the writing ai & claude code · 23 essays

AI & Claude Code

Claude Code, agents, MCP servers and the models behind them. How I use them to build and run this site every day, and where they still fall short.

28 oct 2026 WordPress MCP vs Headless CMS MCP: Which Fits AI Agents? A vendor-neutral comparison of WordPress MCP, headless CMS MCP servers and git-based Jamstack, scored on permissions, drafts, audit trails and security record.
20 oct 2026 Is AI Sentient? What Nature's Scientists and Anthropic Say Nature's scientists say today's AI is not conscious. Anthropic says the question is open. A fair look at both positions, the evidence behind each, and what to do now.
19 oct 2026 Claude Code Slash Commands: A Practical Cheat Sheet A task-first lookup for Claude Code slash commands, plus how to tell built-ins from skills and why a command goes missing in your session.
20 sept 2026 Claude Code Routines vs GitHub Actions: Where to Schedule Agents Claude Code Routines (launched April 2026) run scheduled AI agents on Anthropic's cloud with zero server setup. GitHub Actions cost less and give you more control. Here's how to pick the right host for each job, wire up the blog queue worker, and handle retries without guessing.
19 sept 2026 Ralph Loops in Claude Code: Verifiers, Budgets and Stop Conditions A Ralph loop is not magic. It's a bounded iteration pattern: Claude works, a verifier checks, the loop continues or stops. Here's how to wire up the goal, the budget and the exit conditions properly.
18 sept 2026 MCP Server Security: Vet Tools Before You Install Installing an MCP server is not a one-click convenience. It grants executable code access to your credentials, filesystem, and agent context. Here's a reusable checklist to review any server before it touches your stack.
17 sept 2026 Claude Code Plugins: Build a Private Marketplace for Your Team Anthropic added private plugin marketplaces to Claude Code, announced February 24 2026. This post covers the full packaging and distribution workflow: bundling skills, hosting a private repo, controlling versions and rolling back cleanly.
16 sept 2026 Claude Code Skills, Hooks, Subagents or MCP: Which to Use Skills, subagents, hooks, and MCP servers all extend Claude Code, but they are not interchangeable. This post maps each mechanism to the job it actually solves, includes a six-task decision matrix, and flags the wrong choices teams make most often.
15 sept 2026 Claude Code Usage Limits: Session Caps, Weekly Limits and Options Claude Code enforces two overlapping limits simultaneously: a rolling 5-hour session cap and a separate weekly quota. Both share one pool across every surface. Here's how to read which one you've hit and what your actual options are.
14 sept 2026 Agentic Commerce for Headless Stores: A Readiness Checklist Anthropic published the Claude for Commerce agent blueprint on September 2, 2026. It's reference code, not automatic enrollment. Here's the catalogue-to-checkout readiness matrix your headless store needs before any agent touches your checkout.
13 sept 2026 Claude Code Worktrees: Parallel Tasks Without File Conflicts Git worktrees let you run two or more Claude Code sessions simultaneously, each on its own branch, with zero file conflicts. This post covers setup, port and database isolation, branch merging, and safe cleanup when you've got uncommitted work to preserve.
12 sept 2026 Claude Code Permissions: Modes, Rules and Team Settings Claude Code's permission system has five precedence layers, multiple modes, and a few sharp edges most teams hit only after something goes wrong. Here's how it all fits together before that happens.
10 sept 2026 Build an MCP Server in TypeScript on Supabase Build a read-only MCP server in TypeScript on Supabase Edge Functions using the official SDK's WebStandardStreamableHTTPServerTransport, with paginated tools, RLS-enforced caller isolation, and Claude Code verification.
8 sept 2026 Claude Code Context Window: When to Compact, Clear or Split Claude Code's context window holds your conversation, file reads, tool output, MCP definitions and memory files all at once. When it gets crowded, you have three options: compact, clear, or split into a new session. Pick the wrong one and you either lose critical decisions or drag noise into your next task. Here's how to choose.
7 sept 2026 Claude Code Skills: Write and Test Your First SKILL.md A Claude Code skill is just a folder with a SKILL.md inside it. The description field is the trigger. Get that wrong and the skill never fires. Here is how to write one that does.
5 sept 2026 Claude Managed Agents: Setup, ant apply and SDK Trade-offs Claude Managed Agents and the Agent SDK are both real options, and picking the wrong one costs you weeks. Here's what the service actually hosts, how ant apply works in practice, and when to stay on your own infrastructure.
27 aug 2026 Vercel AI Gateway: Routing, Failover, and What It Actually Saves Vercel AI Gateway promises unified LLM routing, automatic failover, and token-level caching. I ran it across three production projects to find out what it genuinely saves and where it quietly lets you down.
26 aug 2026 Claude Code Hooks: The Automation Layer I Wish I Had Set Up Sooner I ignored Claude Code hooks for three months after launch. Big mistake. Once I wired them up properly at Seahawk, they quietly fixed problems I'd been blaming on the AI itself, formatting drift, untested commits, missing linting passes.
25 aug 2026 Base44, Lovable or Claude Code: Pick by Exit Cost A three-tier map for AI building: managed app builders, agentic coding in your repo, and raw open models, judged by what it costs to leave.
21 aug 2026 Claude Code Subagents: How I Split Work Across Parallel Agents I started running Claude Code subagents in parallel after a particularly brutal WooCommerce sprint. Here's the actual workflow I use to split tasks, avoid conflicts, and ship faster without losing control of the codebase.
17 aug 2026 The Claude Agent SDK in Practice: Building Agents That Work Most "AI agent" demos are theatrical. They look great in a Loom recording and collapse the moment a real user touches them. Here's how I actually build agents that hold up under production load using the Claude Agent SDK.
17 aug 2026 Designing Software With Claude: My Brainstorm-to-Spec Workflow I used to spend three days writing a spec that a client would ignore anyway. Now I do it in two hours with Claude, and the output is genuinely better. Here's the exact process I used at Seahawk.
25 jul 2026 Why opening more AI models makes you worse (until it doesn't) More models feels like more capability. Past about three it stops being capability and starts being a committee that cannot remember what it agreed to. Here is the curve, drawn from a very expensive month of finding out.