





A folkfox special report.
Grok Imagine layoffs: why SpaceXAI cut 60 staff for Grok Bot agents
SpaceX Super Intelligence has cut roughly 60 staff from Grok Imagine, leaving a skeleton maintenance crew while compute and headcount pivot to Cursor coding tools and Grok Bot 1.0 agents inside Zoom, Microsoft Teams and Slack.
By Katie Delaney. Research checked 14 November 2026. Illustrations are conceptual.






Sixty seats vanished from the cinema wing in a single week. On 13 November 2026, reporters Aaron Tilley, Claudia Kim and Juro Osawa broke the news in The Information that Elon Musk’s artificial intelligence division had executed sweeping Grok Imagine layoffs.
Roughly 60 engineers, researchers and product specialists building the consumer video and image tool Grok Imagine were dismissed during the first week of November 2026. Those Grok Imagine layoffs removed more than half of the dedicated media-generation unit inside SpaceXAI, leaving only a small maintenance crew to keep existing endpoints alive.
Follow-up dispatches from AI Weekly and Startup Fortune confirmed that the Grok Imagine layoffs are not a retreat from frontier computing. Instead, SpaceX Super Intelligence LLC (the post-merger entity operating xAI, tracked across industry filings as SpaceXAI and @SpaceXSI) is redirecting GPU clusters and headcount toward Cursor coding tools and Grok workplace agents led by Grok Bot 1.0.
Elon Musk’s SpaceX AI group has laid off roughly 60 employees working on Grok Imagine, leaving just a handful to maintain the product. The cuts signal a shift toward AI coding tools and agents as Chinese companies pull ahead in video generation. [...]
Scoop: SpaceX has axed the team (roughly 60 people) behind AI video generation tool Grok Imagine. It's still live for now, but the company is shifting resources elsewhere [...]
From Aurora and Imagine 0.9 to a quiet freeze after the Grok Imagine layoffs
For creative technologists prowling the generative thicket, the speed of the reversal inside the Grok Imagine layoffs is striking. As documented in the Latent Space Grok Imagine technical archive, the product lineage began in December 2024 with the autoregressive Mixture-of-Experts image model code-named Aurora.
In July 2025, xAI rolled out Grok Imagine 0.9 across X and its standalone mobile apps, bundling text-to-video, image-to-video, native audio synthesis and viral presets including a controversial Spicy Mode. By January 2026, the lab opened public API access and touted 1.2 billion generated videos over a 30-day window, followed by Elon Musk’s March 2026 pledge to double down on Grok Imagine long before the Grok Imagine layoffs took shape.
Even in mid-2026, shipping velocity looked healthy: June 2026 brought Grok Imagine Video 1.5 and 1.5 Fast with native 1080p resolution at 24 frames per second and 30-second clips, while August 2026 delivered Grok Imagine Image 2.0. Yet behind the scenes, high inference burn per clip and rapid user churn left the product vulnerable once the Grok Imagine layoffs began.
Headcount discrepancies in the Grok Imagine layoffs and what remains in Palo Alto
How large was the multimodal burrow before the Grok Imagine layoffs landed? BigGo Finance and 36kr reporting noted that roughly 70 specialists worked across the core multimodal generation group prior to November, whereas The Information counted around 60 cuts in the Grok Imagine layoffs representing more than half of an extended 100-to-120-person unit including product and video data staff.
Would you be sad to see Grok Imagine disappear? [...] The hype was mind-blowing, and I saw so much amazing art made with it. Now The Information reports that about 60 people from the Grok Imagine team were laid off, with only a handful left to keep it running. It's no longer a priority. The focus has shifted to AI coding tools and Grok Bot.





To understand why the Grok Imagine layoffs arrived in November 2026, follow the corporate trail back to 2 February 2026. On that date, SpaceX completed an all-stock acquisition of xAI at a $250 billion valuation (exchanging each xAI share for 0.1433 SpaceX shares), forging a $1.25 trillion combined group that surged toward $1.9 trillion after SpaceX’s June 2026 IPO.
