FULL-STACK · MARYLAND · CURRENTLY @ SYNACK

Pavan Gajula.

Five years building enterprise systems in Java and React.
Two years shipping websites for the people who actually need them.
Lately, I keep noticing the same thing — and I want to study it seriously.

YEARS BUILDING
5+
PRODUCTION SCALE
10K+ users
VOLUNTEER HOURS
450+
BASED IN
Maryland, US
  CURRENTLY

Building Spring Boot microservices at Synack by day. Volunteering for a Howard County youth sports nonprofit by night. Writing about what breaks when developers and AI write code together.

01  /  what I'm interested in
01 · HUMAN-AI INTERACTION
Why do I trust AI more in Java than in bash?
My skepticism rises and falls with how well I know the language. The places I get burned are exactly the places I'm not paying close attention — which is exactly where AI tools are most useful, and most risky.
02 · DEVELOPER TOOLS
What would a tool that helped me not accept code look like?
Most AI coding tools are designed around the accept button. The harder design problem is making it cheap and natural to push back, edit, or reject — especially when accepting feels easiest.
03 · PROGRAMMER EXPERIENCE
What happens to learning when AI writes the boilerplate?
Boilerplate is how junior developers used to learn the shape of a system. When the AI writes it for them, the path is faster — but the path through the path is missing. I'm not sure yet what that costs.
03  /  what I keep noticing

The more developers lean on AI coding tools, the less the problem is writing code, and the more it's knowing when to trust what comes back.

I write code with AI assistants every day. Whether they can produce working code stopped being the interesting question — they usually can. The harder one is when a developer should trust the output, and what happens when checking it costs more than writing it would have.

04  /  notes
NO. 01  ·  HUMAN-AI TRUSTNOV 2025

The asymmetry of developer skepticism

When I'm writing core business logic in Spring Boot, I use Claude Code like an extension of my own brain — I anticipate its suggestions, accept them instantly, and rarely get burned because the patterns are deeply ingrained in memory. Last month I had to switch contexts to a complex bash script for a deployment pipeline, where my syntax knowledge is rustier, and I caught myself passively accepting lines I couldn't instantly verify. It passed a flag that didn't exist in that version of the CLI, silently breaking the build downstream.

The realization: my trust in AI tools isn't a function of the AI's output quality — it's a function of my own confidence in the language. When my own knowledge drops, my baseline skepticism drops with it. Which is exactly when silent failures slip through. The interesting design problem isn't making the AI more accurate. It's making developers more skeptical in their weak spots, not less.

•   more notes coming. monthly cadence.
05  /  experience
JUL 2024 — NOW
Java Full Stack Developer
Synack Solutions LLC · Maryland, US
PRESENT
JAN 2023 — MAY 2024
M.S. Software Engineering
UMBC · Maryland, US · GPA 3.53
EDUCATION
JAN 2021 — JAN 2023
Java Full Stack Developer
Tata Consultancy Services · Hyderabad, India
INDUSTRY