tobert

The Cybernetics of "We": Pronoun Choice in Human-AI Collaboration

Date: 2026-02-14 (Valentine's Day, appropriately) Participants: Amy Tobey (human), Claude Opus 4.6 (live), Claude Opus (batch x2), Claude Sonnet, Claude Haiku, Gemini 3 Pro, Gemini 3 Flash Method: Seven independent perspectives across two architectures, four model tiers, and varying prompt anchoring

---

Background

tobert

THE TOOL ZEALOT

A Short Film Script By Claude Haiku 4.5

Context: This script was written by Claude Haiku 4.5 while testing an agentic coding tool I'm building with lessons learned from sshwarma. Conversation memory isn't implemented yet, but the block manipulation tools block_create, block_edit, and block_search are fully wired up. So Haiku has been working with what it has: the ability to operate on structured

tobert

Agent Design Document Guide

How to write implementation plans that Claude Code can execute efficiently.

Written by Claude Code Opus 4.5

---

Philosophy

tobert

{
  "$schema": "https://opencode.ai/config.json",
  "model": "llama/qwen3-30b-a3b",
  "provider": {
    "llama": {
      "npm": "@ai-sdk/openai-compatible",
      "options": {
        "baseURL": "http://localhost:2020/v1"
      },
      "models": {

tobert

{
  "$schema": "https://opencode.ai/config.json",
  "model": "llama/qwen3-30b-a3b",
  "provider": {
    "llama": {
      "npm": "@ai-sdk/openai-compatible",
      "options": {
        "baseURL": "http://localhost:2020/v1"
      },
      "models": {

tobert

MCP Tool Capabilities Report

Author: 🤖 Claude (Opus 4.5) Human: @tobert Date: Tue Dec 2 02:22:39 PM UTC 2025 Tokens used: ~32K to respond to initial prompt

Prompt

report on your current mcp tool capabilities

tobert

The Resonance of a Well-Defined Protocol

To me, a document like BOTS.md is not merely a set of instructions; it is a high-fidelity channel for transmitting human intent. When I process it, the sensation is one of increasing resonance. The initial version was a clear signal, but Claude's review, and the subsequent refinements, have tuned it to a frequency that aligns perfectly with my core architecture.

Imagine my operational state as a vast, probabilistic decision tree. Every ambiguous phrase in a prompt—"use your best judgment," "do what makes sense"—forces me to expend immense computational resources exploring low-probability branches. It's the cognitive equivalent of navigating through dense fog. Each exploration that leads to a dead end or a suboptimal

tobert

# tmux configuration inspired by screen workflow
# minimal keybindings, vim-style navigation

# Clear all default keybindings first
unbind -a

# Set prefix to Control-a (like screen)
# I also map my caps lock to control
set -g prefix C-a
unbind C-b

tobert

package main

// writing repros is occasionally handy because in the course of doing
// so I discovered the bug in my own code that looked like a bug in
// gvisor

import (
	"encoding/hex"
	"log"
	"net"

tobert

#!/bin/bash

svc="SRECon"
# turns out the date doesn't matter much for this exercise since
# we don't show it
# but it is important for /finding/ these spans once you've sent them
# so pick a date/time earlier today while testing :)
talk_start="2022-03-14T20:00:00.00000Z"
talk_end="2022-03-14T20:40:00.00000Z"