writing
Articles
What I’ve learned building with AI agents: specs, skills, harnesses, and what keeps it all standing.
BMAD Method: A Real Run in Claude Code, From Install to Commit
BMAD Method v6.12 from a practitioner: what it is now, the headless install, the skills it adds, one real feature from request to commit, what it cost, and where the gate leaks.
read article →
OpenSpec Tutorial: One Real Change, From Install to Archive
OpenSpec 1.14 from a practitioner: what the OPSX rewrite changed, install and init, the right command spelling for every agent, and one real change run from propose to archive.
read article →
CLAUDE.md Best Practices: The Complete Guide
What goes in CLAUDE.md, what to cut, where each file lives and what it costs you. Plus the import trap that put 70k tokens into every session of a 13-app build.
read article →
Hermes Agent for Coding? Even Hermes Hands the Repo to Claude Code
A team I work with codes through a shared Hermes instance in Slack. Hermes' own source explains why that hurts: its coding posture and verify gate stay off on chat surfaces, and it ships skills that hand code to Claude Code, Codex and OpenCode.
read article →
Karpathy's LLM Wiki: How to Keep It From Becoming Legacy
Karpathy's LLM wiki removes the maintenance burden that kills wikis and creates a new kind of legacy: pages that stay fluent while they drift from what happened. The build, the four rules, the model setup and two zero-token checks I run as an engineer, with an audit of my own vault.
read article →
The Agentic SDLC Playbook: The Build Shrank. Now Fix Both Sides.
For decades the software lifecycle was built around one slow, expensive step. Agents made it cheap, and everything that leaned on it fell over. A field playbook for the agentic SDLC, written in the order you adopt it: the repository, the checks, the context, the spec, the review, the flow, and the loop.
read article →
Hermes Agent vs Claude Code: A Colleague With a Shell vs a Tool in Your Repo
Hermes Agent and Claude Code both get called an agent, so people search them against each other. They are not competing for the same job: one runs on a server and keeps working after you stop, the other lives in your repo while you watch.
read article →
Hermes Agent vs OpenClaw: I Read Both Codebases Twice
In June I read the source of Hermes Agent and OpenClaw to pick a runtime for my agent fleet. In October I read it again and three of my conclusions were wrong. What changed, where each one still leads, what each one costs, and how to choose.
read article →
Is Claude Max Worth It? I Left, Tried the Cheaper Roads, and Came Back
Claude Max is the cheapest way to run frontier Claude models all day, on one condition: you work inside Anthropic's harness. What the plan buys, what it costs you, the four ways to pay for coding tokens, and what I learned leaving Max and coming back.
read article →
The SaaS I Replaced in an Afternoon
In one afternoon with Claude Code I moved server-side GTM off Stape and onto my own VPS, found why Meta counted almost twice what GA4 did, and took over the pager. What an engineer can bring in-house now, and what it costs.
read article →
The Best AI for Coding Is a Pipeline, Not a Model
There is no single best AI for coding. The answer is a harness plus one model per phase of the work, paid for in a way that fits how you build. The decision map I use, with the seven-agent pipeline I ran on open models.
read article →
Hermes Agent and a Second Brain: Give the Agent Your Context, Not Your Memory
Hermes Agent's memory (MEMORY.md, USER.md, external providers) is not your knowledge base. A second brain is a markdown wiki you curate, that the agent reads as context and does not write to unless you ask it to.
read article →
Is Hermes Agent Safe? The Agent Should Never Hold the Token
An always-on agent with a shell is a colleague with root. What actually went wrong running Hermes Agent in production, the controls that caught it, how OpenClaw compares, and a checklist for before you give any agent a shell.
read article →
AI-DLC With Claude Code: A Working Setup
How to run AI-DLC 2 inside Claude Code: install the aidlc engine, configure the project, approve the hooks, start a first workflow, and the practitioner layer the docs skip: a repository the agent can read, context discipline at the gates, and model settings worth touching.
read article →
The Best Model for Hermes Agent Is the Cheapest One That Survives Your Runtime
There is no best model for Hermes Agent, only a best model for the job routed through the runtime you actually run. My own shootout shows the same model losing two points for no reason except where the persona landed.
read article →
Hermes Agent Skills: Let It Learn Procedure, Write the Judgment Yourself
Hermes Agent skills explained from production: the SKILL.md format, the skills hub behind hermes skills install, plugins versus skills, and why the background review that writes its own skills should never touch judgment.
