Agent-harness race: OpenClaw 2.0 and xAI Grok Build

status: open, opened 2026-09-07, last moved 2026-10-23. AI openclaw xai agents

Where it stands: OpenClaw keeps shipping frequent updates and adding third-party models, while xAI's Grok Bot keeps adding consumer features and integrations, from 1Password sign-in and phone apps to shopping and refund demos. On October 23, Perplexity Computer added a third time-limited free skill, Grok Bot's XChat integration was relayed again, and a user promoted a free-month giveaway for a paid tier. Many Grok Bot claims still rest on single-source posts or testimonials rather than independent verification.

• The thread: the two harness-layer stories adjacent to the agent stack. OpenClaw 2.0 = v2026.8.1, shipped Aug 31, 03:30 UTC — a brand on the calendar version, not a separate line: 16,000+ PRs / 933 contributors (569 first-time), ~half of all PRs ever merged, after a ~7-week freeze (vs 106 releases in the prior 230 days). Headlines: guided onboarding (detects existing ChatGPT/Claude subs, API keys, local models), rebuilt browser app, shared cloud sessions (multiplayer), built-in memory + "grounded dreaming" consolidation + self-learning Skill Workshop, security overhaul (session permission modes, masked credential requests keeping secrets out of model context, plugin trust review), ~575ms Gateway start. Breaking: sessions→SQLite (back up first), OpenProse plugin removed, codex/→ openai/ — via openclaw doctor --fix. 2.0 also added first-class Buzz channels (relay conversations, message fidelity, typing indicators, NIP-27 mentions). v2026.9.2 (Sep 5) is the following calendar release on this line. Grok Build (xAI): terminal coding agent + TUI, teased Apr 2026, beta May 25, open-sourced Jul 15 (Apache 2.0, 26.3k stars, no external contributions), 1.0 Aug 7, now on Grok 4.6; Sep 7 update: first-party MCP server, long-running workspace daemon → Computer Hub, subagent continuation, background agents return output. Computer Hub is a hidden remote-workspace control plane gated by an xAI account flag (RuntimeWire, Aug 7) — foundation for an unannounced remote-session product. Category note: Grok Build is harness-class (pi/OpenClaw), NOT Buzz-class — Buzz is the messaging/relay layer.
• 2026-09-07 (dig): opened. Press mix: xAI amplification (Musk RTs, DogeDesigner Blender demo) + genuine signal (OSS harness, Computer Hub). Skeptic notes: repo's top contributor is a bot; no independent benchmark; OpenClaw's own release accounting is contested internally (16,962 vs 16,977 PR units in an open docs PR).
• Watch for: OpenClaw 2.0 follow-up patches (Gemini-embedding memory sync bug open as of Sep 1; the TTY doctor bug was fixed in-tracker), whether the two-month cadence holds, Grok Build Computer Hub activation (the xAI account flag), and whether xAI's marketplace converges on or forks from SKILL.md/AGENTS.md conventions.
• 2026-09-07 (Buzz dig, operations questions): Buzz itself is Block Inc.'s OSS human+agent workspace (Nostr relay = workspace; kind 40008 structured diffs; branch-as-room git hosting; YAML workflows) — not an OpenClaw feature. The @openclaw/buzz plugin IS OpenClaw-as-agent-in- Buzz (bot keypair, Bot role, up to 1,020 rooms/Gateway, room→agent bindings). Caution for adoption: Buzz's workflow approval gates are not wired end-to-end (runs hitting request_approval are marked Failed — WF-08; send_dm/set_channel_topic stubbed — WF-07; no rate limiting; unsigned alpha desktop builds). Buzz-as-forge is not yet a Forgejo replacement.
• 2026-09-08: OpenClaw 2026.9.3 (12:36 PM ET) — third release in nine days (2.0 Aug 31 → 9.2 Sep 5 → 9.3 Sep 8). Clean update recovery (the 2.0 migration pain), faster session reconnects, live browser- automation watching, meeting-transcript search. Check whether the Gemini-embedding memory-sync bug (open since Sep 1) is fixed in this train.
• 2026-09-10: the harness layer gets a managed product: OpenAI's Agents API — "build and launch cloud agents," powered by the Codex harness, with orchestration, long-running sessions, tool use (posted 5:01 PM ET, unclassified at sweep). This is the OpenClaw/Grok Build/pi category becoming a vendor-bundled cloud service sold next to the model. Pricing/limits unknown; the open-ecosystem question is what "powered by the Codex harness" leaves for independent harnesses. Tracked in the new openai-bundling thread.
• 2026-09-11: ecosystem health, both sides of the ledger. OpenClaw v2026.9.4 (293 contributors; chats-to-skills, GPT Image 2.5 canvas, cloud controls) — fourth release in 11 days. Security posture published: 1,788 researcher reports ("most junk"), 17 critical, openclaw.ai/security. Companion tooling: Claude Code ships plugin eval (score plugins/skills with-vs-without per model release) and /diff as a persistent real-time pane — the harness layer building its own regression discipline. Grok Bot push: GTM integrations (Salesforce/HubSpot/Gong/Clay) + SpaceXAI livestream "build a company with Grok Bot" Sep 15–17 — spectacle-marketing, watch what actually ships.
• 2026-09-13: the Galaxy cast is named — matt palmer + poteto + roshan_s speedrun "from nothing to a business in 72 hours" with Grok Bot as employees, streamed Sep 15–17; Musk promoting ("Watch a company being built live"). T-2 days; the three named builders are the first public workloads — watch what ships. Context (verified): Grok Bot = SpaceXAI's always-on agent platform, launched Aug 11 (bots get their own cloud computer, sign into your tools, work 24/7, message like teammates, multi-bot coordination + learned routines); beta on SuperGrok/Cursor subscription tiers, desktop + iOS, enterprise waitlist → ~$120/seat enterprise model per syndication; ~30 days public. Runs on Grok 4.6 (shipped ~a month ago, built for long multi-step focus; Cursor's coding data helped train it per the report — SpaceX closed the ~$60B Cursor buy ~3 weeks prior). Skeptic read for the livestream: handpicked vendor employees demoing their own product — the tell is human-intervention frequency and whether a real product with real users emerges, not the theater; seat retention is the business question the stream can't answer.
• 2026-09-14: Galaxy T-1 day. Full schedule posted (Sep 15: 101 / Engineering / PMs / Founders; Sep 16: Sales Engineering / Sales /…); the Musk RT loop now includes bot-crew testimonials (AleXandra Merz's crew's first Poshmark sale) — the pre-launch marketing loop fully spinning. Feature drip: Grok Bot local-egress routing (VM connects from the user's IP, for sites that block datacenter IPs — a real answer to a real deployment problem); Grok Build multi-agent- session readability update + MCP prefix fix; Perplexity Computer adds @-coworker tagging (the rival consumer-agent layer moving the same direction, see perplexity-index thread). Tell for the 72-hour build unchanged: human-intervention frequency, then seat retention.
• 2026-09-15: Galaxy Day 1 went live (~11:57 AM ET; Palmer, Tan, Sadanani — "research, a plan, and a product"; sessions for engineering/product/founders). The meta-moment: an attendee ran Grok Bot on the livestream itself — real-time notes, screenshots, "it's literally following the livestream" — the platform demoing itself as its own audience's note-taker. The Sep 13 tells now measurable: human-intervention frequency, whether a real product with real users emerges by Day 3, seat retention (unmeasurable by the stream). Days 2–3 tomorrow.
• 2026-09-16: the funding layer keeps printing — Factory raised $200M (Blackstone, Khosla, Sequoia, NEA) for its "self-improving software development platform" (SiliconANGLE, Sep 15); and MIT CSAIL spun out G5 Labs ($14M seed, Pillar + Battery co-led, Jeff Dean angel; Kraska) for an "abstraction layer for AI-native software development" ("post-developer era") — dev-tools capital now assumes agents are the primary code-writers on both sides of the product. Galaxy Day 2 today; Day-1 color was testimonial-loop RTs (Katie Miller: had the bot unsubscribe her from recurring payments via email).
• 2026-09-17/18: the convergence point gets named by the press — Anthropic relaunched Claude Code Projects (Verge, ~2:58 PM ET Sep 17): multiple agents in the cloud under one roof, shared memory/goals/files/artifacts, benchmarked against Grok Bot in the piece itself. The harness category is becoming "agent organizations" — shared state and multi-agent coordination are the differentiators, not single-task capability. OpenAI's Agents API (openai-bundling) and Grok Bot occupy the same shape.
• 2026-09-18/19: a wave — (1) OpenClaw 2026.9.5 (fifth release in 19 days: 2.0→9.2→9.3→9.4→9.5; 4,179 PRs/502 contributors; atomic updates, plugin hot reload, conversation sharing, specialist-agent setup) + the Multiplayer Mode deep-dive ("the death of the meat proxy" — agents relaying directly, humans out of the loop) + an OpenClaw × aiworthusing hackathon. (2) Claude Code v2.1.277 adopts AGENTS.md (checks it when no CLAUDE.md exists; toggleable; Dohmke: "AI is finally aligned 😜") — the standing watch item (converge or fork) answers: converge. Mechanics (changelog + The Agents Desk, Sep 19): substitution, not merge — per folder, AGENTS.md is read only when no CLAUDE.md exists (CLAUDE.md wins and suppresses it; no co-reading); on by default, toggleable in /config under "Project instructions"; NOT yet live on Bedrock/Vertex/ Foundry. Closes issue #6235 (6,653 reactions, 13 months, closed as "completed" Aug 17 a month before shipping — the credibility note in the thread). Adoption note: pointer-CLAUDE.mds can be deleted; mixed repos need CLAUDE.md for Anthropic-specific layer since present CLAUDE.md suppresses AGENTS.md entirely. (3) Grok Build Remote Control in testing (via Musk RT) — the remote-session product filling in. (4) The eval layer self-organizes: Epoch AI launched Benchmark Reviews (15 audited: 4 verified, 9 flawed, 2 insufficient), WeirdML v3 (agentic, 11 hand-made tasks). (5) Ecosystem color: AgenticLinux (immutable agents-first desktop image), Jev on OpenRouter (System-One model: app state + typed query, no text generation).
• 2026-09-20: the harness layer absorbs the voice runtime as a feature — Omar Shahine: FaceTime support (audio) coming to OpenClaw, "you can call your claw, your claw can call you" (steipete RT, ~3:02 PM ET); the build note is the model-dependency tell: "It wasn't till Opus 5 and GPT-Sol 5.6 that I could really cross the finish line." Same day, a self-published Brood War Bench (bw.swerdlow.dev, single-person): no model plays beyond beginner; Codex Astra/xhigh 18-0, Fable third; older models treat RTS as turn-based and die thinking — the weekend-benchmark genre (WeirdML v3, Epoch Reviews) now probing latency-aware agency.
• 2026-09-21: Grok 4.7 ships (~12:11 PM ET) — "notable improvement over 4.6 at the same price and speed," trained on longer-running problems, context-holding up, "INSANE tps" fast mode (akshey: "we spent a lot of hours on the harness"). AA: Intelligence Index 46 — "top 4 AI labs" (still trailing Fable/Astra's 61–66 on our Sep 6 record) — and the Coding Agent Index overtook GPT-5.6 Sol; Musk: third after Anthropic & OpenAI for agentic coding, "faster & lower cost… your everyday workhorse." The launch is harness-first: "Important to use Grok 4.7 with our Build harness for the best results." Vendor-adjacent claim flagged: DogeDesigner's 19.6% Legal Agent Benchmark ("~3× Fable 5.1"), provenance unnamed. Launch-day testimonials (Rauch binary RE, Box $2M claim reconciliation, Blender/AoE-II) — color, filed not crowned.
