Agentic Systems / 12 min read
The Agentic Prototype Is a Lie
Why your AI agent demo works perfectly and your production system doesn't — and the six engineering gaps that explain it.
Read articleInsights
Field notes from building agentic systems and shipping product work.
Practical writing on the engineering decisions that separate useful systems from demo-stage theatre.
AI Product Engineering / 10 min read
Your 6-Month AI Pilot Should Take 4 Weeks
Most AI pilots fail from scope creep, not technical limitations. Here is the 4-week framework for reaching a real go/no-go decision fast.
Read articlePlatform Engineering / 9 min read
How We Cut Cloud Infrastructure Cost by 75%
How CoEdify reduced monthly Azure spend at EquippedAI from Rs40 lakh to Rs10 lakh — and what architectural decisions actually drove the saving.
Read articleAgentic Systems / 8 min read
Your Coding Agent Can't Refactor Legacy Code — Here's Why
Coding agents don't fail on legacy code because the model is weak. They fail because the code has no safety net.
Read articleAgentic Systems / 8 min read
The Context Window Problem Nobody Talks About
A larger context window gives the model more text to read. It does not give your agent a memory. That distinction breaks most production systems.
Read articleAgentic Systems / 9 min read
How to Evaluate Agent Quality Without a QA Team
You don't need a QA department to evaluate agents well. You need deterministic checks, rubric-based review, and regression testing on every meaningful change.
Read articleAI Product Engineering / 10 min read
If Your Agentic MVP Takes More Than a Month, You're Solving the Wrong Problem
The bottleneck in most agentic MVPs is not the agent — it is the missing infrastructure layer underneath it.
Read articleAgentic Systems / 9 min read
Multi-Agent Coordination Without Framework Lock-In
The problem with framework-first multi-agent design is not the frameworks — it is what happens when they become the system of record.
Read articleAgentic Systems / 9 min read
Why Your RAG Pipeline Keeps Disappointing
Most RAG systems disappoint in production for the same reason: they flatten reference knowledge, live state, and accumulated memory into one retrieval layer.
Read articleAI Product Engineering / 8 min read
The Spec Is the Prompt: Why Vibe Coding Breaks at Month 3
Coding agents are fast. Underspecified intent is expensive. For production work, the spec is the real prompt — not the one-liner you type into the chat.
Read article