Operating inside SpaceX Super Intelligence LLC (SpaceXAI), leadership reorganised the entire engineering warren into four explicit divisions prior to the Grok Imagine layoffs: Grok Main Chatbot and Voice, Coding Tools and ML Infrastructure, Grok Imagine, and the multi-agent enterprise automation programme known as Macrohard.
That post-merger reorganisation triggered a wave of senior departures that foreshadowed the Grok Imagine layoffs. More than 50 researchers and half of xAI’s 12 original co-founders exited during 2026, including co-founders Tony Wu and Jimmy Ba in February alongside core Grok Imagine architects Haotian Liu, Guodong Zhang and Ethan He, several of whom left to launch independent video ventures.
The Grok Imagine layoffs also mirror two earlier pruning exercises across Elon Musk’s companies: xAI’s September 2025 dismissal of roughly 500 generalist data annotators in favour of specialist STEM tutors, and Tesla’s decision to wind down its custom Dojo supercomputer silicon once commodity GPU clusters proved faster to deploy.
Elon Musk once promised to double down on Grok Imagine. Now SpaceX has gutted the group behind it, echoing OpenAI’s retreat from costly AI video generation as ByteDance and other Chinese companies gain ground. [...]
Argues that AI video generation burns massive compute per clip with high consumer churn, whereas coding assistants and workplace agents deliver recurring monthly enterprise revenue, leaving Grok Imagine on maintenance life support after the Grok Imagine layoffs.
Why US frontier labs are retreating from standalone AI video in the wake of the Grok Imagine layoffs
The unit economics of standalone consumer video generation explain the Grok Imagine layoffs immediately. Rendering a single ten-second 1080p diffusion clip consumes orders of magnitude more FLOPs and HBM bandwidth than serving hundreds of lines of code completion or summarising a 45-minute meeting transcript, while casual consumers frequently churn after a month of novelty prompts.
SpaceXAI is not the only American lab to pull back as the Grok Imagine layoffs unfold. OpenAI shut down its standalone Sora consumer application between March and September 2026 after internal accounting showed the service burning more than $1 million per day in inference compute against thin recurring revenue after peaking near one million downloads.
Meanwhile, Chinese labs backed by integrated short-video social ecosystems have pulled ahead on both release cadence and unit economics, intensifying the commercial pressure behind the Grok Imagine layoffs. ByteDance’s Seedance 2.5, MiniMax’s H3 and Kuaishou’s Kling ship monthly upgrades wired directly into Douyin, TikTok and e-commerce ad pipelines, alongside Google DeepMind’s Veo 3 inside YouTube Shorts.
Opportunity cost sealed the fate of the division during the Grok Imagine layoffs. Throughout late 2026, SpaceX has leased excess NVIDIA GB300 cluster capacity to external frontier buyers, including a reported $1.25 billion-per-month compute lease with Anthropic and parallel capacity contracts with Google. Every megawatt burned rendering loss-leader Grok Imagine clips is a megawatt that cannot earn hard lease revenue or power enterprise Grok workplace agents.
folkfox compute & headcount ledger
How the Grok Imagine layoffs free silicon for Grok Bot and C/C++ rollouts
| SpaceXAI workstream / metric | Share / rate | Scale benchmark | November 2026 status |
|---|---|---|---|
| Grok Imagine video & image unit | ~55.0% cut | ~60 staff dismissed in Grok Imagine layoffs (~70 in multimodal per 36kr) | Skeleton maintenance crew; feature freeze |
| Grok Bot 1.0 daily active engagement | 75.0% DAU/WAU | >600,000 DAU out of 800,000 paid WAU | Priority expansion into Zoom, Teams & Slack |
| Grok 4.8 C and C++ RL & serving stack | 62.5% MFU | +21.5 pts over SGLang; +38% rollout speed | Week-1 soft rollout to Grok Bot users |
| Legacy JAX training stack | 51.2% MFU | Prior xAI pre-merger training baseline | Replaced by bare-metal Starlink C/C++ |
| Legacy SGLang serving stack | 41.0% MFU | Prior Python agent serving harness | Deprecated for multi-turn agent loops |
| Video platform & lab | Flagship model | Nov 2026 operational status | Commercial distribution moat |
|---|---|---|---|
| SpaceXAI Grok Imagine | Video 1.5 / 1.5 Fast & Image 2.0 | Maintenance crew only (~60 staff cut) | Bundled inside The X Subscription & legacy xAI API |
| OpenAI Sora | Sora standalone app | Shut down Sept 2026 ($1M+/day burn) | Folded remnants into core multimodal research |
| ByteDance Seedance | Seedance 2.5 | Active monthly feature velocity | Native Douyin, TikTok & CapCut creator monetisation |
| MiniMax | MiniMax H3 Video | Active rapid iteration | Low inference cost per second & high motion coherence |
| Kuaishou Kling | Kling 2.x studio suite | Active commercial expansion | Prosumer camera controls & e-commerce ad generation |
| Google DeepMind Veo | Veo 3 | Active inside YouTube & Vertex AI | YouTube Shorts creator pipeline & Google Cloud enterprise |





Where are the compute budgets and engineering seats freed by the Grok Imagine layoffs actually going? Follow the scent directly to x.ai/bot and the general availability rollout of Grok Bot 1.0. First previewed in beta on 11 August 2026, expanded to shared Team Bots on 28 September and promoted to 1.0 in early November 2026, Grok Bot is SpaceXAI’s flagship workplace platform.
