Rolling Out AI to a Two-Speed Team
Half my team adopted AI within months. The other half kept working as before. The Skeptic Conversion Matrix and a 4-week pilot for the slow half.
Articles, tutorials, and thoughts on web development, tech, and building cool things.
Half my team adopted AI within months. The other half kept working as before. The Skeptic Conversion Matrix and a 4-week pilot for the slow half.
Six months moving my engineering work onto Claude Code agents: merged PRs up 10x, cycle time down 65%, what I built, what broke, and what I'd do again.
The RBI raised the repo rate to 5.50% on 7 October 2026, yet SBI's 5-year FD still pays less than its 2-year one. The math on waiting versus locking in.
Revenue per employee rose 3-4% at India's top IT firms in FY26 after two flat decades. What caused it, and what the first half of FY27 shows so far.
TCS cut 23,000 jobs, Infosys hired 20,000 freshers. Updated with Q2 FY27 results: both now grow headcount, but they still bet differently on AI.
AI made writing code far faster, but cycle time barely moved. Amdahl's Law, DORA data and my own numbers on why, and where to aim AI to speed up delivery.
Enterprises run about 12 AI agents and half work in silos. Why connecting them is a distributed-systems problem, and the order to fix it in, from my setup.
Three tools on this site shipped with real bugs: a calculator that lied, a leaked API key, a converter broken on every input. How I found and fixed them.
Coding agents read your repo, run commands and push code. Real 2026 incidents: prompt injection, leaked secrets, slopsquatting, plus a 12-point checklist.
Build a working MCP server with the v2 TypeScript SDK: tools, resources, prompts, stdio and HTTP transports, Inspector testing and Claude Code setup.
AI token prices fell about 280x in two years, yet compute spend keeps setting records. Why cheaper intelligence means bigger bills, seen from my own usage.
A build log from two browser games on this site: Phaser scene architecture, React lifecycle bugs AI could not fix, and what a deterministic sim taught me.
The Market Mood Index on my SIP/FD calculator is three inputs and a weighted average. The exact arithmetic, worked by hand, and the things it cannot see.
Three agent failures from my own team's rollout, the four ways agents break in production, the fixes that hold, and a 10-item readiness checklist to copy.
Most SIP vs FD comparisons use pre-tax numbers. Taxing both correctly widens the gap instead of narrowing it. Worked math at the 20% and 30% tax slabs.
The 2023 tax change ended indexation on debt funds. How PPF, NSC, SCSS and SSY compare to debt funds now, at the official Oct-Dec 2026 government rates.
Agentic AI isn't killing IT services, it's killing the slack. Q2 FY27 data on the margin squeeze at TCS, Infosys and Wipro, and the pivot that works.
Indian IT after Q2 FY27: TCS, Infosys and HCLTech are hiring again while Wipro shed 6,716. What the data means for your job, bench risk and your next move.
EU AI Act deadlines after the Digital Omnibus: what applied on 2 Aug 2026, what lands in Dec 2026, Dec 2027 and Aug 2028, plus a developer checklist.
Why a second model catches what the first misses: panel, judge and generate-then-refute patterns, the review agents I run, and working Python code.
Real numbers from 400 hours of agent work: 86% of sub-agent work on cheaper models, a proxy that cut tokens 34%, and the cost rules I'd give any team.
AI coding feels faster than it is. How to measure real ROI with DORA and SPACE, with my own before-and-after numbers and the confounders I found.
Will AI replace developers? An engineering manager maps which tasks AI automates, what the 2026 hiring data shows, and the moves that keep you valuable.
Untangling Indian IT's manager titles: Project, Program, Product, Delivery and Engineering Manager. What an EM actually does, and how AI changes the job.
How my AI job risk calculator scores 45 roles, including Indian IT ones like PM, BA and Scrum Master: every point it adds or removes, and its limits.
Install Claude Code in 5 minutes, then build a first project with plan mode, CLAUDE.md and the commands that matter. Checked against November 2026 docs.
How I built this site with Claude Code: about 70 posts, nine tools, Money Pulse and FamilyRoots. What AI did well, where it failed, what I'd change.
How I write prompts for code, reviews and analysis: six elements, before and after rewrites, what reasoning models changed, and the mistakes to avoid.
I built this site with AI writing about 90% of the code, then found three broken tools in an audit. The vibe coding workflow I use now, and what changed.
Where rubber duck debugging comes from, the research on why explaining aloud works, a step-by-step method, and how to use it when the duck is an AI chat.
AI jobs in India without the hype: skills that pay, what ML and GenAI engineers earn, and who is hiring, from an engineering manager who hires.
Build a research agent in about 180 lines of Python on the Claude API: web search, saved notes, memory, and the tool-use loop, explained step by step.
Claude Code's creator said Claude wrote all of his code for a month. What that claim did and didn't mean, set against my own six months of real numbers.
Copilot vs Cursor vs Devin Desktop (ex-Windsurf) vs Claude Code, rechecked in November 2026: plan prices, models, agent features and which one to pick.