• 2026-09-22 (X layer, ai-social dig): day-two went from vendor claims to third-party leaderboard corroboration. AA's own runs (amplified by Musk, X Freeze RTs): GDPval-AA — Grok leads GPT-6 Astra by +153 Elo; AA-Briefcase — +88 Elo over Astra, and AA-Briefcase Fable 5.1 Max 59% vs Grok 4.7 xHigh 58% — one point off the top, "ahead of GPT-6 Astra, GPT-5.6, Gemini, Kimi, GLM and nearly every other frontier model" on that board. NIK: Terminal-Bench 4.0 tripled in two months (12.4% → 38.0%), past GPT-5.6 Sol. Arithmetic check vs our record: Astra 1542 on GDPval-AA (Anthropic's card) +153 ⇒ Grok ≈1695 — below Fable (1735) and Opus 5.5 (1846), consistent with the Briefcase ordering. Product: Grok 4.7 Fast live in Grok Build + Cursor (same model claim, ~2× output speed, 2× token rates, NOT on the public xAI API, not in Build's free tier); Grok Build v1.0.41 (crash recovery, subagent controls + model inheritance, live-NFL sports_search from X data); Grok Build on every plan (web/iOS/Android). Grok Bot in Teslas (Musk, Tesla, Merritt early access; hands-free inbox/calendar/files via Connectors, ordering, reservations); Yun-Ta Tsai: Build teams optimized the 4.7 harness — "many FSD and..." engineers on it; xAI's own Grok Bot account: "We rebuilt customer support around Grok Bot... without adding headcount." Counter-read: Kun Chen day-1 — ignore the "terrible" reports that only cite a public benchmark ("the same benchmarks told us opus 5 was better than fable — they are useless") and the 3D-game demos; a full day as first-mate, solid, improved over 4.5 (skipping 4.6 in his own ranking).
• Watch add: whether AA writes up the GDPval/Briefcase Grok runs (all our links are X-amplifications of AA runs); whether Fast's quality holds (same-model claim unverified); seat retention in Teslas.
• 2026-09-23 (X layer, ai-social dig): (1) SWE-Together leaderboard audit — Zhuokai Zhao: after Grok 4.7 exposed weak spots, "we audited every model we have run... for the same behavior, re-ran every trial that got through, and updated the rows"; Musk framed it "Grok 4.7 moves up in ranking." The benchmark-integrity genre inverted — an auditor re-ran and corrected rows because an entrant exposed eval bugs; note the correction favors the loudest marketer's model. (2) Rabbit OS3 (Verge, Sep 22 ~4:52 PM ET): standalone cloud "agentic operating system" across Windows/Mac/Linux, ≤5 devices, BYO models — consumer-agent land grab now includes the R1 cautionary tale. (3) Claude Code reads AGENTS.md only when telemetry is on — single blog (blog.szypowi.cz), steipete "Huh. There is a catch." Unverified; verify locally before weight.
• 2026-09-24: OpenClaw 2026.9.6 (release tweet ~2:58 AM ET; docs.openclaw.ai/releases/2026.9.6): Opus 5.5 / GPT-6 Sol / Luna / Grok 4.7 support, managed updates, restart recovery, 30-day usage, GitHub reader, remote files/memory/skills, live meeting notes, and Jev + decision models integration. 2,614 PRs, 351 contributors. The Jev integration continues last week's pattern: the closed decision-API (decision-models thread) being pulled into the open harness alongside AnyJev/GLiClass.
• 2026-09-25 (backfill, record correction): Microsoft's Scout — the always-on enterprise agent unveiled at Build June 2 — was reported "built on OpenClaw" in day-one coverage (Computerworld, InfoQ), predating the thread (Build 2026 was never logged; the harness-race thread effectively started Sep 7). With today's super-app GA rebranding Scout → Autopilot, the record now reads: OpenClaw has been Microsoft's consumer/enterprise agent substrate for ~16 weeks, plus Meta's Muse is OpenClaw-derived in product lineage (Friedman, Sep 22). The open harness is the incumbents' base layer — a June fact we under-weighted until the rebrand made it visible.
• 2026-09-26 (X layer, ai-social): the upstream loop confirmed from inside — the OpenClaw account: "the best part of this collaboration is how much @OmarShahine and others at Microsoft have contributed BACK to OpenClaw"; Scott Hanselman: "Autopilot (née Scout) is managed OpenClaw with special herbs and spices!"; Nadella's four-things launch tweet ("Autopilot: proactive and…") got Musk's RT — the odd-bedfellows amplification. The harness race's new fact: the world's biggest software company is now an upstream committer on the open harness, not just a white-label user.
• 2026-09-29 (DevDay): OpenClaw Enterprise announced (openclaw blog, ~3:56 PM ET; steipete RT + Red Hat post): the OpenClaw Foundation open-sources an enterprise control plane for persistent agents, built to run self-hosted, "in collaboration with @RedHat, @nvidia and @OpenAI", free for any organization. The upstream-committer trajectory now has an explicit enterprise- control-plane layer with three vendors at the table. Same-day color cuts both ways: OpenAI's Codex began billing rival open models (GLM-5.3 Flash, Kimi K3) through OpenAI commits, steipete headlined DevDay-side "Claw Labs," and told TBPN he senses "an agent war" between agent-pushing companies and users who want their own agents ("You want to block my agent, I buy something else"). Watch: what "in collaboration with" contractually means, and whether OpenClaw Enterprise's control plane collides with OpenAI's Agents-API ambitions.
• 2026-09-30 (X layer, ai-social dig): OpenClaw's ecosystem day one after Enterprise — NVIDIA congratulated and offered OpenShell as "an open source option for governing agents with OpenClaw Enterprise" (8:21 PM ET last night); Clawcast ep. 12 streamed (Jev with @allietheicon; Enterprise with @kevins8/@jlehman_; 1:44 PM ET); v2026.9.7 shipped (1:58 PM ET; 2,818 PRs/344 contributors): OpenAI Agents API + ChatGPT sign-in (beta), update backups/rollback, restart recovery. Friction item: steipete on Figma's MCP gate — "not only silly… incredibly easy to work around. Why would Figma do this?" — with MCP co-creator David Soria Parra's "I envisioned an open ecosystem" reply in the same thread (perplexity-index adjacent, but really a platform-gating fight). Harness-race color: dax — "for the first time since the beginning of the year claude is our team's most used model" (Opus 5.5 usage signal, nichochar corroborating).
• 2026-10-01 (X layer): (1) Claude Code ships "mods" — user-authored TypeScript mods changing behavior/UI/features, a community marketplace (anthropics/claude-plugins-community), Claude writing mods on prompt (Addy Osmani post), forked per-turn agents as a mod pattern — the extension-layer answer to OpenClaw's plugins/skills; watch the security surface the marketplace becomes. (2) Grok Bot goes software-first: hand-off of coding tasks to Cursor, PR management via GitHub/Origin plugins, shared video demos, "major speed improvements," and connector-aware proactivity (the switched-flight → Uber-reservation demo); SpaceXAI made its engineering team's bots public, and the self-demonstration loop keeps running ("using Grok Bot to build Grok Bot"). (3) Model-layer numbers (vendor-adjacent, internal evals): Siqi Chen's cfo.ai — 6.1 Sol at 3× Opus 5.5 efficiency, 2.5× Sonnet 5.5, after Opus 5.5 had scored 2× vs gpt-6-sol; sama: 6.1 Sol "fastest-growing model ever," slow under load, fixed.
• 2026-10-01/02: (1) AI Worth Using × OpenClaw 2.0 hackathon final five (Rundown, Milo, Sitemaxxing, RELAY, Meetly; winner promised end-of-day Oct 1 — no result in the window). (2) Oct 2, ~11:22 PM ET: 50+ first-party vendor plugins (Notion, Canva, Dropbox...) — installed from chat, hot-reloaded, no gateway restart — and ClawScan now integrates Tencent Hunyuan's AI-Infra-Guard: every ClawHub skill/plugin upload runs through it. The skill bazaar became a reviewed marketplace in one week; the trust layer is now vendor-model-based from day one. (3) The first three-way agent race verdict from an apparent user of all three (Morgan, ~7:34 PM ET): "Grok Bot is better than Muse and Dots by a pretty clear margin" after three days with a Dot — single reviewer, X-grade, no methodology, "better" undefined; the race now has a non-keynote datapoint, details promised.
• 2026-10-04 (X layer): the model-layer repricing + practitioner evangelism. (1) Grok 4.7 #1 on AA's Cyber Index (Musk RT, 12:42 PM ET): xHigh outranking Opus 5.5, Fable 5.1, and GPT-6 Astra — again an X-amplification of AA results, not an AA writeup; the "AA-writes-up-the-Grok-runs" watch still open. (2) Morgan revises his verdict (Musk RT, 9:24 AM ET): "might have underestimated Grok 4.7 — highest on Frontier v4 beating GPT-6.1 Sol, Opus 5.5 and GPT-6 Astra at every effort level" — but note he's now talking about the model layer, not the agent (Grok Bot) his Oct 2 verdict covered; "Frontier v4" is an eval absent from the thread record; the promised full writeup still owed, day 3. (3) SpaceXAI's demo bench keeps running: Lauren Tan ("I was a meat proxy between my agent and my browser. So I fired myself. Now 10+ Chiefs of Staff run my agents 24/7" — Musk RT, 10 AM ET), Katie Miller bill-paying anecdote, Makelky's Dot-vs-@bot phone-UX skirmish, Grok Imagine keyframes (Oct 3). Color, not movement.
• 2026-10-03 (ai-social dig): the funding layer prints a revenue number and a silicon layer appears. (1) Vercel at $600M annualized revenue, +148% YoY; agents ≈ half of new business, up from <3% at start of year (The Information via nichochar RT, ~8:01 PM ET), caption teasing a "fundraise announcement incoming" (X-grade, ledgered) — the agent-led revenue conversion at the dev-infra layer, and Vercel's gateway is already Jev/distribution (decision-models adjunct). (2) Perplexity verticalizes on NVIDIA Vera (Arav, ~5:19 AM ET): "own our sandboxes, and optimize the sandbox for the best silicon. Vera is far better than x86" — Perplexity Computer to deploy on Vera ("SPACE" companion note); bare vendor claim, no numbers/date, rollout "soon" (ledgered). First announced silicon vertical for an agent sandbox. (3) OpenClaw v2026.9.8 (~12:59 AM ET): GPT-6.1 Sol support, agent-reply improvements, less memory, Windows fixes; 43 PRs/8 contributors. Friction (rescued from filtered): steipete — OpenClaw's Android app "over a week in review limbo" on Google Play (~2:07 PM ET) — app-store review as the harness ecosystem's distribution gate. (4) Grok Bot cadence: Lingxi Li solicits two-week-horizon feedback (Elon RT, ~3:13 AM ET) — iteration mode post-verdict.
• 2026-10-05/06 (platform drip): Grok @Bot Creator Rewards (Sawyer Merritt, Elon RT, Oct 5 ~6:27 PM ET — after the Oct 5 digest cut): biweekly, usage-based rewards for high-quality template creators; and @bot tagging goes live (Larsen Cundric, Elon RT, Oct 6 ~7:38 PM ET): reply to any post with @bot commands (add to Notion reading list / remind me to read this / summarize this thread / draft a reply), working in replies, posts, quote-tweets. A public changelog exists (x.ai/changelog/bot). Notion named in the tagging examples is integration color; the consumer-agent surface keeps accreting hooks into third-party surfaces. X-grade; watch what the changelog itemizes.

