Writing

Essays, audits, engineering notes.

Long-form pieces on bot-readability, the strange afterlife of dynamic rendering, the May 2026 cross-bot audit, and the day-to-day of running a prerender layer in production.

All posts

EXPLAINER

WebMCP, explained — your website hands the AI a menu of buttons

The plain-language version of WebMCP: what it is, the HTML attributes and JavaScript it's built from, why the page has to render first, and how PrerenderProxy brings tools to a CMS that can't add them itself.

8 min · 2026-07-02
DEEP DIVE

WebMCP, examined — the spec, the GitHub, and what's actually shipped

A citation-first teardown of WebMCP in mid-2026: the imperative and declarative APIs, the Draft-CG-Report reality, WebKit's oppose, the one real attack paper, an FAQ, and the honest good and bad.

16 min · 2026-07-02
ANALYSIS

The 2MB line: what Google actually changed about Googlebot (and what it didn't)

Google's documented Googlebot crawl limit now reads 2MB, down from 15MB. The fetch-vs-index nuance most write-ups get wrong, who is actually at risk, and why byte order rewards lean, content-first HTML.

10 min · 2026-06-20
PRACTICE

Stuck with legacy — when the CMS can't ship Product schema, the edge can

Inject schema, fix canonicals, convert soft 404s, shape responses per-bot at the prerender layer when the origin CMS won't. Cloaking risk addressed directly with production-grade safety rails.

12 min · 2026-05-19
OBSERVABILITY

What to log when you serve bots — schema + Combot.ai integration

The structured-log schema PrerenderProxy emits per request, the Vector → ES → Combot.ai pipeline, five anomaly alerts with thresholds. GDPR PII handling and VRL syntax notes baked into the recipe.

13 min · 2026-05-19
ENGINEERING

Reverse-DNS bot verification — copy-paste recipes for seven platforms

Working code for nginx, Cloudflare, Fastly, Vercel, AWS, Apache + OpenResty Lua. The three-step protocol, the published vendor IP-range JSON sources, the operational tips most rDNS posts skip.

14 min · 2026-05-19
POSITION

Don't block GPTBot if you're a brand — the 2026 case for AI-memory presence

The decision framework that says "block training, allow retrieval" was written for publishers. For brands, the math inverts. The parametric-recall mechanism, the 2026 numbers, the strongest counter-argument addressed directly.

11 min · 2026-05-19
DEEP DIVE

The strange afterlife of dynamic rendering, 2018–2026

Eight years from Google I/O recommendation to formal deprecation to AI-era revival. The compute-economics re-reading of 2024, the strongest counter-argument, and a 2027 forecast.

12 min · 2026-05-19
RESEARCH

62 of 100 — what we learned auditing the world's largest stores

62 sites block AI crawlers wholesale. 6 single-target ClaudeBot. 5 still dynamic-render. The full pattern map from our May 2026 audit.

9 min · 2026-05-19
DECISION

Should you block AI bots? A 3-step decision framework

The Cloudflare "Block AI Bots" preset blocks both training and live-retrieval bots. Here's how to write a more honest robots.txt — and what it costs.

7 min · 2026-05-19
ENGINEERING

Building PrerenderProxy in 2026

Fastly VCL + Puppeteer + a custom warm-cache crawler. Why we picked each piece, what we'd do differently, and the operating cost at our scale.

8 min · 2026-05-19