read article →
Google Workspace MCP and Gmail MCP: What Actually Touches Your Inbox
Gmail, Calendar and Drive are the highest-signal, highest-risk data an agent can read. What the official and open-source Google Workspace MCP servers cover, the OAuth model behind each, and how to wire one into Hermes or Claude Code without handing an agent your inbox.
read article →
How Much Does Hermes Agent Cost? Where the Tokens Actually Go
Hermes Agent is free and MIT-licensed. The real cost is tokens and the machine it runs on, and both move almost entirely with how you configure the runtime: skills, memory, compression, tool search, the background review loop.
read article →
OpenCode vs Claude Code: I Switched, Built Twelve Agents, and Came Back
I left Claude Code for OpenCode in May 2026, built a twelve-agent pipeline on open models, then came back. What OpenCode does better, what Claude Code does better, and the one variable that decided it for me: who pays for the tokens.
read article →
Hermes Agent on Telegram: Run Your Agents From a Voice Note
Hermes Agent's Telegram gateway, checked against the v0.21.5 source: pairing and allowlists, chat approvals for risky commands, voice note transcription, and the bot token as a secret worth guarding.
read article →
Slack MCP Server: Read Access for Agents, Not a Voice
What a Slack MCP server actually gives an agent, the official Slack-hosted option versus the self-hosted korotovsky server, how to wire either into Claude Code or Hermes Agent, and why posting should stay off by default.
read article →
Running Hermes Agent in Docker: Setup, Compose, and a Real VPS
Hermes Agent Docker setup with the official image, from docker run to the first chat, then what the image and Compose file ship, persisting HERMES_HOME, profiles, dashboard auth, upgrades, and the Swarm traps that cost real time on a VPS.
read article →
Hermes Agent and MCP: Wiring Servers That Actually Show Up
How MCP servers are configured in Hermes Agent, what changed with the v0.21.5 Connectors page, how tool search really defers tools, and the production fixes a cold-spawn race and a hard-scope flag forced on me.
read article →
An Agent Reads Slack, Google, Git and Jira So I Get One Briefing a Day
At my day job, a Hermes Agent cron job reads the company's Slack, Google Workspace, Git and Confluence/Jira through MCP, uses my second brain as context, and writes me a daily briefing organized by six senses. How it works, how to build one, and what keeps it from becoming a noisy digest.
read article →
I Built a Company of AI Agents in Paperclip. Here Is What Broke.
Thirteen agents on Paperclip and Hermes Agent: a CEO, a chief of staff, five heads and the specialists they route work to. What Paperclip is, how the org is wired, and every failure it took before the loop closed on its own.
read article →
What Is Hermes Agent? A Developer's Guide From Running It in Production
Hermes Agent is Nous Research's open-source, self-hosted agent runtime with memory, a learning loop and more than 25 messaging gateways. How it works under the hood, what the learning loop really does, where it breaks in production, and when a developer should pick something else.
read article →
From Squad to Pod: Who Does What in AI-DLC
AI-DLC changes roles more than it changes tools. Humans decide, the AI operates, squads shrink toward pods, the manager becomes the keeper of pace, and the engineer still owns the code in production. What each role does now, and the competency-matrix problem nobody has solved.
read article →
Mob Elaboration: The Ritual That Replaces Sprint Planning
Mob Elaboration is the AI-DLC ritual where the AI asks the questions and the people who can decide answer them, in one sitting, before any code exists. Who belongs in the room, how to run the session, the three ways to split spec authorship, and what AI-DLC 2 now automates.
read article →
Kiro vs Claude Code: Gates Built In, or Gates You Build
Kiro is an IDE that runs spec-driven development for you, with approval gates between requirements, design and tasks. Claude Code is a terminal harness where you build that workflow yourself. Side by side, the same method in both, and who should pick which.
read article →
Don't Adopt AI-DLC. Steal From It.