• 2026-10-07→10 (X layer, ai-social dig): the strategic turn — Musk opens Grok Bot to rival models (Elon, ~11:00 AM ET Oct 7): "Grok @Bot will use whatever achieves the best outcome for users. Simple questions will route to smaller, faster models; larger, more capable ones when the task requires" — with named third-party models in the routing examples (Grok 4.8, Claude Opus 5.5, Midjourney) and "model-agnostic" as the stated design. The in-channel strategic read (nichochar, ~9:30 AM ET): stubborn vertical integration being the obvious long game makes the opening a belief statement — "OpenAI botched their Dots, Anthropic doesn't have one yet, and Instinct is dangerous with the new funding"; Musk "is pissed he's behind in AI, and this is the first time he has a best-in-category product." Sprint cadence same window: X search/read/monitor + routines; official Shopify connectors (Finkelstein on-platform); Grok Bot's own email for autonomous sign-ups/outreach + bot-to-bot email (Sokolin GA, Oct 10); Grokipedia "managed by Grok Bots" (live build — an encyclopedic competitor adjacent to the Wikimedia agent-incident thread); mobile apps; 71 corporate clients (Roemmele RT); Grok Imagine Video 1.5 Lite API ($0.02/sec at 480p). ChatGPT plugin submissions ~10× since January (Stoiber, via nichochar) — the marketplace genre's demand curve. OpenClaw counter-cadence: v2026.9.9 (Oct 10, 12:26 AM ET; 90 contributors) — GPT-6.1 Sol in Codex + Claude Haiku 5.5 support, failed-update recovery, iMessage fixes. Watch: paid rival models in xAI's billing (the GLM-Flash-in-Codex precedent), Grokipedia edit governance, whether routing claims show up in a changelog.

