Prompt guardrails fail right when coding agents get dangerous. How Docker Sandboxes make YOLO mode safe, plus a ready-made kit for infrastructure work.
Engin Diri
Manage PagerDuty teams, schedules, escalation policies, and services with Pulumi, and wire alerting to the AWS infrastructure in the same program.
Engin Diri
Google's Open Knowledge Format: a one-page spec for the LLM wiki. Markdown, one required field, git. Why knowledge as code is your agent loop's missing layer.
Engin Diri
A zero-touch Ollama GPU inference server on AWS, Azure, or Google Cloud as one Pulumi program, with OIDC credentials from Pulumi ESC and no static keys.
Engin Diri
Deploy a self-hosted Hermes agent as one Pulumi program across Render, Modal, and Tailscale — a code-executing AI agent with nothing on the public internet.
Engin Diri
The unit of work moved from the prompt to the loop. The five pieces of loop engineering, the memory that makes it compound, and what it won't do for you.
Engin Diri
AI coding has two shapes: 2x is mostly prompting, 10x is mostly plumbing. The parallel coding playbook, translated to Pulumi.
Engin Diri
In a large codebase, the model is the smaller variable. The harness around it does the work: CLAUDE.md, hooks, skills, LSP, MCP, subagents.
Engin Diri
A year ago, building an AI agent meant frameworks, RAG, and glue code. The Claude Agent SDK, Codex SDK, and skills replaced most of that middle layer.
Engin Diri
What the dark factory pattern looks like when the factory floor is your Pulumi state graph, and where to start without burning down a prod account.
Engin Diri
Page 1 of 5
