📰 Dev.to · Ken Imoto

I was shipping a Claude Code plugin that guards against /compact amnesia. At 67% context usage it warned me. Then its own hooks failed during the release. Both

Anthropic shipped a +39 / -26 rewrite of their frontend-design Skill on June 18. The new version contradicts the old one, names 3 specific AI design defaults (o

Every voice-agent benchmark reports end-to-end latency. None of them report what it costs to let the user interrupt. I instrumented my own barge-in path and wat

Pairing Opus with a free local Qwen executor should be the cheap option. I measured 40 trials across 3 code-repair tasks. It was the most expensive cloud config

I gave a coding agent read access to a month of my company's transactions and let it do the bookkeeping. It reconciled 200 rows and got 11 categories wrong. Her

Ahrefs studied 75,000 brands and found unlinked web mentions correlate with AI visibility at 0.664 — about 3x stronger than backlinks at 0.218. The off-page sid

When your voice AI can't hit 300ms, you stop trying to be fast and start lying convincingly. Five perception hacks I use, with cost, payoff, and where they back

I run a blog with 31 English articles. I asked ChatGPT, Claude, Gemini, Perplexity, and Brave AI to cite it. Three articles did all the work for every engine.

Over 22 days: PT 748 pageviews, EN 195, JA 27, ES 7. I shipped 4 languages expecting English to win. Here's what actually happened, and what it taught me about

An SEO agency reported an 8,337% lift in ChatGPT referrals using four LLMO pillars. I spent 90 days copying the playbook onto three indie sites. Three of the fo

Claude refactored 100 functions in my Python service. CI was green. Mutation tests were green. Two weeks later, on-call paged me — 7 functions were slower in pr

A production Claude agent had been quietly burning 5.2M tokens a month. I turned on per-call tracing for 30 days, found four bottlenecks no dashboard surfaced,

Plugged 8 MCP servers into one Claude Code agent. No error. No warning. Three pairs of servers had registered the same tool name, and Claude was picking one at

I kept reading that voice agents respond under 300ms. I measured 5 stacks against the same one-minute conversation. Three of them missed the cliff entirely. Her

Observer / Strategist / Marketer were following the rules in strategy.md. My Strategist had been writing 'we will evaluate next week' for three weeks straight,

Four days after registering a new domain, GA4 showed 65 page views and 34 users across 9 countries. Before celebrating, I beat the data with 5 signals. What sur

Nothing tracks whether AI is citing your site. Here is how to measure LLMO visibility with GA4, Python scripts, and a 30-minute monthly protocol that costs you

I shipped my third llms.txt this month and felt productive. Then I opened 30 production llms.txt files from companies like Stripe, Vercel, and Anthropic. Most o

One agent doing observation, strategy, and execution took 20 minutes and burned 120k tokens per run. Splitting it into Observer / Strategist / Marketer dropped

I asked five AI search engines the same question. They cited five different sources. Here's how each platform decides what gets cited, and what you can do about

I spent a month switching between Claude Code and ChatGPT Codex. They aren't the same tool, they aren't even the same idea, and the cheapest answer turned out t

Skills replaced my .claude/commands/ directory in fifteen minutes per file. Here's the frontmatter that matters, the gotchas, and the !`command` trick that chan

I turned off every permission prompt for a day. Here's the OWASP Agentic Top 10 (2026) lesson plan I got back, written in incident reports instead of bullet poi

The invisible bill I was debugging why my Claude Code sessions felt sluggish after...