17/10/2026
A Developer’s Journey to Better Software.
I started building websites more than ten years ago with PHP, MySQL and a lot of curiosity. Back then it was WordPress themes, Joomla extensions, shop plugins and the odd Facebook app. Every project taught me the same lesson: good software is less about the code you write and more about the problem you understand.
Over the years the problems grew. I integrated payments with Stripe, PayPal and Authorize.net, connected systems through REST and SOAP APIs, spent two years on Oracle Service Cloud, and led teams to ship on deadline. Laravel became my home, and TypeScript and React joined it later.
Then AI coding tools arrived, and I saw two ways to use them. One was autocomplete: faster typing, same habits. The other was to rethink how software gets delivered. I chose the second.
Today I build agentic workflows. On Claude Code I designed a spec-driven process where reusable AI skills carry a ticket from scope and plan through implementation, self-review, PR review, tests, QA notes and archive. Agents review code against our standards, fix PR comments, write browser smoke tests and dig into production issues with real data. They reach GitHub and issue tracking through MCP, and each task goes to the model that suits it, from Claude to local LLMs on Ollama.
One rule never changes: agents propose, people approve. The team still owns every line that ships.
The results showed up in my own pull requests. Reviewer change requests dropped by about 70% per 1,000 lines changed, even as my PRs grew about 3.7× larger.
What hasn't changed is how I work: clean code that follows the standard, careful reviews, clear documentation and a calm head when deadlines get close.
If you're exploring agentic AI or developer tooling, or just want to swap notes on shipping better software, I'd be glad to talk.
- sanjoypaul.com