Introducing Grok Bot 1.0. AI teammates that work alongside your whole team. Share Bots, collaborate with them in Slack and Teams, invite them to meetings, and watch them learn and become more helpful.
Grok Bot can now join meetings. Ask it questions, assign it work, or have it keep track of what your team decides.
How Grok Bot works under the bonnet after the Grok Imagine layoffs: persistent Firecracker microVMs
According to SpaceXAI’s technical manual on how Grok Bot works, every Bot funded by the Grok Imagine layoffs resource shift runs inside its own isolated Linux virtual machine hosted on US Firecracker microVMs (*.cursorvm.com). Unlike stateless chat tabs that vanish when closed, Grok workplace agents maintain a persistent file system, a full headless browser, a bash shell, scheduled cron jobs and markdown memory across weeks.
Teams can equip a Bot with its own email address, GitHub identity, Linear seat and SaaS logins via an encrypted Vault, a 1Password integration or a one-time Secure Form that injects credentials without exposing raw passwords in chat logs. Inside SpaceXAI’s own engineering burrow following the Grok Imagine layoffs, internal instances such as tinkabot and dr eggbot triage build alerts, open pull requests and manage ad budgets around the clock.
How a Grok meeting agent behaves inside Zoom, Microsoft Teams and Google Meet
Two days before the Grok Imagine layoffs broke in the press, SpaceXAI launched the capability documented in Grok Bot Meetings: sending a live Grok meeting agent directly into Zoom, Microsoft Teams and Google Meet video calls. Unlocked on self-serve Team plans ($40/seat), Enterprise contracts and individual $200-per-month Ultra subscriptions after the Grok Imagine layoffs, the meeting workflow follows four concrete rules:
- Calendar or ad-hoc join: Invite a Grok meeting agent by forwarding a calendar invite, pasting a Zoom, Microsoft Teams or Google Meet URL into chat, or clicking Join Meeting in the dashboard.
- Visible lobby badge: The bot enters the call roster with the display suffix
[Bot Name] - transcribing(for example,finance-bot - transcribing) so attendees see an automated scribe in the room. - Live in-call Q&A and private Zoom replies: Participants can message the Grok meeting agent in the call chat to pull live CRM figures or assign tickets; on Zoom, private direct messages receive private replies, and hosts can toggle
Disable Interactivityfor silent note-taking. - Post-call Pages and Slack follow-up: Once the call ends, the Grok meeting agent publishes structured minutes, decisions and owner-tagged action items into collaborative Pages and posts follow-up tasks into Slack or Microsoft Teams.
Shared Team Bots in Slack and Teams, proactive initiative and @bot on X
Beyond video conferences, the Team Bots specification and Microsoft Teams integration guide show why SpaceXAI prioritised collaborative agents during the Grok Imagine layoffs. Even more notable is the behaviour documented in When Your Bot Takes Initiative: once a Bot learns a channel’s rhythm, it can reply unprompted when it spots a broken link, a duplicate ticket or a question already answered in its memory.
