Writing
Build logs, essays, and notes on agentic AI, Indian retail, and what I'm learning along the way.
Security Practices and Tools in the Age of LLMs
How to actually secure LLM and agentic applications in production. Why the model isn't your attack surface, the one rule that predicts agent breaches, and the practices and tools that hold up.
Testing Practices in the Age of Agents
A deep, code-first guide to testing LLM and agentic flows: contract tests, hermetic mocking, statistical gates, LLM-as-judge, multi-turn simulation, and the CI setup that ties it together.
Inside the Claude 4.7 System Card
A practitioner's reading guide to the 200+ page Anthropic document almost no one reads in full. What the launch post hides, where the load-bearing safety numbers live.
Inside the Mythos System Card
Anthropic published a 245-page system card for a model almost nobody can use. Here's why it's the most important Anthropic document of 2026 to read carefully.
How a Diffusion Model Works: A Practitioner's Read of the 2026 Image Stack
Modern image models aren't U-Nets running 50 denoising steps. They're transformers running 4 steps of a straight-line flow. Once that lands, every product surface starts making sense.
How a Vision LLM Works: A Practitioner's Read of the 2026 Multimodal Stack
Vision LLMs don't see images. They tokenize them. Once that lands, the cost, the failure modes, and the design space all fall out cleanly.
How a Video Model Works: A Practitioner's Read of Veo 3.1 and Seedance 2.0
Modern video models aren't image models in a loop. They're diffusion transformers on spatiotemporal patches, with audio now riding the same train.
What Manchester City Taught Me About Building Systems
A lifelong City fan on what watching Guardiola's teams build, break, and rebuild has taught me about engineering systems that actually last.
What Fashion Manufacturing Taught Me About AI's Blind Spots
AI fails wherever the cost of being wrong is high, feedback is slow, and the decision is irreversible. Fashion manufacturing reveals the pattern clearly.
Prediction Is a Crutch for Slow Systems
Retailers spend millions on demand forecasting that's wrong 30-40% of the time. The better investment: systems fast enough to make prediction unnecessary.
The 2,000-Store Problem: Why Point Solutions Collapse at Scale
AI works beautifully for one store. At 2,000+ stores across 15 channels, every point solution breaks. What works is system-level architecture, not better individual models.
Why Retail AI Fails at the Vocabulary Layer
Before you can forecast, recommend, or allocate anything across brands and channels, you need a shared vocabulary. This is a context engineering problem wearing a retail costume.
An Engineering Org of One: Software Engineering in 2026 and Beyond
The minimum viable team for shipping production software has collapsed to one person. Not because the work disappeared, but because it got encoded into a software factory.
Making Design AI Native: From Pixel Perfection to System Thinking
Part 3 of 3. How designers transform their practice when AI is doing the building: design system enforcement, accessibility automation, UX audits at scale, and the anti-patterns that erode craft.
Making Product AI Native: From Ticket Writers to Decision Architects
Part 2 of 3. How the PM role transforms when engineering is AI-native: scope docs over tickets, prototyping with AI, and the anti-patterns that kill product thinking.
Making Engineering AI Native: A Playbook for the Transition
Part 1 of 3. The techniques, adoption sequence, and anti-patterns for transforming an engineering team from traditional development to AI-native workflows.
GTM 101 for EPD: Why Distribution Is the New Bottleneck
As AI makes building faster, the constraint shifts to distribution. Here's the GTM literacy every engineering, product, and design leader needs: and how it changes what you build.
Skills 101: Building Your First Claude Code Skill
A step-by-step guide to building a Claude Code skill from scratch, using a meeting notes pipeline as the example: Granola transcript in, Slack-ready notes out.
Skills Are the New Org Chart: Agentic Engineering with Claude Skills (Part 1)
EPD orgs were built around vertical skillsets: backend, frontend, QA, design. Each of those can now be encoded as a skill in your repo. Here's what that means for how we build and organize.
Evals in Practice: Agents, Retail, and Multimodal Pipelines
Worked examples of evaluating AI agents: trajectory evals for tool-use, catalog enrichment scoring, customer support resolution, and multimodal pipeline evals.
Evals for AI Agents: What They Are and How to Build Them
A technical guide to AI evaluations: what evals are, why agents need them, the three scorer types, and how to build an eval suite from scratch in Python.
Spec-Driven Engineering: How to Write Specs That Agents (and Humans) Can Build From
The practices that make a spec actually useful. 10 patterns from writing dozens of specs for production systems, with real examples and a framework for what every spec should contain.
Agentic Engineering with Claude Code: What It Actually Looks Like
A build log of how I use Claude Code to ship production software. Spec-first workflows, living documentation, multi-agent sprints, and the real numbers behind the productivity gains.
Context Engineering > Prompt Engineering: Patterns That Actually Work
The techniques that produce decision-quality output from LLMs have nothing to do with clever phrasing. They're about structuring context. Here are the patterns I use every day.