Research · 11 · 13 · 16 August
Copying week
Last updated: 2026-08-12
Confirmed timeline (August 2026)
❓ Who shipped which teammate idea first, and what was already old by then?
| Date | Event | Source quality |
|---|---|---|
| Years prior | Hermes: profiles, gateway “bots”, skills, cron, memory, Kanban | Official Hermes docs[8][28] |
| 2025-12-09 | AAIF: MCP, goose, AGENTS.md | Prior KG milestone |
| 2026-08-11 | Grok Bot early beta launch; persistent computer; named teammates; group chat; routines[1] | Primary |
| 2026-08-11 | Elon: widen beta after basic fixes and Grok 4.6 later that week | X lead (not used as SKU proof) |
| 2026-08-13 | Teknium: Bot Mode one-day public plugin beta for Hermes Desktop[22][23] | Primary (X + later docs) |
| ~2026-08-16 | PR #87886 bundles Bot Mode default-on; teammate protocol into core prompt (not SOUL.md)[23] | Primary |
| 2026-08-17–19 | v0.20.4 / v0.20.5: group-room threads, avatars, Bot Mode hydration, cron memory…[12] | Primary (GitHub releases) |
| 2026-08-21 | Grok Bot included with more plans[24] | Primary |
| 2026-08-25 | This overlay | — |
Hermes did not invent named teammates on 13 August. It put a Grok-like glass on primitives it already had. Grok did not invent skills, cron, or MCP. It productized a persistent VM + coworker UX and shipped it to SuperGrok/Cursor users in one SKU.
That is borrowing in both directions, on different layers.
What Grok appears to have taken from the open-harness world
❓ Which Grok Bot features are hard to explain without Claude Code / Hermes / OpenClaw existing first?
From official Grok pages, without claiming a lawsuit-grade copy:
- Named persistent teammates instead of a single chat session.[2][20]
- Message like a colleague / no workflow builder onboarding.[1]
- Skills as saved procedures;
/invoke.[18] - Routines as schedules/events.[18]
- Group chat of agents with a coordinator Bot.[1]
- Approvals on sends/purchases; least privilege language.[19]
- MCP + connectors.[3]
- SOUL-like role description (job + operational boundaries).[20]
These were in the water: Claude-class harnesses, OpenClaw, Hermes profiles, ChatGPT scheduled tasks. Grok’s unique combination is those ideas plus the always-on shared VM plus computer use for non-API SaaS.
What Hermes appears to have taken from Grok Bot (and the same water)
❓ What in Bot Mode is new to Hermes versus a UI remix?
New-as-product (Aug 13–19):
- Roster glass replacing session-first mental model.[21]
- Canonical forever-chat (
/new→/compact).[21] - Needs you badge; 2–6 Bot rooms with round caps.[21]
- Core teammate protocol injected into every Bot Chat, including headless CLI, so handoffs do not require editing SOUL.md.[23]
- Visual avatars / petdex as identity (Desktop craft).[21]
Not new: profiles, cron, @-routing via CLI, multi-profile gateway, Kanban.
The speed is the story: two days after Grok Bot’s public launch, Hermes showed a competing roster. A week later it was default-on in Desktop. Grok widened SKUs ten days after launch. Open-harness commit velocity (hundreds of PRs per tag) vs closed-SKU plan velocity (beta → more plans).[12][24]
Separate product: Grok Build
❓ Are we accidentally comparing a teammate SKU to a coding CLI?
Grok Build is SpaceXAI’s coding agent (TUI, headless, ACP).[14] Hermes is often used as a coding harness too. Fair peers for coding are Grok Build / Claude Code / Codex / Hermes CLI. Fair peers for teammate bots are Grok Bot / Hermes Bot Mode / OpenClaw-style rosters. This package is the second comparison. Mixing them inflates Grok’s “computer” as if it were a repo loop.
X and video as treasure, with a filter
❓ Which viral claims survive the docs?
Survives:
- Always-on when laptop closed.[3]
- Shared computer, own screens.[2][3][Cursor forum language matches docs]
- Teach / demo → skill.[18]
- SuperGrok/Cursor bundling.[4][24]
- Hermes Bot = profile.[21]
- People stacking Hermes inside some other cloud box (conceptually possible; not a Grok feature).
Does not survive as fact:
- “Each bot gets its own computer” (common blog FAQ). Official: one computer per member.[5]
- “Grok killed the parameter race because of Bot” — opinion.
- Dollar prices and “20–30 agents at once” — anecdotal X.
- SpaceXAI/Cursor “partnership” as a legal structure — Grok Bot uses Cursor accounts; deeper corporate form is out of scope.
X search synthesis is lead discovery only. Primary URLs above are the mechanism.
Pace classification
❓ Is this exponential shipping or a launch spike?
Classification: stepwise (product) + high-frequency craft (Hermes git).
- Metric: user-visible teammate features; Hermes merged PR counts per tag.
- Period: days-to-weeks in Aug 2026; Hermes compounding over 2026.
- Not exponential: no stable doubling-time of a capability metric here.
- Mechanism: looking-glass patterns are software and copy in days; hosts and isolation copy in quarters.
- Bottleneck: security review, identity (MDM/IdP), audit logs.
- Next paradigm: durable boards (Hermes Kanban) vs shared desktops (Grok) may merge — a host with per-bot tenants, or a board that owns a VM pool.
- Reach: consumer “digital employee” UX is no longer scarce; governed side effects still are.
Unresolved
Independent eval of overnight Grok routines vs Hermes cron-on-VPS. Teach-a-task availability fraction. Hermes Cloud desktop-vs-container roadmap.