SpaceXAI also enabled Tag on X, allowing users to summon @bot in public X replies even as the Grok Imagine layoffs reshape the company’s consumer offering. However, UK and European readers should note a strict compliance block in the official manual: tagging @bot on X is unavailable in the United Kingdom, the European Union, Australia, Iceland, Liechtenstein and Norway.
Try Grok @Bot today!





Here is the central irony behind the Grok Imagine layoffs: even after gutting its video division to free up GPUs, SpaceXAI is racing to keep up with the token appetite of Grok Bot 1.0. Speaking on Y Combinator’s Lightcone podcast on 14 November 2026, core engineers Lauren Tan and Roman Ugarte revealed that Grok Bot went from first line of code to launch in seven weeks and has already reached 800,000 paid weekly active users, with more than 600,000 daily active users.
That 75 per cent daily-to-weekly engagement ratio collided head-on with serving costs. As reported by Business Insider, Elon Musk admitted on X on 1 November 2026 that the cost of serving Grok Bot was unsustainably high relative to revenue, prompting tighter enforcement of The X Subscription weekly usage limits.
@DogecoinNorway The cost of serving Grok Bot was unsustainably high relative to revenue
[...] You should get ~7% more usage out of your plan, and up to ~15% more on routine runs! Enjoy!
How The X Subscription shared pool and Team Bot billing work after the Grok Imagine layoffs
Under SpaceXAI’s unified subscription ledger after the Grok Imagine layoffs, X chat, remaining Grok Imagine prompts, Cursor coding tasks and Grok workplace agents all draw from one shared weekly pool. Relative to the $8-per-month Plus baseline (1x), Premium ($30/mo) provides 4x, Super ($100/mo) provides 15x and Ultra ($200/mo) provides 50x plus meeting-bot access. Despite Sam Sokolin’s 13 November harness patches adding roughly 7 per cent general efficiency and 15 per cent on routine runs in the same week as the Grok Imagine layoffs, three billing traps frequently catch buyers off guard:
- Caller-pays Team Bot metering: Every message sent to a shared Team Bot in Slack or Microsoft Teams bills the invoking user’s weekly pool; if a Bot runs via an unlinked Slack user or webhook, the burn falls back onto the Bot owner’s personal quota.
- Top-up credit pricing gap: Once a user hits 100 per cent of their weekly pool, pay-as-you-go top-up credits clear at a steeper per-token rate than the base plan: one subscriber documented getting 247 million tokens from his $30 Premium subscription over two days, whereas $121.49 in extra credits bought only 196.9 million tokens.
- Opaque sub-processor routing: As stated in the Grok Bot FAQ, there is no manual model picker even after the Grok Imagine layoffs; the harness routes steps across Grok checkpoints and external sub-processors including Anthropic’s Claude Opus 5.5, Midjourney and Suno.
@mattyp Hey Matt, I hit my usage limits in 2 days of usage and Grok Bot started eating away at my credits even though my additional usage is set at $0. That's fine, BUT my issue is that $121.49 in credits only got me $196.9M tokens whereas my subscription of $30/mo got me 247M tokens.
folkfox subscription & quota benchmark
How shared weekly usage pools gate Grok Bot after the Grok Imagine layoffs
| Tier | Monthly price | Weekly pool multiplier | Grok meeting agent & Team Bot rules |
|---|---|---|---|
| Free | $0 / month | Minimal trial cap | Basic chat and X search only; no meeting joins |
| Plus | $8 / month | 1x baseline pool | Shared Grok Bot and Grok Imagine pool; no meeting joins |
| Premium | $30 / month | 4x Plus pool | ~247M tokens reported before weekly cap; no meeting joins |
| Super | $100 / month | 15x Plus pool | Heavy background Bot tasks; no meeting joins on individual Super |
| Ultra | $200 / month | 50x Plus pool | Full Zoom, Microsoft Teams & Google Meet bot joins + priority routing |
| Team / Enterprise | $40 / seat or custom | Per-seat pooled & metered | Shared Slack & Teams bots + meeting joins (Enterprise adds SSO, SIEM & network controls) |
The $60 billion Cursor engine and the Grok 4.8 C and C++ link
The second beneficiary of the Grok Imagine layoffs is automated software engineering. On 15 August 2026, TechCrunch confirmed that SpaceX closed its $60 billion acquisition of Cursor, wiring the AI code editor directly into SpaceXAI’s Coding Tools track. Internally, every SpaceXAI staffer pairs with an EPD Teammate and hundreds of Cursor Cloud Agents that merge more than 100 pull requests per day (roughly 4,000 PRs per month), while external customers like commercial insurance brokerage Harper report saving $120,000+ per year.
