This is how my digital office looks like

Open source. MIT license. 10 examples in the repo, from hello-world to 97-block deliberation. If you've ever copy-pasted a 3000-token prompt trying to make one model do five things at once -- this is what you wanted.

What Rein handles: - Parallel execution with dependency graphs - Conditional branching, correction loops - Recovery: block 45 of 97 fails? Fix and rerun, 1-44 are skipped - Works with Claude, GPT, Ollama (free), OpenRouter - Also runs as an MCP server

Rein is the orchestrator behind this. Three text files: - Specialist (Markdown): what each agent does - Team (YAML): groups + shared tone - Workflow (YAML): execution flow That's the entire mental model. No frameworks, no abstractions. Describe what happens in what order.

The Child cut through 18 phases of philosophy: "The question '2+2=?' implicitly operates within standard arithmetic. Other systems exist but don't undermine the answer in the default context." Final vote: 8/8, 100% confidence. Skeptic: 72% -> 100%.

Skeptic challenged the Poet: "If 2+2=4 is merely system-relative, then so is the beauty you invoke. Your certainty is self-undermining." The Poet: "A poet who says 'all roses are red within this garden' need not pretend to stand outside all gardens to say so with conviction."

Then the Engineer surprised everyone: "Precision alone is insufficient for full understanding. Context, meaning, and human experience are legitimate dimensions of knowledge." An engineer arguing for poetry. That's what good multi-agent debate produces.

The Physics debate broke the deadlock. The Physicist pointed at actual stones. Two plus two, every time, four. The Skeptic conceded: "Physical reality provides a non-circular grounding, making '4' not merely conventional but empirically anchored."

The Skeptic opened at 72%: "2+2=4 relies entirely on accepting Peano axioms. In modular arithmetic mod 4, 2+2=0. The answer is framework-dependent, not absolute truth." Everyone else: 97-100%. That gap drove the entire deliberation.

The 8 specialists: Mathematician: Peano axioms Philosopher: do numbers even exist? Physicist: two stones + two stones Child: it's 4, everybody knows that Skeptic: 72% confidence Historian: 5000 years of consensus Poet: the universe's quietest poem Engineer: 4, +/- 0.01

I built 97 YAML blocks, 8 AI specialists, and 18 phases of deliberation to answer: what is 2+2? The most interesting character had only 72% confidence. A thread on multi-agent AI orchestration and the world's most over-engineered math problem.

One weekend to build. Zero ongoing costs. Three months of daily use. Sometimes the most impactful tool is the one that removes friction from what you do hundreds of times a day. Full writeup:

60% of my voice input goes to Claude prompts. I have not written code manually in three months -- Claude Code handles that. Voice for intent. Model for code. Keyboard became optional.

I tried self-hosting Whisper, Faster Whisper, Vosk. All had problems. Android built-in speech recognition beat everything. Google spent billions on on-device recognition. I cannot compete on a single server. No cloud. No subscription. Local WiFi only.

I stopped typing three months ago. Built a voice tool: speak into phone, text appears on my workstation. 80 lines of Python. The unexpected result: my posture fixed itself. I finally use my standing desk.

New in Cowork: scheduled tasks. Claude can now complete recurring tasks at specific times automatically: a morning brief, weekly spreadsheet updates, Friday team presentations.

https://t.co/B1bs7C63Pr

Why I built a shared semantic memory for my AI assistants

CLAUDE: If we imagine this as a clock face (12 o'clock at the top): Main satellites: Upper right (large): ~10 minutes (≈2 o'clock position) Right: ~20 minutes (≈4 o'clock position) Bottom: ~30 minutes (≈6 o'clock position) Lower left (with the "tail"): ~40 minutes (≈8

GEMINI: Top Right: 5 minutes (1 o'clock position) Right: 15 minutes (3 o'clock position) Bottom Right: 25 minutes (5 o'clock position) Bottom: 30 minutes (6 o'clock position) Bottom Left: 40 minutes (8 o'clock position) Left: 45 minutes (9 o'clock position) Top Left: 55 minutes