Latest

Sandboxing LLM-Generated Code Without Docker-in-Docker

August 20, 2026 Code Security Sandboxing

Running agent-generated code in a plain Docker container trusts a boundary built for reviewed workloads. MicroVMs give you a real kernel per execution at roughly 125ms boot.

Read article

All posts

AI Agents

TDD in the Agent Loop: Mirror or Actual Check?

A test written by the same agent that wrote the implementation is a mirror, not a check. Which TDD invariants survive agent authorship, and the mutation-testing gate that catches the rest.

August 18, 2026
AI Tools

AI Context Caching Is Cache Invalidation With Worse Tooling

Context engineering borrowed cache vocabulary but skipped the part that matters: key design. Here's how versioned, provenance-based keys fix stale retrieval where TTL clocks can't.

August 17, 2026
AI Agents

750 Tokens/Sec Won't Fix Your Agent Loop

OpenAI's Ultrafast tier hits 750 output tokens/sec on Cerebras hardware. Whether that changes your agent's wall-clock time depends on one ratio you can measure in 20 lines of Python.

August 14, 2026
AI Agents

Open Weights Don't Mean Open Behavior: Audit Muse Glimmer First

Meta's Muse Glimmer is 30B params under Apache 2.0 and ships with no documented tool-call schema. Three concrete audit steps to run before it touches your agent stack.

August 10, 2026
AI Agents

Temporal Policies for AI Agents: Stateful Authorization in Plain SQL

Per-call RBAC can't say "deny this refund because a cancellation already happened." Sequence-aware authorization for AI agents is two Postgres tables and one plpgsql function.

August 10, 2026
AI Agents

Role Confusion Is the Prompt Injection You Didn't Patch

Role confusion exploits how text sounds, not which role tag it carries. The fix isn't a filter — it's a trust-level envelope in your message construction code. Here's the structural patch.

June 24, 2026
AI Agents

Benchmarking Open Models on Your Own Tool Schemas Before You Commit

Public leaderboards score tool calling on clean synthetic schemas, not the nested mess your MCP server exposes. Here's the ~50-line Python harness that settles the debate on your own stack.

June 19, 2026
AI Agents

Context Window Accounting: Why Your Agent Gets Dumber After Turn 8

Long agent loops collapse around turn 8 — not because the window is full, but because tool results silently eat the budget. Here's the audit pattern and checkpoint prompt to fix it.

June 01, 2026
AI Agents

Coding Agents Create Decision Fatigue: The Review Loop That Cuts It

Coding agents relocated your cognitive load from writing to reviewing — and judgment degrades fast. Here's a preamble prompt and a three-tier triage that cut the fatigue.

May 28, 2026
AI Agents

The Pre-Ship Security Checklist for Vibe-Coded PRs

An AI agent resolved an RLS error by making the table publicly readable. CI stayed green. Here's the grep triage and checklist I run before any LLM-generated code ships.

May 22, 2026
AI Tools

pgvector Hybrid Search vs. Pure Vector: The Query Patterns That Flip the Winner

Pure cosine similarity returns garbage on version numbers, SKUs, and exact phrases. The fix is hybrid search in Postgres — two indexes, one CTE, and a rule for which leg to weight.

May 18, 2026
Workflows

AI Evals Are a Bottleneck: The Minimal Harness I'd Wire Into CI Today

Most LLM features ship on vibes — skim three outputs, merge. Here's the 40-line Python eval harness with JSON fixtures and a judge-model gate you can wire into CI today.

May 06, 2026

About

Vasileios Laios

I've spent 30 years designing and building software systems — from early enterprise architectures to modern distributed platforms. Over that time I've learned that the best solutions aren't the most complex ones. Clean architecture, clear thinking, and the right tool for the job. That's always been the philosophy. For the past few years I've been going deep on AI Engineering — not the hype, but the actual craft of building systems where AI agents do real work. Workflows that think, pipelines that decide, tools that act. I build these things, I write about them, and I share what actually works in practice. CyberDevTech is where I publish my thinking — on AI agents, software architecture, and the practical side of integrating intelligence into the systems we build every day.
LinkedIn — Vasileios Laios