Running hundreds of thousands of persistent Firecracker VMs and multi-step tool loops demanded a radical serving overhaul, which connects the Grok Imagine layoffs directly to our earlier report on Grok 4.8 reinforcement learning in C and C++. Led by SpaceX Starlink systems engineers, SpaceXAI dumped its legacy Python, JAX and SGLang stacks and rewrote its entire reinforcement learning and inference engine from scratch in bare-metal C and C++.
By stripping out Python interpreter stalls during multi-turn agent tool calls, the Grok 4.8 C and C++ stack lifts Model FLOPs Utilisation (MFU) to 62.5 per cent versus 51.2 per cent on JAX and 41.0 per cent on SGLang. As Elon Musk confirmed on 10 November 2026, Grok Bot users are the very first cohort receiving the Grok 4.8 soft rollout, letting SpaceXAI serve roughly 50 per cent more agent steps per megawatt on the silicon freed by the Grok Imagine layoffs.
At @SpaceXSI, Lauren Tan (@poteto) and Roman Ugarte (@romanugarte_) are on the core team that took Grok Bot from first line of code to launch in seven weeks. It lets you spin up AI teammates with their own identities and logins, and they just keep working in the background. Their marketing team gave a bot its own ad budget. Lauren built an entire feature with her bot in one Slack thread without ever opening Cursor. [...]
We have dumped JAX/SGLang and rewritten the entire RL and inference stack from scratch in C/C++, which took some extra time. Training (aka pre-training) is done. Lot of cleanup was needed on the SFT data to get that mostly right. Now, 4.8 is in RL and the evals are starting to climb. We will do a soft rollout to Bot users later this week and see how it goes. [...]





Where does SpaceXAI prowl next after the Grok Imagine layoffs? On 14 November 2026, replying to Robert Scoble on X, Elon Musk summed up the 2027 product roadmap in five words: There will be many agents. Internally, Musk has told engineers that autonomous agent-to-agent traffic will eventually outstrip human internet traffic by 1,000 to 1 as every corporate department spins up specialised digital workers.
@Scobleizer There will be many agents
Seven security and GDPR hurdles inside the Grok Bot Security FAQ after the Grok Imagine layoffs
Before Chief Information Security Officers or UK and European data-protection leads invite Grok workplace agents into sensitive calls in the wake of the Grok Imagine layoffs, inspect SpaceXAI’s Grok Bot Security FAQ alongside independent governance notes from European consultancy DataNorth AI. Seven architectural realities stand out:
- US-only Firecracker VMs (zero EU or UK data residency): Every Grok Bot virtual machine and persistent disk is hosted exclusively in United States data centres, creating immediate GDPR transfer hurdles for European public-sector and regulated financial workloads.
- External sub-processor routing: Because Grok Bot automatically delegates sub-tasks to Anthropic (Claude Opus 5.5), Midjourney and Suno, prompts and meeting transcripts can cross multiple third-party sub-processors.
- Team model allowlists bypassed: The Security FAQ explicitly notes that team model allowlists configured for Cursor or standard chat do not restrict Grok Bot, which picks its own execution models.
- Broad Microsoft Teams channel read scope: Adding Grok Bot to a Microsoft Teams team grants the bot read access to every standard channel inside that team, not just the single thread where it was tagged.
- Minimal call-recording notice: When a Grok meeting agent enters Zoom, Microsoft Teams or Google Meet, its only disclosure is the display name suffix
- transcribing, with no audio consent prompt for two-party consent jurisdictions. - No admin kill-switch or SIEM logs on self-serve Team plans: Self-serve Team plans ($40/seat) have no workspace-wide admin off-switch to block personal or Team Bots, no SIEM Audit Logs and no egress Network Controls; all three require an Enterprise contract.