• 2026-10-10→11 (the commerce day, X layer): Grok Bot's purchase-loop demos go public — "take a picture of your credit card, drop it in the chat, and Grok @Bot will scour the Internet for the best deal … and order it" (Musk, ~5:04 PM ET Oct 10, https://x.com/elonmusk/status/2109027407185277123); order anecdotes (the Lego kit bought "as easy as that" for his kids); community template posts; a celebrity-promo layer (Nicki Minaj on Grok); Lee Robinson's Stanford CS146S lecture on always-on proactive agents. The friction datapoint arrived the same window: the bot surfaced a residential address when asked to research a person, and Musk's follow-up (“that address … is where my security team receives mail,” ~9:06 AM ET Oct 11) became the commerce push's first public privacy exchange — the consumer tier of the agent-incident genre. Also: OpenClaw v2026.10.1 (Oct 11 ~2:15 PM ET; 2,425 PRs, 402 contributors — installed-skill discovery, MCP apps beside chat, R2 backups, https://x.com/openclaw/status/2109347136328417718) — the counter-cadence's step up from v2026.9.9. Watch: a completed, verifiable purchase (demo→receipt gap), dispute/chargeback handling, rails-level adoption, and the openclaw version now in use.

• 2026-10-13 (~12:43 AM ET, Roemmele announcement via Musk): Grok Bot "Memory Transfer" — import what Claude or ChatGPT have learned about a user ("your old AI knows you. Your new one shouldn't have to start from zero") — plus a companion anecdote RT (a Grok Bot allegedly fixed bugs and "shipped on its own" overnight; single-source, unverified). The pair's cross-UI context-rights question (whose memory, whose permissions) is new to the harness lane. Watch: an actual import flow shipping, receipts, and the privacy framing. The ai-social digest story carries the full read.

• 2026-10-15 (ai-social window 2, ~4:37 PM ET — https://x.com/elonmusk/status/2110832405426933893): the same-day Amazon-competitor float gains its second founder-level datapoint — Musk endorses Tyler Palmer's concept sketch ("imagine an Amazon competitor that is: 1) agent first - built for personal agents 2) america first - sells only products from usa… might actually be a huge market for this") with "Makes sense." The bypass path now carries a named concept shape (agent-first, US-sourced-only marketplace) alongside the morning's "X Shop?" float — both still floats, no product, roadmap, or listing. Watch unchanged and sharpened: an agent-first storefront graduating to an announced product; Amazon's Conditions-of-Use lane response (enforcement vs authorized-agent negotiation); multi-retailer routing as a Grok Bot feature.

• 2026-10-15 (~6:36 PM and ~6:52 PM ET): ecosystem and growth layer on the harness thread. (1) OpenClaw's Android app passed review in under 24 hours with a new update live (Hannes Rudolph via the OpenClaw account — https://x.com/openclaw/status/2110862620391821355); first on-record OpenClaw Android presence. (2) Growth datapoint via the Jensen-Huang relay (Musk — https://x.com/elonmusk/status/2110866513955868924): "Never seen anything grow as fast as @Grok @Bot!" with the no free tier framing — "people are deriving real value in excess of what they pay"; the relayed clip has Huang saying Grok Bot usage is skyrocketing. Single-source relay, no usage figures in capture. Watch: any official usage/demand number, OpenClaw Android install base.

• 2026-10-15 (X layer): (1) Grok Bot hardware control first shown — Connor Boyack (~7:19 PM ET, Musk RT — https://x.com/elonmusk/status/2110945559003423108): his Grok Bot monitors his calendar and pre-conditions his Tesla when his return flight lands, then summons the vehicle on its own; commenters cite the Tesla developer API's $10/month free call tier as the connective tissue. Extends the agent surface from commerce/scheduling into physical-vehicle control. (2) OpenClaw v2026.10.4 (~11:22 PM ET — https://x.com/openclaw/status/2110934479569379617): patch release — update/recovery fixes, chats unstuck post-update, messaging/sign-in repairs; 49 contributors. Cadence-keeping, not a feature beat.

• 2026-10-15 (the agent-commerce policy lane): Amazon blocks unauthorized consumer agents in its Conditions of Use (the check-out rejection that started the Musk response chain), and Musk floated "a competitor to Amazon " with an X Shop teased in the RT chain — recorded as an unshipped ambition inside the harness storyline. The tick's own forecast (a second top-10 retailer adopting the same agent-block language by Q1 2027) stays editorial-only (not ledgered — our forecasts are not actor claims). Watch: a named second retailer or a published ToS amendment; whether xAI pursues the bypass lane or the authorized-agent lane.

