Essay · AI Stack
Agents vs Skills vs MCP
A pragmatic framework for deciding how orchestration, knowledge, and integration layers work together in 2025, including real architecture diagrams and ready-to-use configurations.
Sowmya Raghunathan
Engineering & Architecture Leadership
I build large software systems and spend most days examining how technology, people, and incentives collide in the messy real world. I care about clarity, first principles, strong design patterns, and decisions that hold up under real-world pressure. I’m a builder by nature, a mentor by habit, and skeptical of buzzwords and performative certainty.
Outside of work, I’m a mother, a lifelong learner, and someone constantly balancing ambition, family, health, and meaning. I run small experiments in how I learn, work, and live, and I write to make sense of those experiences.
This website is where I write - articles, essays, blogs, and short reflections on software architecture, leadership, AI, modernization, systems thinking, execution, learning, and the human side of building things. Some pieces are polished. Some are exploratory. All are written to think clearly, not to market or perform.
Recent & Featured
No infinite scroll. Each piece stands on its own. Come in, read, and leave with a clearer mental model.
Essay · AI Stack
A pragmatic framework for deciding how orchestration, knowledge, and integration layers work together in 2025, including real architecture diagrams and ready-to-use configurations.
Framework · Leadership
How I run the first 72 hours of any modernization effort - problem definition, success metrics, ROI, and risk before a single line of code is written.
Essay · Context Systems
Why the best AI practitioners act like context architects - designing reusable instruction stacks instead of chasing clever prompts.
Guide · Claude Code
Step-by-step instructions for writing reusable SKILL.md files, optional folder structure, and the five skills worth building this week.
Long-form
Structured arguments, annotated diagrams, and decisions tested against messy constraints.
AI Stack
Field notes from building AI-assisted development workflows, with decision criteria, architecture layers, and templates ready for production systems.
Frameworks
A step-by-step playbook for defining business context, success metrics, investment theses, and risks before teams start shipping.
Context Systems
The mindset shift from clever prompts to reusable context stacks that make AI work trustworthy and compounding.
A no-nonsense walkthrough for capturing your preferences in SKILL.md files so Claude stays consistent.
Short notes
Building a scorecard for architectural cleanups so leadership sees compounding value instead of invisible toil.
Teams want to skip the context work, so we run short daily standups during Phase 0: alignment, metrics, blockers, sign-off. Keeps the urgency without losing rigor.
Replaced “prompt reviews” with “context reviews.” Once a month we prune anchors, templates, and examples - treating them like code.
Themes
Design patterns, integration scars, modernization playbooks.
Trust, incentives, meetings that matter, unglamorous execution.
Pairing with copilots, evaluation criteria, production guardrails.
Small bets, reflection prompts, routines that survive real life.
Balancing care work with leadership without pretending it’s easy.
Slow progress logs, what I’m reading, how I reset.
Archive