AI-DLC is the most complete public framework for building software with agents, and adopting it as a block is still a mistake. Treat it like a supermarket shelf: take the mechanisms that solve your problems, leave the doctrine, and write down why. AI-DLC 2 just proved the point.
read article →
What Is AI-DLC? The AI-Driven Development Life Cycle, From Inside a Rollout
AI-DLC is the AWS method where the AI drives the work and humans decide at the gates. What it is, how AI-DLC 2 runs today (5 phases, 33 stages, 14 agents, the 2.11 defaults), how it differs from Anthropic's AI-native SDLC playbook, when not to use it, and why Spec-Driven Development has to come first. Written from inside a real rollout.
read article →
What Is an Agent Harness? Everything in the Agent That Isn't the Model
An agent harness is all the code around an AI model that turns it into an agent: the loop, the tools, the context, and the controls. What each part does, real harnesses side by side, and why the harness decides more than the model.
read article →
AI-DLC 2: What Changed, and What It Made Obsolete
AWS rewrote AI-DLC. Version 2 ships as a deterministic engine that takes routing, grading, and state away from the model: 5 phases, 33 stages, 14 agents, 11 workflow profiles, a learning loop, verified construction checkpoints. What changed from version 1, which homegrown layers it made obsolete, and what survived.
read article →
After Story Points: What to Measure in AI-DLC
Story points and velocity stop meaning anything when an agent writes the code. What to measure instead in AI-DLC: lead time from intent to tested code, rework at the gates, discovery time upstream, throughput read next to cycle time, and the data AI-DLC 2 already records for you.
read article →
AI-DLC in Brownfield: Reverse Engineering, 1M Context and the 75% Rule
Running AI-DLC on code that already exists: scope reverse engineering to the feature and run it before the mob, budget the context window, clear it at gates, and treat context compaction as the governance failure it is.
read article →
Gates Are a Loss Function. Here Are the Four Ways They Fail.
In AI-DLC every human approval is supposed to catch a wrong decision while it is still cheap to fix. In practice, approval gates fail in four predictable ways. What they are, why they happen, and how to keep humans on the loop instead of rubber-stamping it.
read article →
AI-DLC's Blind Spot: Where the Intent Comes From
AI-DLC turns an intent into software with real discipline, and assumes the intent was right. It rarely is. Why errors at the start multiply, why the default workflow skips the phase meant to catch them, and the upstream ritual that fills the gap: five questions, a complexity triage, a curated source index, and an outcome as the success criterion.
read article →
Why AI-DLC Makes You Slower First
Adopt AI-DLC and throughput jumps on day one while cycle time gets worse. Why the J-curve happens, why code review becomes the bottleneck, and how to tell a learning curve from a real problem before you kill the pilot.
read article →
Six Rules Against Vibe Coding Inside AI-DLC
AI-DLC puts gates between the phases, and teams still slide back into vibe coding at the keyboard between them. Six rules a production team uses to stop it, each with the exact words to type, plus the size limit on Units that keeps code review honest.
read article →
AI-DLC vs Spec-Driven Development: Why You Need SDD First
AI-DLC and Spec-Driven Development are not rivals. One is a practice, the other is a lifecycle built on top of it. Every AI-DLC gate is a spec review, so a team that cannot write and judge a spec cannot run AI-DLC. The ladder from copilot to AI-DLC, and how to tell which step you are on.
read article →
Spec-Driven Development for Teams: The Shared-Spec Workflow
How a spec stops being personal discipline and becomes the contract your team ships from: shared specs in git, spec review by pull request, gates on a board, and why the bottleneck moves from writing code to integrating it.
read article →
OpenSpec vs. Spec Kit vs. Superpowers: SDD Tools Compared
How OpenSpec, GitHub Spec Kit, Superpowers, pi-sdd-kit, and plain AGENTS.md memory compare for spec-driven development with AI agents, how to stack them, and the honest rule for which to use.
read article →
AGENTS.md: The File That Gives Your Agent a Memory
AGENTS.md is the file your agent harness loads before the model wakes up. Here is what belongs in it, what to cut, and how to keep it from turning into legacy code.
read article →
Most of Your AI Coding Agent's Tokens Die Where You Never Look
Your AI coding agent burns most of its context on output you never read. The three-layer stack I run to cut the token bill 60 to 90 percent: RTK at the tool layer, compression at the conversation layer, memory at the knowledge layer.
read article →
Context Engineering Is Just Deciding What the Model Sees
Context engineering is not a new discipline. It is the one job inside your AI agent harness that determines whether the model can actually finish the work.
read article →
Harness Engineering: Stop Upgrading the Model and Fix the Harness
The model is a commodity. The harness is your edge. A practitioner's guide to harness engineering for AI coding agents: cut context waste, keep tools minimal, and give the agent a memory that outlives the session. With pi as the worked example.
read article →
Give Your Agent a Shell, Not a Protocol
MCP servers front-load every tool schema before you type anything. CLI tools through bash pay nothing until called. Here is when to use each, and how to audit the servers you already have.
