Writing
Longer pieces on how the work actually goes: what broke, what held, and what I would do again.
Small‑Model Hierarchy Beats the Giant AgentWhy a stack of small, specialized models beats a single large LLM for agentic work, and how to build one today.aiThe Scaling Trap: Why Agentic Systems Need Small Language ModelsWhy one frontier model running every step of an agent loop breaks on cost and reliability, and how to build a heterogeneous system where a large model plans and small language models execute.aiSpec-Driven Development: Why Vibe Coding Is a Debt Trap for Agent TeamsVibe coding makes one developer faster and the team slower. Spec-driven development makes the specification the contract that governs coding agents. What that looks like in a repository, and where it still falls short.aiBuild the factory, not the appOne machine at home runs a pipeline of AI coding agents that plans, builds, tests, reviews and documents whole projects on its own, and I steer it from my phone. Here is the shape of the machine, what each project actually cost, and the places it still leaks.ai agentsAgent Permissions Are the New Attack Surface: Least Privilege, MCP and Memory PoisoningWhy least privilege is harder for AI agents than for people, how MCP servers and poisoned memory turn one compromise into many, and the identity, token and tool-call controls that contain the damage.aiModel Routing for Agents: A Three-Tier Router Built on Complexity, Latency and CostA lightweight orchestrator that picks a model per request by complexity, escalates on sustained failure and stops runaway loops, with runnable Python and the numbers to check it against.aiPrompt Engineering Is Dying – Embrace Agentic OrchestrationPure prompt pipelines stop scaling on real tasks. Redesign AI services around state, tools and permissions instead.aiMy most-used agent skills: what I forked, what I pulled, what I wroteA month of my own Claude Code sessions, counted skill by skill. Which skills I reach for, where each one came from, what I changed or wrote myself, and how to pick a starting set of your own.claude-codeA 27 Credit AI Product Ad: Price Before You Generate, Prompt Positively, Animate LastA four cut vertical product ad, built on a 100 credit Higgsfield budget for 26.97. The pricing grid that has no pattern in it, why negative phrasing made every output worse, where brand consistency actually comes from, and the rule for when a video model is worth 5 credits and when ffmpeg will do.generative-aiAgentic coding shifts the bottleneck to review, so redesign your CI/CDAI-generated code has moved the slowest step in delivery from writing to reviewing. Here is how to rebuild the pipeline so its controls hold at agent speed, with a working circuit breaker and a 90-day rollout.ci/cdHow to Build a Daily AI News Briefing with n8n, Ollama and Telegram on a MacA plain-language, step-by-step guide to a free, private morning briefing: n8n in Docker runs two news searches and a GitHub search, a local model in Ollama writes it up, and a Telegram bot delivers it, with every node's settings, a flow diagram and the traps that cost the most time.aiHow to Set Up Hermes Agent with Ollama and Telegram: A Private AI Assistant on Your Own ComputerA plain-language, step-by-step guide to texting an AI agent that runs on your own computer: Ollama for the model, Hermes Agent for the work, a Telegram bot for your phone, with the config, a flow diagram, safe skills and scheduled reports.aiAgentic AI in the Enterprise: A Four-Pillar Playbook from Demo to ProductionWhat an enterprise agent needs before it touches production data, pillar by pillar: reliability, observability, security and human-in-the-loop governance, with the controls, code and an eight-week pilot plan.aiBuilding Portable, Governed Skill Packages for Microsoft 365 Copilot AgentsHow to package agent skills as self-contained folders, upload them to Agent Builder, govern them at the tool calls that load them, and check them in CI.copilotSmarter vs Faster in Microsoft Copilot Studio: An ROI Framework for Agent CreditsHow Copilot Credits stack up when an agent reasons, and a cost-per-request calculation that tells you when paying for the smarter path is worth it.copilotAI writes fast. Someone still has to check it.What I learned helping my engineering team start using AI: the writing got quicker, the checking did not, and that is where the real work is.aiAgentic Postmortems: Let the Logs Talk, Not the LLMWhen an AI agent breaks production, its own account of what happened is a hypothesis. A logs-first postmortem process for SREs, with the template fields and autonomy tiers that make it stick.sreEngineering Reliable Agentic Loops in ProductionIteration limits, idempotent tool calls, the cost-accuracy numbers behind four orchestration patterns, queue-depth autoscaling and pass^k, with runnable Python for each guardrail.aiLocal AI Setup on Mac StudioImagine having the power of a supercomputer on your desk, capable of running state-of-the-art artificial intelligence models without breaking the bank or consuming excessive energy. The Mac Studio, powered by Apple Silicon's unified memory architecture, makes this a reality. This guide will show you how to set up and run large language models (LLMs) locally on your Mac Studio.testingCost per Successful Task: how to measure what an AI agent really costsToken prices say what a call costs, not what a finished job costs. How to calculate Cost per Successful Task from production logs, and which levers lower it.aiOperational intelligence for SREs: treating incidents as reasoning problemsAlerts tell you something is wrong. Agents that recall past incidents, and a deterministic policy that decides how far they may act, close the gap to knowing what to do.sreAgentic Refactoring Playbooks: How Agents Took 300K Lines of Legacy C to Code Health 10What the CodeScene Street Fighter III case study shows about refactoring legacy code with agents, and a playbook-driven loop you can pilot on your own hotspots.aiMCP Moves the Security Boundary: What Integration Engineers Must Change in ConnectorsWith MCP the model picks a connector by reading text a server author wrote. Where the trust boundary now sits, how connectors change, and the controls to put on the host.aiPostgres SKIP LOCKED as a Job Queue: When You Don't Need RedisBuild a transactional job queue on Postgres with FOR UPDATE SKIP LOCKED, keep it healthy under MVCC, and know the numbers that say it is time for Redis.postgres