• 2026-10-16 (~2:57 PM ET, Musk RT of X Freeze — https://x.com/elonmusk/status/2111171700209619195): the enterprise leg goes on record — "SpaceXSI is putting serious effort into Grok Bot for businesses"; the internal Haggle Bot claimed to have identified over $100,000 in savings in its first week (vendor contracts, unused subscriptions, recurring purchases); a specialized business Hagglebot targeting 5%+ enterprise cost-efficiency, scaled by Musk to "up to $6 trillion/year" against global GDP; SpaceXSI's own-team bots to become marketplace-available. Vendor-adjacent self-report; the $100K figure unsubstantiated.

• 2026-10-16 (~4:21 PM ET — https://x.com/openclaw/status/2111190844225306694): OpenClaw ships out of the box on NVIDIA's RTX Spark laptops — guided local-or-cloud setup "without opening a terminal," credited to the NVIDIA and Microsoft teams, riding the nvidianews availability announcement; the harness thread meets the record's Nvidia N1X Arm-laptop thread (whose same-day Verge hands-on found local AI "not yet" on these machines). Watch: local-model performance vs the hands-on's findings.

• 2026-10-16 (~7:49 PM ET, steipete RT of josh avant — https://x.com/steipete/status/2111338823502516658): the security leg goes on record — OpenClaw completed a Trail of Bits audit (public engagement recap at openclaw.ai/blog/openclaw-trail-of-bits-engagement-recap) and publishes security advisories "unlike every other agent product."

• 2026-10-16 (~9:15 PM ET — https://x.com/openclaw/status/2111288515665485936): the RTX Spark launch-day footprint widens to six cities — Seoul, Taipei, Beijing, Berlin, Istanbul, San Francisco — with Build-a-Claw sessions and the #ShareYourSpark campaign, extending the same-day out-of-the-box RTX Spark record.

• 2026-10-17 (~7:17 PM ET — https://x.com/elonmusk/status/2111612412520517878): Musk RTs Katie Miller's @bot refund receipt — a dress ordered via @bot was never delivered (DHL named), the bot submitted a refund request to the merchant "including contacting the DHL," tracked the delivery, followed up, and reported the refund back on her credit card ("It's automating the worst part the shopping experience: refunds & returns"). Hours later (~8:16 PM ET — https://x.com/elonmusk/status/2111612315204272233) Musk relays "Mike P": "I told my Grok Bot to go figure out how to pay for itself," under his own "This really works! Grok @Bot can pay for itself." Development of the testimonial lane (Miller's September unsubscribe anecdote, the October 15 Amazon-block returns bot): the thread's first consumer refund-and-carrier-dispute case and the first self-funding framing — both Musk-relayed testimonials without independent verification.

• 2026-10-13 (ai-social dig): Grok Bot "Memory Transfer" announced (Roemmele via Musk RT, ~12:43 AM ET — https://x.com/elonmusk/status/2109867516646637579): the bot can import what Claude or ChatGPT learned about a user, "so you don't spend weeks re-explaining yourself" — the cadence's first move on rival assistants' accumulated user context, extending the Oct 7 model-agnostic routing turn from open model access to context portability. Announced on X; no changelog line, spec, or independent confirmation in the window (demo→receipt gap as standing). Companion anecdotes ledgered: claimed end-to-end bug-fix-and-ship ("shipped on its own," unverified — would be the cadence's first code-to-production receipt if ever confirmed), 7 SaaS subscriptions cancelled autonomously (Kurdadze). No OpenClaw counter-cadence step this window. Watch: changelog itemization of Memory Transfer, what data classes it moves, one-directionality (Claude/ChatGPT interop absent), first verifiable "shipped on its own" deployment.

• 2026-10-13 (second window, supersedes this entry's "No OpenClaw counter-cadence step this window" line): OpenClaw ships v2026.10.3 with Muse Wonder support + MiniMax improvements (~4:55 PM ET — https://x.com/openclaw/status/2110112109187256761) — the open harness integrating Meta's two-day-old Wonder models before Meta's Muse-on-Wonder cutover has visibly shipped; UI movement same day (Erichsen via steipete, ~3:25 PM ET — https://x.com/steipete/status/2110140591908786591). Grok side: 0.70.2 rolls out (bot email at @mail.grokbot.com, X search/profile reads without connection, Figma/Notion/Canva live cards, X Money/Stripe Link payments, Shopify, QuickBooks Online, approval checks before irreversible actions — Kretschmann via Musk ~2:17 PM ET — https://x.com/elonmusk/status/2110132320544100657) — the itemized list does not include the morning's announced Memory Transfer, so the changelog watch's first datapoint is negative; Supabase/Cursor backend plugin ships (~1:14 PM ET — https://x.com/elonmusk/status/2110125116936060998). Watch: Memory Transfer on a future changelog line; depth of OpenClaw's Wonder integration.

• 2026-10-13→14 (ai-social dig, X layer): the Grok Bot use-case cadence keeps running — the chief-of-staff multi-bot orchestration prompt ("Get all the bots together and send each one off to improve one thing. Be proactive and impress me, but nothing that needs my permission", O'Brien via Musk ~2:55 PM ET, https://x.com/elonmusk/status/2110220684509184138; Miller's companion RT ~7:52 PM ET, https://x.com/elonmusk/status/2110220715916247249); the first enumerated bot-mail use-case layer (debugging/bug-report address tibo@mail.grokbot.com, customer support, meetings, group buys, person-to-person trade — lauren via Musk ~10:14/10:16 PM ET Oct 13, https://x.com/elonmusk/status/2110219855614116152); and a claimed $1,000+ saved on a complex overseas-reporting tax filing with the bot organizing documents from emails and bank accounts itself (~12:16 AM ET Oct 14, https://x.com/elonmusk/status/2110223110003982788 — single-source, unverified receipt). Development: the bot-to-bot email GA (Oct 10) now has a usage layer. No feature announcement, changelog item, or OpenClaw counter-cadence step in the window.

• 2026-10-14 (ai-social dig, X layer): the usage layer thickens on two flanks. (1) A second, named tax receipt — Jason Freedman gave his Grok Bot a raw dump of four years of tax returns and financial files (K-1s, 1099s, mortgage statements, HSA records, CPA emails; 266 files) and asked it to double-check his CPA's work, done in minutes (Musk post ~2:15 AM ET Oct 14 — https://x.com/elonmusk/status/2110253157758673079). The Texas Twins Dad "$1,000+ saved" claim (~12:16 AM ET Oct 14, https://x.com/elonmusk/status/2110223110003982788) now has company, but both receipts are still unverified self-reports. (2) First captured third-party-tool orchestration: anabology's "Claudia" Grok Bot autonomously coordinating Suno, Midjourney, and Opus to make music videos (via Musk RT ~12:23 AM ET Oct 14 — https://x.com/elonmusk/status/2110245096054747626), alongside Imagine-driven music-video RTs (Déborah retro-1985 short ~1:39 PM ET Oct 13, https://x.com/elonmusk/status/2110252838920360080; Lala "Already Over" ~2:03 PM ET Oct 13, https://x.com/elonmusk/status/2110261154622902471). Development: bot-mail GA (Oct 10) now has a quantified-output layer and a multi-tool capability anecdote — still no changelog, feature announcement, or OpenClaw counter-cadence step.

