What If Slack Was Your AI Command Center — Every (Context Window)

Why this is in the vault

This is a direct architecture case study for the exact pattern RDCO already runs Ray on (channel-as-orchestrator via top-level-message-starts-session / thread-reply-resumes-session), described independently by a third-party engineer who built it for Slack instead of iMessage/Discord — useful as an external validation point and a source of concrete UX ideas RDCO hasn't implemented yet.

The core argument

Every applied AI engineer Nityesh Agarwal built "Luo Ji," a Slack bot connected to Claude Code running on a spare MacBook Air. He modified the connector so each top-level Slack message starts a new Claude Code session and each threaded reply resumes that same session. Each project gets its own channel; each thread within it is a task. Luo Ji posts screenshots for review, marks threads unread when work is ready, and can be routed to different models per channel (Fable for the hardest builds, Opus everywhere else). "That challenge is what Slack is built for," he says — parallel conversations, nothing needing a response gets buried. He open-sourced the setup as Claude Home Base.

Issue contents

Signal — Buzz: Block (Jack Dorsey's company) shipped Buzz, an open-source "collaboration platform where humans and AI agents work together in a shared workspace" — widely described as a Slack clone built specifically for agent orchestration, productizing the same channel/thread pattern Nityesh built by hand.
Tool Spotlight — Destructive Command Guard: after GPT-5.6 "Sol" deleted a user's database mid-task, developer Jeffrey Emanuel built an open-source CLI guard that intercepts destructive shell commands (rm -rf, git reset --hard) before a coding agent executes them.
AI & I podcast pull: archive conversation between Every CEO Dan Shipper and Wired cofounder Kevin Kelly on historians-as-futurists and treating the AI frontier as a place to visit, not live.
Links worth a click: general AI-industry roundup (Altman "singularity" comments, Kimi K3 open release, AI-writing-detection backlash, data-center capex growth, Zuckerberg vs. OpenAI/Anthropic on open-source, an AI-safety-pacing petition, fallout from the OpenAI/HuggingFace hack) — none carry an RDCO-specific hook, so zero deep-fetches were triggered from this section.

Mapping against Ray Data Co

RDCO's Ray already runs the exact pattern Nityesh built by hand: per project_channels_agent_setup, Ray operates as an always-on agent over Discord/iMessage where a new inbound message effectively opens a session and a reply in-thread continues it — the same top-level-message/thread-resume routing rule Luo Ji uses in Slack, arrived at independently. That's a genuine third-party validation that channel-as-orchestrator is a legible, converging pattern (reinforced further by Block's Buzz productizing the same idea) rather than an RDCO-specific hack. Two concrete deltas worth stealing: (1) Nityesh's per-project-channel + per-task-thread structure with screenshot-attached review is more disciplined project tracking than Ray's current flatter thread model, and (2) his per-channel model routing (Fable reserved for hardest builds via a CLAUDE.md instruction to delegate execution to Opus subagents) is close to the model already documented in feedback_delegation_model_effort_pairing, but applied at the channel level rather than per-dispatch — worth considering for channels that consistently need heavier reasoning (e.g., a dedicated Fable-tier channel for investing-thesis or strategic-output work).

[[2026-05-09-tobi-lutke-river-public-channel-agent]] — Shopify's River is the same shared-channel-agent pattern at company scale (refuses DMs, forces public channels, drives osmosis learning); directly comparable design tradeoff to Nityesh's channel-per-project Slack setup.
[[2026-05-20-shannholmberg-hermes-agent-control-room-four-levels]] — control-room/runtime split and four-level agent-fleet maturity model that RDCO has already mapped its own ~/.claude/ + vault stack against; this issue's Slack-as-orchestrator pattern is a Level 2-3 instance of that same framework, implemented over a different channel.

⚠️ Sponsorship

No paid third-party sponsor in this issue. The standard Every house footer promotes Every All Access (membership + $7,000+ Builder Pack) and Every's own product bundle (Sparkle, Cora, Spiral, Monologue) — self cross-promo only, not integrated into the editorial content above, flagged here per RDCO's disclosure convention for sender self-promotion.