search.

jun → oct 2025

1 big thing

terminal-style daily goal app. 155 users, all organic from tiktok slideshows. still getting installs months later with zero effort.

late 2025 - tools & open source

aug 2025

finance tracker

personal cli for my own money. built it because i wanted to use claude code to analyze my finances.

oct 2025

dev-ports-tray & klippy

two macos menu bar apps built in rust. dev-ports-tray monitors and kills processes on dev server ports. klippy is a clipboard manager with keyboard shortcuts. both lightweight, native, and open source.

dec 2025

mcp-reddit

open source mcp server. built in 3 hours after seeing a tweet. lets claude browse reddit without api keys.

early 2025 was all vibe coding - accept all, poor documentation, no architecture understanding. by late 2025, proper planning from day one. understanding auth, migrations, component organization. using claude code with skills and mcps. being specific with prompts.

the growth wasn't just in commits. it was learning to ship despite uncertainty, building in public without polish, and finding what works through iteration. most projects went nowhere, but each one taught something.

every new model release makes building easier. problems that blocked me for days become solvable in hours. baseline capability keeps expanding. the gap between "idea" and "shipped" keeps shrinking.

what a time to be alive. thank you ai, long may it continue.