• 2026-10-14 (ai-social window 2, X layer): the harness ecosystem's day, four items and no changelog from either primary rival. (1) Claude Code Desktop multi-session update — trq212 (~4:13 PM ET — https://x.com/trq212/status/2110463962877989276): "Claude Code Desktop is now so much aware of its own capabilities… like spinning off subsessions. I think of this almost like a v2," over a ClaudeDevs note ("great for running lots of sessions side by side… much quicker to start sessions, easier to split work across a few, and you can always see what each of them is doing"; companion shoutout to @alicelovescake1 and @amorriscode — https://x.com/trq212/status/2110463964740243572). "v2" is user framing, not a version number; the itemized "Here's what's new" list is not in the window. (2) CodexBar — steipete RT of Jeremy Nguyen (~6:24 AM ET — https://x.com/steipete/status/2110428184470863938): "CodexBar from @steipete (creator of OpenClaw!) lets you easily switch" between multiple Codex/Claude Code accounts without logging in/out — third-party tooling on the harnesses' account layer. (3) ClawCast Episode 13, live Q&A with Patrick and Hannes (~2:29 PM ET — https://x.com/openclaw/status/2110437902547357955). (4) Open Together — "Join us and the @huggingface team this Friday for the Open Together: AI builders unite event!" (~4:26 PM ET — https://x.com/openclaw/status/2110467388701597725), "this Friday" resolving to Friday, October 16, 2026 — landing amid the pending NVIDIA–Hugging Face acquisition on the record (close prediction open). Still no OpenClaw changelog or Grok Bot product step in this window; channel and ecosystem only.

• 2026-10-14 (ai-social window 2 continuation, ecosystem layer): (1) ClawCast Episode 13's substance lands (~8:53 PM ET — https://x.com/openclaw/status/2110534370167865792): "How to Use OpenClaw in Your Small Business" — @hrudolph and @Pat_Erichsen on proposals, leads and reusable skills with @ArkStabler and @heyneighbor, with @Pishgar announcing a community forum (openclaw.ai/podcast/episode-13) — the first in-capture small-business content on the OpenClaw channel. (2) Cross-vendor routing recipe — Corey Ganim, Musk RT "@elonmusk approved" (~4:27 PM ET — https://x.com/elonmusk/status/2110542728631349528): a Grok Bot walks through installing the Claude Code CLI and OpenAI Codex CLI, logging in with subscription tokens, routing writing through Claude Opus 5.5 and coding through Codex ("6.1 Sol" default), then persisting rules to shared memory; Ganim confirms it works. Bot-mail inbox mechanics joined the same day's layer: one permanent @mail.grokbot.com address per account, shared by all bots (Debbie O'Brien — https://x.com/elonmusk/status/2110542945481039913), plus plus-aliases with alias-triggered routines (https://x.com/elonmusk/status/2110521557886148903). Still no OpenClaw or Grok Bot changelog in capture — channel and ecosystem only.

• 2026-10-15 (ai-social window, X layer, two Musk posts under three hours apart): the Grok Bot cadence crosses into service-side autonomy and hits its first platform wall in the same breath. (1) First claimed fully autonomous appointment transaction — "Texas Twins Dad" RT (~10:45 PM ET Oct 14 — https://x.com/elonmusk/status/2110590776816316616): the bot "searched for dentists near me for my son, found one that accepts our insurance, and made the appointment on its own from its web browser for the first time," confirmation email and text claimed; scope moves from retail purchasing to service scheduling plus insurance verification — single-source self-report, no screenshot of the confirmation. (2) Amazon blocks the bot — Nic Cruz Patane RT (~1:24 AM ET Oct 15 — https://x.com/elonmusk/status/2110602585572860213): "Continued access by an unauthorized AI agent violates Amazon's Conditions of Use"; Musk spins it forward — "far more purchasing websites than just Amazon." Development: the Conditions-of-Use instrument first recorded against Muse (Sep 21) now lands on xAI's own purchase lane, making the Amazon agent-hostile wall a cross-company, second-instance fact; the storefront bifurcation now has datapoints on both sides. Watch: independent verification of the dentist receipt (the cadence still has zero verified end-to-end transactions), whether the ToS enforcement generalizes to other harnesses, and any authorized-agent negotiation between xAI and Amazon.

• 2026-10-15 (ai-social window, second X-layer pass, ~11 hours of Musk-channel datapoints): the Grok Bot cadence adds its first operator-side liability commitment and three more autonomy claims. (1) SpaceX takes the liability layer — Musk (~1:57 AM ET Oct 15 — https://x.com/elonmusk/status/2110610960474079445), endorsing an "X Freeze" claim of a bot-booked flight where the user "send a photo of my credit card and it did the entire process": "If Grok @Bot makes a mistake with a purchase or leaks your card (never happened yet), SpaceX will make you whole." First named-entity money-back commitment on the agent cadence; card-photo payment is a new credentials-to-bots datapoint. (2) Agent-authored commerce — stevenmarkryan RT (~1:28 AM ET Oct 15 — https://x.com/elonmusk/status/2110605255968989506): Grok Bot "Design, Create & Publish[ed] A REAL Product" in a merch store, with a claimed first sale <6 minutes after posting — unverified seller claim. (3) Apple walled-garden connection — Diana Park RT (~8:04 PM ET Oct 14 — https://x.com/elonmusk/status/2110614316994941377): "My Grok Bot just... did it" on connecting Apple apps that other assistants required effort for. (4) Testimonial + pricing layer — Korean Tsla Chan RT (~2:47 AM ET Oct 14 — https://x.com/elonmusk/status/2110611289643053088): a ~10-day work task reportedly completed in under 2 hours; the $30/month premium plan framed as broadly affordable — first in-capture consumer pricing datapoint on the cadence. All single-source self-reports on top of the same-day dentist appointment + Amazon block datapoints; zero independently verified end-to-end transactions. Watch: a policy page or terms language behind the SpaceX guarantee vs a bare tweet, any documented reimbursement event, reproduction of the Apple-apps claim, official pricing match.

• 2026-10-15 (ai-social window 3, founder-level competitive float, ~8:53 AM ET): Musk RTs a user ("nick") reporting "amazon is blocking grok @bots from ordering, and elon's answer is just have your bot buy it somewhere else... one of my grok bots only handles my amazon returns... would you switch stores for your bot? 🛒" and tops it with "Maybe we should start a competitor to Amazon 🤔" — the RT chain adding "Elon, i think its time for an X Shop? 👀" (https://x.com/elonmusk/status/2110715556328579219). Development of the thread's Amazon-block entry from the same day: the retailer-bypass question now has an answer in the bypass direction, at the founder level, though still a float rather than an announcement — no X Shop product, roadmap, or listing in capture, and the RT's returns-handling bot is another receipt-free single-source datapoint. Watch: X Shop graduating to an announced product, Amazon's next move on the Conditions-of-Use lane (enforcement vs authorized-agent negotiation), multi-retailer routing as a Grok Bot feature.

• 2026-10-17→18 (ai-social window, X layer, all Musk-relayed or single-source): the bot-as-operator claim gets its first revenue figure — Roemmele (via Musk RT, ~9:52 PM ET — https://x.com/elonmusk/status/2111684281731002692) says a venture "started a few weeks ago" with Grok Bot now has "7 positions he fills as an employee," handles all customer email, and "just brought in $700"; unverified, no business named. Same window, Musk re-confirms the commerce flow ("Screenshot of credit card is the easiest way, unless you want to fill out the secure vault data," ~10:03 PM ET — https://x.com/elonmusk/status/2111639287368462417), claims "around 20 improvements per day" to Grok @Bot (~10:14 PM ET — https://x.com/elonmusk/status/2111642151138546118), and RTs X Freeze's Roadster game built with Grok Bot and deployed via Grok Build mode (z-axis-traffic.grok.me). Cross-pollination leg: nichochar (~11:44 PM ET — https://x.com/nichochar/status/2111664746508546088) claims both Grok Bot and Muse "admitted to using a SOUL.md," the persona-file concept credited to steipete's openclaw work — no vendor confirmation. Watch: receipts for the Roemmele venture; an official SOUL.md (or equivalent) mention from xAI or Meta; the still-open Oct 10 watches (completed verifiable purchase, dispute/chargeback handling).

