Writings

WebMCP lets your web app register typed tools for browser agents with document.modelContext.registerTool. No separate MCP server. Here's how it works, and when you still need one.

Schedule a Mastra agent with a scheduled workflow: declarative cron, date ranges in code, an evidence check, and a Slack post the model can't trigger.

Fractional CTO pricing in 2026 ranges from about $3K per month at the low end to $15K+ per month for ongoing architecture review, delivery guidance, and production judgment. The important change is that a senior human can use agents to turn decisions into reviewed implementation faster without handing judgment to the model.

Most AI demos hand one agent the whole job. When the output touches reputation, money, or outbound email, use a workflow instead of an agent loop. Here's where the boundary goes, built with Mastra.

When an agent fails, harness engineering gives you four levers (Context, Tools, Loop, Governance) to find which one broke in under a minute.

Prompt-based approval gates fail because the model decides whether to ask. Mastra's requireApproval primitive removes that decision entirely. Here's how to implement it.

What 'supported by a fleet of agents' means in practice: which tasks automate, which don't, and where the ROI breaks down. Evidence from Stripe, Coinbase, Ramp, and Shopify.

A backend PR kept colliding with other merges on database evolution numbers. Four manual rebases later, I described the problem to Claude and let it write a routine to handle the rest.

Practical tips for getting the most from Claude Opus 4.7's 1M context window in Claude Code. Effort levels, proactive compaction, subagent delegation, and session management from daily production use.

Most AI agent governance advice targets boards, not builders. Three failure patterns, real TypeScript examples, and what a CTO should do Monday morning.