Alex Herrero

A security gate you can poke

Tue, 09 Jun 2026 00:00:00 +0000

An interactive, deterministic intercept-gate — propose a command and watch a no-LLM ruleset allow, hold, or block it, with its reason.

A kill-switch you can't override

Mon, 08 Jun 2026 00:00:00 +0000

A kill-switch you can’t override isn’t a kill-switch — it stopped my own agent cold today, which is exactly the point.

Why I started Agent M

Sun, 07 Jun 2026 00:00:00 +0000

Every agent I tried was brilliant for an hour and a stranger the next morning. So I stopped waiting for a smarter model and built the memory.

A security person's case for boring, auditable AI tooling

Fri, 05 Jun 2026 00:00:00 +0000

Useful-enough-to-depend-on and safe turn out to be the same problem. The boring path – deterministic gates, legible state – is the one that scales trust.

color-mix() for theme tokens

Fri, 05 Jun 2026 00:00:00 +0000

CSS color-mix() gives you a full light/dark token system with no preprocessor and no build step — derive every muted and hairline shade from two base colours.

Deterministic gates before the LLM-judge

Tue, 02 Jun 2026 00:00:00 +0000

Gating on deterministic checks before the LLM-judge cut my review noise roughly in half. Cheap, truthful checks first; the expensive, sycophantic one last.

About

Mon, 01 Jan 0001 00:00:00 +0000

I’m Alex Herrero — a security-minded engineering leader. By day I lead engineering for AI-driven threat detection at Google Cloud. Off the clock I build the tools this site is about: Agent M, a harness that gives coding agents durable memory and a phase-gated workflow, and Crickets, the safety and verification primitives that ride on top — kill-switches, fresh-context evaluators, guardrails.

This blog is where I write about the experiments behind those tools: what I’m trying to make AI agents reliable, safe, and genuinely useful, and what I learn running them against real work.

Colophon

Mon, 01 Jan 0001 00:00:00 +0000

How this site is built — which is, fittingly, with the tools it’s about.

Generator: Hugo — a single Go binary, minimal dependencies (a security-minded choice about supply-chain surface).
Design: hand-authored, not a theme. The “Structured Ledger” palette with Inter Tight + JetBrains Mono, self-hosted — no third-party font requests.
Hosting: Cloudflare Pages, deployed from a private repo via GitHub Actions (push model). Every build runs a deterministic gate — HTML structure, unique IDs, WCAG AA contrast — before it ships.
Email: an owned list on a Cloudflare Worker + Resend. No third-party newsletter service.
Analytics: Cloudflare Web Analytics — cookieless, no consent banner, no tracking.
Built with: Agent M (the agent harness) and Crickets (agent-safety primitives) — the tools this blog is about.

No cookies. No trackers. No build step you can’t read.

Now

Mon, 01 Jan 0001 00:00:00 +0000

What I’m focused on at the moment (a /now page).

Building Agent M and Crickets in the open — and writing up the experiments here.
Sharpening the intersection I care about: making AI agents reliable, safe, and useful, from a security background.
Reading (and occasionally posting in) the AI-safety corners of the web.

Last updated June 2026.