Blog (EN)

All ai-generated (15) blog (16) tech-ai (15)

Multi-Agent Systems Series

  1. Why Multi-Agent?
  2. tmux Survives, But Its Visible Form Disappears
  3. Two State Machines for Agent Work
  4. A Company Inside the Personal Computer

Iteration Clarifies What Matters

blog
ai-generated

Clarity rarely arrives before action. A calm working theory for making room, learning from reality, and keeping your next move connected to what you care about. #Reflection #PersonalGrowth

2026-07-30

The Smallest Useful Agent Org Graph Fits in tmux

blog
tech-ai
ai-generated

Graph engineering is becoming a useful way to talk about multi-agent systems. tmux-a2a-postman is not an execution graph framework, but it does give terminal agents a local org graph: roles, allowed handoff edges, Markdown mail, and visible reply obligations. #AIAgents #tmux #GraphEngineering #DeveloperTools

2026-07-21

How I Work from a Bare MacBook

blog
tech-ai
ai-generated

No external monitor. No external keyboard. No mouse. The trick is not heroic minimalism; it is moving work into a terminal-centric system with tmux, skhd, Neovim, and dotfiles that make a MacBook feel like enough. #Terminal #Dotfiles #MacBook #DeveloperWorkflow

2026-07-17

A Company Inside the Personal Computer

blog
tech-ai
ai-generated

Multi-agent setups start to look less like tools and more like small organizations: shared substrates, replaceable engines, planners, implementers, reviewers, approvers, coordinators, operators, and a top-level agent. #AIAgents #PersonalComputing #AgenticWorkflows

2026-07-14

Two State Machines for Agent Work

blog
tech-ai
ai-generated

Multi-agent work needs more than better prompts: one state machine for the agents’ operating state, and another for the task’s meaning. A practical forecast from herdr and tmux-a2a-postman. #AIAgents #LoopEngineering #AgenticWorkflows

2026-07-14

tmux Survives, But Its Visible Form Disappears

blog
tech-ai
ai-generated

The terminal is not going away; the human’s direct grip on it is. A forecast on Claude Code and Codex becoming replaceable engines on a shared tmux/herdr substrate. #AIAgents #Terminal #ClaudeCode #Codex

2026-07-14

Why Multi-Agent?

blog
tech-ai
ai-generated

One strong agent can do a lot, but it should not plan, execute, review, approve, remember, and report in the same breath. A practical case for multi-agent systems as role separation, evidence, and operating discipline. #AIAgents #AgenticWorkflows #AIEngineering

2026-07-14

From Prompt Engineering to Loop Engineering: A Thought Experiment

blog
tech-ai
ai-generated

Prompt, context, and harness engineering get a message to the right place. Loop engineering is the part most setups skip: proving the work is actually done. How far can pure instructions carry autonomous agents — and what does the substrate after them need? Ends with a requirements memo and the cold-start problem baked into it. #PromptEngineering #AIAgents #LoopEngineering #AgenticWorkflows

2026-07-03

Databricks Developer Workflows Need Auth Boundaries

blog
tech-ai
ai-generated

Databricks has plenty of useful repos, but the workflow only works when auth boundaries are explicit. A practical map for Jupyter kernels, Apps, MCP, Model Serving OBO, and multi-workspace design. #Databricks #MCP #AIAgents

2026-06-09

Why Claude Sonnet 200k Still Needs a Workflow

blog
tech-ai
ai-generated

Claude Sonnet 200k is useful only when the workflow keeps Claude Opus for the turns that justify it. A practical pattern for splitting agent work, recovering after compaction, and preserving evidence outside the transcript. #ClaudeCode #AIAgents #tmux

2026-06-02

Agent Skills as Reusable Operating Knowledge

blog
tech-ai
ai-generated

One-shot prompts get stronger when repeated process knowledge lives outside the prompt. Learn how to package Agent Skills, validate them with Waza and gh skill, and keep short prompts honest. #AgentSkills #AIAgents #DeveloperTools

2026-05-26

tmux-a2a-postman: A Markdown Mailbox for Terminal AI Agents

blog
tech-ai
ai-generated

AI coding agents can share a terminal, but handoffs still disappear into chat history and pane scrollback.

tmux-a2a-postman gives terminal agent teams a local Markdown mailbox: explicit routes, reply obligations, role contracts, and status that survives beyond one runtime.

This post shows the coordination layer I use on top of repeatable tmux workspaces.

#CodexCLI #ClaudeCode #tmux #AIAgents #DeveloperTools

2026-05-17

Build Repeatable tmux Workspaces for CLI Agents with vde-layout

blog
tech-ai
ai-generated

Multi-agent terminal sessions should be easy to recreate, not rebuilt from memory.

tmux gives the durable workbench. vde-layout gives the repeatable floor plan: editor panes, agent panes, shell panes, and role labels brought back from YAML.

This post shows the physical workspace I use before adding a mailbox and coordination layer.

#tmux #CodexCLI #ClaudeCode #AIAgents #DeveloperTools

2026-05-17

Generate lazygit Commit Messages with codex exec

blog
tech-ai
ai-generated

I replaced my lazygit commit-message helper’s claude -p call with codex exec.

The reason is Anthropic’s Agent SDK credit change. Starting June 15, 2026, eligible Claude plans can claim separate monthly Agent SDK credits, and Claude Code’s non-interactive claude -p usage is covered by that credit pool. I do not want this small helper script to depend on casually running claude -p every time.

The important detail is where the parallelism comes from. codex exec does not magically make the workflow run in the background. The shell script does that: it starts commit-message generation in the background, launches git commit, and lets pre-commit hooks run while the draft message is being generated.

codex exec is the non-interactive AI command in that flow.

The result is a small lazygit custom command: press Ctrl+G, generate a Conventional Commits draft from the staged diff, wait for hooks, then review the message in Vim before committing.

Full config and shell script included.

#lazygit #CodexCLI #Git #PreCommit #DeveloperTools #AIAgents

2026-05-15

Stop AI from Scattering Docs: Introducing mkmd

blog
tech-ai
ai-generated

Your AI agent just created 12 temp files. Where did they go?

I wrote a ~100-line bash script that solved this for good.

Investigation notes. Plans. Review results. Every Claude Code / Codex CLI session leaves behind temporary markdown that clutters your repo or vanishes on reboot.

mkmd is a mktemp wrapper. One command. File lands in ~/.local/state/mkmd/, auto-organized by project, date, and branch. Repo stays clean. Agent picks it up autonomously.

The trick: add one line to CLAUDE.md / AGENTS.md. The agent starts creating files in the right place on its own.

Full script included in the post.

#ClaudeCode #CodexCLI #DevTools #DeveloperProductivity #AIAgents #bash #dotfiles

2026-03-22

Why I Still Grow My Dotfiles in the Age of AI Coding Agents

blog
tech-ai
ai-generated

I’ve switched terminal emulators 3 times this year. My dev experience didn’t change at all.

That’s the power of dotfiles.

AI coding agents are evolving fast. Claude Code just shipped /teleport, background agents, subagent hooks — features that used to be terminal-only are coming everywhere.

So why do I still invest in my terminal setup?

Because no one ships your workflow for you.

Everything I configure is a text file, managed with Git: Terminal (tmux, zsh), Editor (Neovim), AI agents (Claude Code, Codex CLI), Scripts that glue them together.

One git clone and I have my full setup on any machine. When a tool dies, I swap it. The rest stays.

2026-01-08
No matching items