writing
-
The pleasure of being bad at something
On learning drums, a year of fitness, and finding room for effort and flow as coding agents get better at their work.
-
The Age of Software Factories
Software factories can form teams, discover workflows, build missing capabilities, and adapt their approach as they turn complex problems into production systems.
-
Common Obligations
Six shared obligations for AI builders, and a proposal for making them hold across companies and borders without putting the largest labs in charge.
-
When Outputs Lie
Your AI agent's outputs look composed.
-
Agent Framework Generations
Digging through five generations of agent frameworks, from raw API loops to federated swarms.
-
Overnight Agents: Crunching Tokens
I launched an orchestrator that managed 7 Claude Code peers across repos simultaneously.
-
Agent Drift in AI Systems
Agent drift is when an AI agent slowly strays from its original goal over a long run.
-
The AI Hourglass
The AI stack is shaped like an hourglass — wide at energy and applications, pinched at chips and GPUs.
-
How to Build AI Agents: Inside Out
Build AI agents from the inside out — start with behavior in CLAUDE.md, add capabilities via MCP, then wrap in code with the Claude Agent SDK.
-
Repowire: Multi-Agent Coding Workflow Across Repos
Repowire is a pull-based mesh network that lets Claude Code sessions communicate across repositories in real-time, replacing stale docs with live agent-to-agent queries.
-
The Vibe Bottleneck: AI Coding Workflow Across Multi-Repo
Multi-repo coding agent workflows turn you into the message bus.
-
Agentic Cloud Infrastructure: Beyond 12-Factor Apps
Agents are stateful logic, not stateless apps.
-
AI Agent Memory Management With Databases
AI agent memory management mirrors database design.
-
AI and the Future of Programming: Architects of Intent
Code is becoming a transient artifact — AI-generated bytecode.
-
Ambient AI
Torale monitors the web so you don't have to.
-
ML Infrastructure Strategy: Build for Momentum
Early architectural decisions create a flywheel effect that accelerates your path to production.
-
Machine Learning Project Template with Modern Stack
A modality-aware machine learning project template that gets you from idea to training in minutes using uv, Polars, and LitServe for modern ML workflows.
-
Claude Code Productivity: I Shipped More, Coded Less
A month with Claude Code: I shipped more projects than ever while paradoxically becoming worse at programming from scratch.
-
A Better ML Training Workflow: Stop Training and Praying
Fix your ML training workflow with the Pre-Flight, In-Flight, Post-Flight framework—catch bugs early, monitor runs, and evaluate beyond vanity metrics.
-
Agentic MLOps: AI Agents Automating Your ML Stack
Explore how agentic MLOps uses AI agents and MCP to autonomously handle data drift, model deployment, and ML pipeline orchestration across your entire stack.
-
The A2A Protocol
Learn how the A2A Protocol enables secure AI agent communication.
-
Tech Trends Next Decade: A 6-Shift Framework
Six shifts define tech trends next decade: Autonomous Economy, Radical Abstraction, Composability, Ecosystem Dynamics, Ambient Intelligence, Next-Gen UX.
-
AI Agent Frameworks
A practical comparison of AI agent frameworks — OpenAI Agents SDK, Google ADK, AWS Bedrock, and Smol Agents — covering architecture, multi-agent systems, and emerging trends.
-
ML Experiment Tracking as First-Class Infrastructure
Treating ML experiments as disposable runs undermines reproducibility and model quality.
-
Modern ML Training Infrastructure
Scale ML models faster with the right training infrastructure.
-
MLOps for Startups: A Decision Framework
A decision framework for when to invest in MLOps at your startup — from early-stage flexibility through scaling infrastructure at each growth stage.
-
Building Tech Products for India: A First-Principles Playbook
Building tech products for India from first principles—local habits, scale, price sensitivity, and cultural diversity require a full ground-up rethink.
-
Rise of the Full Stack Machine Learning Engineer
How the histories of full stack web development and machine learning converged to create the full stack machine learning engineer role.
-
ML Ecosystem Fragmentation (It Ain't JavaScript)
The ML ecosystem mirrors JavaScript's 2015 framework chaos.
-
Pipenv vs Pyenv: Python Version Management
Compare pipenv vs pyenv and learn how combining both tools solves Python version management and dependency isolation in one workflow.