Peter Zatloukal

Building toward autonomous coding loops.

[ linkedin | scholar | github ]

Writing

• Agent-First Repos
  October 2026 — Agent-first codebase + agent-aware data + agentic-devops = magic.
• Zero Times Infinity
  September 2026 — The economics of the global AI buildout: who pays, and how it ends.
• Agentic AI @ 2 FPS
  July 2026 — The border to local frontier-class AI is already crossed. The rest is frame rate.
• The Bitter Lesson of Agentic Coding
  April 2026 — Invest in verification and goal-setting, not implementation control.
• Agent-Hypervisors
  June 2025 — An approach toward self-optimizing agent fleets.
• LLM Inferencing Costs are Going to $0
  March 2025 (updated April 2026) — The marginal price of intelligence is trending toward zero.
• Cat Detection from Scratch
  May 2026 (based on writings from 2017) — What training a model from random weights actually felt like in 2017, before Hugging Face and foundation models did the work for you.

Projects

• zat.env
  The Bitter Lesson, shipped as a framework. Autonomous coding loops for Linux + Claude Code.
• OrgSmith
  Synthetic data generator for hallucinating entire companies as browsable file shares of realistic business docs & emails with a deterministic ground-truth answer key.
• daydream
  An agentic-first engine for interactive storytelling: frontier model writes the story at design time, and local AI on a single RTX 4000 GPU handles the game's reflexes, from dialogue to watercolor art (Jev optional).
• KX3-Rig
  Custom Linux rig for my Elecraft KX3 ham radio and mag loop antenna, including digital modes and smart scanning with automatic band condition awareness. Agent-first: a full emulator and a test suite let an agent build and verify new features with no radio attached.
• Blabacus
  Ambient AI that shows who's doing all the talking. State-of-the-art speaker diarization using NVIDIA Nemotron 3, runs on the Apple Neural Engine so nothing leaves your Mac.