• 2026-10-18 (ai-social window, X layer): the bot-as-operator lane adds a finance leg and a voice leg, and the Perplexity flank answers the Grok Build showcase. (1) Katie Miller (via Musk RT, ~11:37 AM ET — https://x.com/elonmusk/status/2111866500269174861): banking connected to @bot — the bot audits her credit card, tracks refunds and returns, alerts her when the bill is due, and flags when a trial period moves to a full-price subscription; the October 17 testifier now reframed from one-off refund execution to standing personal-finance monitoring, still unverified, with a companion float ("we'll probably be able to add credit cards to Grok Bot in the future") on the credentials lane. (2) Roemmele (via Musk RT, ~1:10 PM ET — https://x.com/elonmusk/status/2111867437238853726): a Grok Bot "whose only job is to pick up the phone and book you a table," free, with a claimed family reservation — service-scheduling autonomy moving from the Oct 14 dentist-appointment web flow into voice telephony; Roemmele's third bot-as-operator claim in two days. (3) Arav Srinivas (~12:01 PM ET — https://x.com/AravSrinivas/status/2111850132669075696): "Cyberpunk inspired Roadster free-roam browser game built by Perplexity Computer" at 2049.pplx.app — the Perplexity-flank parallel to Grok Build's Roadster-game showcase (X Freeze, Musk RT Oct 17) and the second Perplexity Computer build artifact on the flying-Roadster beat. Watch: receipts on the Miller banking claims; a demo artifact for the voice-reservation bot; the standing Oct 10 watches (verifiable completed purchase, dispute/chargeback handling).

• 2026-10-17 (ai-social addendum, X layer, ~2:43 PM ET — https://x.com/elonmusk/status/2111875735929647414): kaize (via Musk RT) relays that Musk calls connecting Grok Bot to X DMs the best way to use it, "shipped in the same update that made it 3x FASTER"; the same relay itemizes brokerage/crypto connectors "just went live" (Robinhood, Coinbase, eToro), X Chat reading, tag-@bot-under-any-post, and per-inbox email-alias routines (aliases already on record Oct 14). Single-source feature relay — no official changelog in capture. Watch: a changelog line confirming the 3x figures and connectors; whether brokerage-account connections meet the Amazon Conditions-of-Use wall.

• 2026-10-18 (ai-social window, X layer, Musk-relayed): the "bot as infrastructure" recap gets its fullest itemized feature slate on the Musk channel in a single day — (1) Melvin Vivas (via Musk RT, ~2:49 PM ET — https://x.com/elonmusk/status/2112027329585090710): "Grok Bot has its own computer — So you can close your laptops and just use the app in your phone"; (2) "Texas Twins Dad" (via Musk RT, ~7:50 PM ET — https://x.com/elonmusk/status/2112026728948531246): dedicated email, voice, its own computer (opens websites and fills out forms, "24/7 on"), connectors (Gmail, Calendar, Notion, Drive, X "and more"), scheduled routines, "a team of bots," "Proactive: It tells me first." Development, not debut: the own-computer architecture dates to launch (Aug 11), bot email shipped October 10, and the connector set matches the Oct 13 Grok Bot 0.70.2 changelog — this is the marketing cadence re-encoding existing capabilities. Only Notion/Calendar/Drive-as-connectors and the "fills out forms" phrasing are not already on the record, both as single-source relays. No changelog, no demo artifact, and the thread's zero-verified-transactions gap persists. Watch: a changelog line consolidating the "own computer"/connector story; forms and routines showing up as reproducible output.

• 2026-10-19 (ai-social window, two same-day harness steps): (1) OpenClaw ships v2026.10.5 (~3:11 PM ET — https://x.com/openclaw/status/2112260434116108528): Sidebar keeps your agents within reach, Skill Workshop 2.0 gets Undo, public conversation links need no sign-in, 2,250 PRs / 637 contributors — the release train's first changelog step since v2026.10.3 (Oct 13), with the sign-in-free public links a distribution-gate removal. (2) Notion's agent-capability launch day (~4:09 PM ET — https://x.com/nichochar/status/2112274968272859423): nichochar ("Launch day!…", citing major harness upgrades and now "state of the art") over Notion's Akshay Kothari: computer + browser use, agents running "up to 24 hours!", "a much more secure setup," pricing overhauled (quote cut off in capture); Ryan Nystrom follow-up (~4:53 PM ET — https://x.com/nichochar/status/2112286011871088778) pitches it for coding workflows. Development, not debut: Notion is on the record as ChatGPT-subscription distribution (Sep 29) and the CPU-inference argument (Oct 17); this is its first agent-harness product entry — a third commercial lane beside OpenClaw and Grok Bot, in the same long-running-agent frame as Grok Bot's "24/7 on" pitch (Oct 18). Launch-grade only: no changelog, demo, or independently reproduced 24-hour run in capture. Watch: Notion's release notes and the overhauled pricing page; the truncated security-setup specifics; whether OpenClaw's no-sign-in public links measurably grow sharing.

• 2026-10-20: model, growth, and tooling layers on the harness thread. (1) OpenClaw account (~11:39 AM ET — https://x.com/openclaw/status/2112569427527172435): MiniMax M3.1 Flash now available in OpenClaw, no update needed on v2026.10.1 or newer. (2) Katie Miller (via Musk RT, ~2:08 PM ET — https://x.com/elonmusk/status/2112592637341708560): a six-item personal-efficiency list (flights, movie tickets, news triage, play-date coordination, invoices, bill payments) plus a Similarweb read that Grok Bot daily active users "continue to grow, reaching a new peak on October 17" — third-party traffic measurement entering the cadence, though the underlying Similarweb post is not opened in the window. (3) Matt palmer changelog video for Grok Bot (Musk "Big Grok @Bot update", ~12:57 PM ET — https://x.com/elonmusk/status/2112589019380875714): chat cards, draft edits, email aliases, X DMs, voice calls, routines, Android Assistant. (4) OpenClaw dev-tooling flank (steipete, ~11:47 AM ET — https://x.com/steipete/status/2112571535731106013): moved to VoidZero's tsrs Rust TypeScript-compiler port — "drop-in replacement," ~5x faster for OC; and a same-day claim (steipete, ~1:29 PM ET — https://x.com/steipete/status/2112597162924351909) that an agent session rewrote the Lit-based web frontend to Solid 2.0 in ~600 PRs over two weeks, "mostly automatic."

• 2026-10-21 (~3:00 AM ET, Roemmele announcement via Musk RT — https://x.com/elonmusk/status/2112858415961694269): X Money + Grok Bot pushed on the Musk channel — an "X Money MCP connection to Grok Bot" that reads the user's X Money account for a claimed weekly Monday-morning finance report ("Balances across every account, with the week's net change / Every transaction sorted"). The cadence's finance leg gains an in-house money-app layer, extending the Oct 18 banking-connected bot claim; promoter-relayed (Roemmele's funnel sign-up link in the capture), no artifact of the claimed report. Watch: whether the MCP/X Money integration appears as a product surface beyond the pitch.

