Blog
Case studies and technical deep-dives from building in public.
Adventures in Local LLMs: How We Made Our AI 2.3x Faster for 81 Cents
Same graphics card, same model. With Claude Opus 5.5 orchestrating, our local-server LLM ended up reading prompts 2.3x faster, and the whole run used 81 cents of electricity.
AI Visibility Takes Weeks. Start With a Baseline.
LinkedIn's 2026 AI search guide is written for B2B teams with a marketing department, but its timing and measurement advice fits a ten-person business. Our own pilot shows why checking your own name tells you very little.
Reading the AI Visibility Numbers
LinkedIn's AI visibility guide gives sound advice, but several of its headline numbers measure something narrower than they sound. A small business should plan around its own baseline.
You Are the Bottleneck
Spinning up more agents hides the wait and moves the cost onto you. The fix is to take yourself out of the steps that don't need you and put independent checks between the agent and your attention.
What to Stop Prompting Now
Anthropic's own guidance for its Claude 5 models says to delete lines most of us still type: think carefully, show your reasoning, retype your charts. Here is the stop-doing list, and what to write instead.
Using AI Isn't the Same as Running On It
Two 2026 surveys agree: most small businesses now use AI, and most say it's working. Almost none have it running inside how they actually operate. That gap is the story.
A Change Is Not Done When the Command Succeeds
A write can succeed without being read. A deploy can succeed without changing what's live. Here's the check we run to catch the difference, and two real problems it just found in our own product.
When Your AI Agent's Instructions Contradict Themselves
We audited every instruction file feeding our AI coding agents. Three projects came back clean. One had two files telling the same agent two different failure thresholds.
GDPval Isn't One Benchmark Anymore
A model scoring 85% on GDPval sounds like a fixed, trustworthy number. It isn't anymore, there are two different tests sharing that name, and they disagree.
Why Should You Ever Log In Again?
Cloudflare's CEO says bot traffic will outpace human traffic by 2027. Software vendors are already rebuilding around agents as the operator, not the user. Here's what that shift looks like at our scale.
We Ran Our Own Site Through a Stranger's Checklist
A 20-item website checklist went viral on X. We ran addAI.dev through it ourselves, fixed what we could fix outright and are plain about the rest.
Back on the Open Side
Structured data doesn't make our site more crawlable, it makes the facts in it liftable. Here's the Organization and BlogPosting JSON-LD we just added, live.
The Header We Bet On Just Got Extended
Cloudflare extended the same Content-Signal field our robots.txt already ships. What changed, what didn't and why the two are easy to mix up.
Where Your Context Window Actually Goes
Most agent CLIs never show you what's actually sitting in context. Here's a real breakdown, and the mechanism behind it.
The One Page We Didn't Make Agent-Friendly
Why our Contact Us and Join Us forms are the one part of addAI.dev we deliberately made harder for automation to reach, AI agents included.
The Standards Aren't Written Yet
What's actually standardized in AI discoverability right now, what's still a pre-RFC draft and what's a buzzword wearing a protocol's clothes.
How We Keep Our Own Blog Readable to AI Agents
The skill that converts our posts to Markdown, the Markdown twin serving them today and what we're rolling out next.
Claude Code Imports AGENTS.md - Until It Doesn't
Claude Code doesn't read AGENTS.md, and the recommended CLAUDE.md import doesn't reliably survive compaction either - here's what we found testing both claims ourselves.
How to Make Your Website Visible to AI
How we took a small business site from invisible to AI-agent-ready, one fix at a time.