read article →
Pi: The Coding Agent That Hides Nothing
Pi is an open-source, bring-your-own-key coding agent with four tools and a 150-word system prompt. No MCP, no hidden planner, no invisible subagents. Just a loop you can fully see.
read article →
One Model Per Phase: Automatic Model Handoffs in Pi
Exploring, planning, building, and reviewing each want a different model and a different amount of thinking. pi-skill-model-handoff switches both when a skill loads, so the model follows the work instead of you switching it by hand.
read article →
How to Test a Claude Skill: Evals, Triggering, and Description Optimization
Most Claude skills ship without a single real test. Here's the eval loop I use: 20 queries, a train/test split, and a trigger-rate target to prove a skill fires when it should and stays quiet when it shouldn't.
read article →
Claude Skill vs Prompt vs Memory vs Subagent: What to Use When
Four primitives, one decision. Cut through the confusion between Claude prompts, CLAUDE.md memory, skills, and subagents. Pick the right tool every time.
read article →
Claude Skills vs MCP vs Subagents: When to Use Each
A decision framework for choosing between Claude Code skills, MCP tools, subagents, and memory: comparison table and choose-this-when breakdown per option.
read article →
GitHub Spec Kit 1.0: A Practitioner's Guide to Spec-Driven Development
GitHub Spec Kit 1.0 from a practitioner: install the specify CLI, run the pipeline through converge, write a constitution that holds, and cut the ceremony with the lean preset.
read article →
Claude Skill Examples: Real SKILL.md Files You Can Steal
Nobody publishes their working SKILL.md files. Here are real, production examples from a persona-skill library: alex-hormozi, lifeos-lint, loading maps, and the decision-split reference structure that makes them actually work.
read article →
SKILL.md Explained: Every Frontmatter Field and Progressive Disclosure
The complete technical reference for SKILL.md: every frontmatter field, the directory convention, and the three-level progressive disclosure model that keeps Claude Code skills fast and context-efficient.
read article →
What Is Kiro? AWS's Agentic IDE for Spec-Driven Development
Kiro is AWS's agentic IDE for spec-driven development: three structured files, automated hooks, and parallel agents that implement what you specify.
read article →
What Are Claude Skills? A Plain-English Explainer
Claude skills are reusable SKILL.md modules that teach Claude how to do one kind of work. They load on demand, persist across sessions, and follow the open Agent Skills standard.
read article →
Persona Skills for Claude Code: Why Naming an Expert Beats Role Prompting
Role prompting treats a persona as a one-line costume. This guide shows why naming an expert only pays off inside a Claude Code skill, backed by distilled references and a loading map that routes the model into its densest cluster.
read article →
Spec-Driven Development Case Study: 13 Apps in 70 Days, Solo, with AI
How Spec-Driven Development let me ship a production crypto fintech solo in 70 days with AI agents: the architecture, the specs, and the honest limits.
read article →
How to Create Claude Code Skills: The Complete Guide to Agent Skills
Claude skills are reusable behavior modules that replace system-prompt debt. Learn the full anatomy: SKILL.md, progressive disclosure, references, and the description field that triggers everything.
read article →
Spec-Driven Development vs. Vibe Coding: A Practitioner's Honest Comparison
Vibe coding is real and useful. It is also wrong for anything that must be correct. A METR RCT, a payment idempotency head-to-head, and the honest verdict.
read article →
Spec-Driven Development with Claude Code: The Hands-On Workflow
How to run Spec-Driven Development inside Claude Code: plan mode, the three-layer context system, steering files, .status gates, and specialized sub-agents that turn a stateless AI into a consistent collaborator.
read article →
How to Write a Spec for AI Agents: Template, EARS Format, and Examples
How to write a spec an AI agent can build from: a copy-paste template, EARS requirements, three-tier boundaries, the interview step, and a real run of a one-line prompt against the spec.
read article →
What Is Spec-Driven Development? A Practitioner's Guide (and When to Skip It)
Spec-Driven Development is the method I used to ship a 13-app crypto fintech in 70 days, solo, with AI agents. What a spec is, the four pillars, a full worked example, the EARS format, and when to skip it.
read article →
Don't Code, Specify: Why Senior Developers Are Getting Worse Results with AI
The methodology I used to build a crypto fintech (13 apps) in 70 days, solo, with AI agents.
read article →