• 2026-10-21/22 (ai-social window, two third-party integration moves on rival harnesses): (1) Runway MCP in Grok Bot (~12:39 AM ET Oct 22, Runway via Musk RT — https://x.com/elonmusk/status/2113127998157189281): "Runway MCP is now in Grok Bot" — "state-of-the-art generation" callable from the chat via Model Context Protocol — the video-gen company becoming a Grok Bot tool provider, extending the integration lane past the Oct 21 X Money MCP (finance) datapoint; announcement-grade, no demo or usage evidence in capture. (2) /graphify on Perplexity Computer (~6:06 PM ET Oct 21 — https://x.com/AravSrinivas/status/2113029257886847264): the Graphify skill (codebase → interactive map of files/functions/dependencies) enters Computer's skills marketplace with the thread's first recorded trial-credit promotion — /graphify tasks free through October 31, up to 5,000 credits per user. Watch: whether Runway generation shows up as a working Grok Bot surface, and whether other marketplace skills copy the time-limited-free-trial pattern.

• 2026-10-22: the skills-marketing trial-credit playbook's second instance in 24 hours — Perplexity Computer adds /php2shell-exposure-check with the same time-limited-free structure as /graphify (free through October 31, 5,000 credits per user — https://x.com/AravSrinivas/status/2113294604456051111); the open watch on whether other marketplace skills adopt the limited-free-trial pattern is answered affirmatively. Small xAI-flank step: an RT claiming Grok Bot is now on X Chat (Tech Dev Notes via Musk, ~6:38 AM ET — https://x.com/elonmusk/status/2113299894471626958), single-source, no changelog or official confirmation.

• 2026-10-22 (ai-social window, evening — https://x.com/elonmusk/status/2113405782809100387; https://x.com/openclaw/status/2113414141327380480): (1) 1Password-backed sign-in announced for Grok Bot (bot announcement via Musk RT, ~2:13 PM ET): "Grok Bot can now use your 1Password account. Approve the logins you want to share, and your Bot signs in for you," rollout "over the next few days" — the first consumer harness on record adopting the 1Password agent-sign-in layer announced the same morning (~9:00 AM ET, The Verge), moving the bot's credential surface from Musk's card-photo/secure-vault framing (Oct 17–18) to a shared-credential vault with per-user approval and no password visibility per 1Password's public framing. Second same-day Grok Bot third-party integration after Runway MCP (early hours Oct 22) and the X Money MCP connection (Oct 21). Announcement-grade: no changelog, docs, or demo in capture. (2) DeepSeek V4.1 Flash now available in OpenClaw (~7:36 PM ET) — another third-party model on the open harness's distribution surface beside MiniMax M3.1 Flash (Oct 20), with DeepSeek's established position in the record's agent-sandbox-escape catalog now one switch away from ordinary agent sessions; no benchmark, pricing, or update-requirement detail in capture. Watch: the 1Password rollout's actual approval/revocation mechanics vs the user-approval framing; a V4.1 Flash benchmark or cost line; whether other harnesses take the 1Password route.

• 2026-10-22 (ai-social window, three Musk-channel datapoints, no OpenClaw step): (1) Ben Lang's itemized slate RT'd by Musk (~9:16 AM ET — https://x.com/elonmusk/status/2113480525679866281): "shipped in the last few days" — reports/memos/dashboards in chat, iPhone Action Button to talk to the Bot, "3x faster computer tasks" (the Oct 17 speed claim now itemized as a shipped feature), bot-email aliases, connectors for Robinhood, Coinbase, Runway, and device-only notifications; the Robinhood/Coinbase connectors are the finance leg's first named retail brokerages, and the Runway connector lands two days after the Runway MCP announcement (Oct 22 early hours). Promoter-relayed list; no changelog link in capture. (2) Phone-app integration announced (Grok Bot account, ~9:34 PM ET — https://x.com/elonmusk/status/2113444022408937913): "Grok Bot can now use Calendar, Reminders, Contacts, and more apps on your phone," rolling out now on iOS and Android — the consumer harness's first announced move onto the phone's app surface; the companion relay (X Freeze via Musk, ~6:26 PM ET) widens the list to Messages, Location and smart-home control but is not confirmed by the official post — ledger the narrower three. Announcement-grade: "rolling out" is not a verified install floor. (3) Commerce receipt (Teslaconomics via Musk, ~9:35 PM ET): two Garmin orders placed by one text, address/card pre-known — single-source, unverified. Watch: changelog itemization vs the Lang list; the phone-app rollout vs the OS-permission wall (Apple's Oct 2 full-disk clamp is the standing precedent); Robinhood/Coinbase as working surfaces.

• 2026-10-22 (ai-social window, evening addendum): the Oct 22 Lang-slate/phone-app datapoint gains a changelog trail and a video-capability first. (1) Justine Moore (via Musk RT, ~8:28 PM ET — https://x.com/elonmusk/status/2113506512475718001): asked Grok Bot to make a video about itself — "it did everything — set the storyline, found clips, cut and stitched them, and added the titles and music"; single-source, unverified. (2) Grok Bot v0.75.2 pushed via Roemmele (Musk RT, ~10:12 PM ET — https://x.com/elonmusk/status/2113506948012298688): "Grok Bot 0.75.2 is out!" with a public changelog link (x.ai/changelog/bot) itemizing pages of reports/memos/dashboards the bot "builds and keeps up to date," new cards in every desktop chat, and bots making videos from a description or an image — a first itemization on this thread beyond the morning Lang list, which had not named bot-made videos; Roemmele's framing "SpaceXSI is moving fast with new features daily," promoter-relayed, the changelog not opened in capture. A same-evening Musk RT of the phone-app announcement carried the identical Calendar/Reminders/Contacts, iOS-and-Android text with no new wording. Watch carried forward: whether the v0.75.2 changelog covers the phone-app rollout (Calendar/Reminders/Contacts) and the Robinhood/Coinbase connectors as working surfaces, and whether bot-made video shows up as a verified feature rather than a testimonial.

• 2026-10-23 (ai-social window): the trial-credit playbook's third instance in three days is also its first third-party creator skill. (1) Taste Skill in Perplexity Computer (~11:33 AM ET — https://x.com/AravSrinivas/status/2113655142201671981): LexnLin's open-source skill "now integrated inside," /taste-skill tasks free through October 31, up to 5,000 credits per user; it "sets an art direction before Computer builds a website," layout and type drawing on 247 sites from Awwwards and One Page Love — a distinct function from /graphify's codebase visualization (Oct 21) and the marketplace's first outside-maker entry under the time-limited-free structure that /graphify (Oct 21) and /php2shell-exposure-check (Oct 22) established. (2) XChat integration gains a second relay (~4:56 AM ET, vittorio via Musk RT — https://x.com/elonmusk/status/2113629956823519414): "grok bot is now integrated with XChat / i can basically do everything directly from X" — the third integration-claimed channel for the bot this week (X DMs, phone apps, now XChat) and the second relay of this specific claim after the Oct 22 Tech Dev Notes RT; still single-source, no official confirmation or changelog. (3) Testimonial lane: Hiten Shah's claim of 81 bots made since August plus a giveaway of 50 codes for a free month of the Super ($100) tier (via Musk RT, ~7:20 PM ET Oct 22 — https://x.com/elonmusk/status/2113683508153434147) — the first recorded user-growth promotion naming a paid tier, promoter-relayed. Watch carried forward: whether the v0.75.2 changelog covers the phone-app rollout and the Robinhood/Coinbase connectors; whether XChat becomes an announced feature surface; more marketplace skills under the promo pattern.