- 50-hostname egress blocklist and Zscaler TLS exemptions: Even on Enterprise plans, network filtering uses a 50-hostname blocklist that still permits direct connections to bare IP addresses, and corporate networks using Zscaler TLS inspection must exempt
*.*.cursorvm.com.
| Workplace / meeting agent | Entry pricing | Persistent Linux VM? | Live meeting join & compliance posture |
|---|---|---|---|
| SpaceXAI Grok Bot 1.0 + Cursor | $8-$200/mo or $40/seat Team | Yes (Firecracker microVM per Bot) | Joins Zoom, Teams & Meet on Team/Ent/Ultra after Grok Imagine layoffs; US-only VMs; Enterprise needed for SIEM/SSO |
| Anthropic Claude Cowork + Code | $20/mo Pro; $100-$200/mo Max | Sandboxed desktop & cloud runners | Deep Slack, GitHub & enterprise Zero-Data-Retention; no native video-call attendee bot |
| Microsoft 365 Copilot (Wave 3) | $30 / user / month (annual) | No (Graph SaaS orchestrator) | Native Teams transcription, EU/UK tenant data residency & Purview audit governance |
| Zoom AI Companion 2.0 | Included with paid Zoom licences | No (workspace assistant) | Bot-free native Zoom capture with built-in host consent banners and admin controls |
| Otter.ai / Fireflies.ai | $8.33-$19/mo Pro; $19-$39/mo Biz | No (transcription & CRM sync) | Mature Zoom, Teams & Meet bot joins with explicit consent notices & SOC 2 controls |
| Granola AI | $14/mo Individual; $35/mo Biz | No (local macOS/Windows capture) | Zero-bot meeting notes capturing local system audio without placing a bot in the call lobby |
Five buyer checks after the Grok Imagine layoffs
- Decouple production pipelines from Grok Imagine APIs: With the Grok Imagine layoffs leaving only a skeleton maintenance crew, treat
Grok Imagine Video 1.5as a legacy endpoint and test fallback adapters to Google Veo 3, Seedance 2.5 or MiniMax H3. - Ring-fence shared weekly token pools: Because every Slack or Teams reply after the Grok Imagine layoffs bills the invoking user’s pool (and unmapped triggers bill the Bot owner), disable proactive initiative in high-traffic channels so casual chatter does not drain your coding quota in 48 hours.
- Mandate verbal consent before using a Grok meeting agent: On Ultra or Team plans, require hosts to announce recording aloud at the start of Zoom and Teams calls rather than relying solely on the
- transcribingdisplay suffix. - Segregate sensitive Microsoft Teams channels: Because Grok Bot can read every standard channel of any Microsoft Teams team it joins, keep board, HR, legal and M&A work in separate teams or private channels, or consider self-hosted open weights like Inkling 2 Small open weights and high-speed enterprise APIs like Gemini 4 Hydrogen.
- Require an Enterprise contract before loading Vault secrets: Never store production database credentials or regulated customer PII in Grok Bot Vaults on self-serve plans that lack SSO, SIEM audit logs and an admin kill-switch following the Grok Imagine layoffs.
If your organisation needs an independent technical audit spanning AI search strategy, agent governance and inference TCO after the Grok Imagine layoffs, our AI consultancy practice benchmarks every stack against raw telemetry and transparent published retainers.
Quick answers
Direct answers on the Grok Imagine layoffs, video API continuity, Zoom and Microsoft Teams meeting joins, weekly token pools and enterprise security controls.
Why did SpaceXAI carry out the Grok Imagine layoffs in November 2026?
SpaceX Super Intelligence (SpaceXAI, formerly xAI) executed the Grok Imagine layoffs across roughly 60 employees in early November 2026 to reallocate GPU compute and engineering budget toward enterprise AI coding tools (following its $60 billion Cursor acquisition) and Grok Bot 1.0 workplace agents.
Is Grok Imagine shutting down completely after the Grok Imagine layoffs?
No. After the Grok Imagine layoffs, Grok Imagine remains live inside the X app, the standalone Grok mobile apps and the xAI API, kept running by a small maintenance crew. However, with more than half of the division dismissed, major feature upgrades beyond Grok Imagine Video 1.5 and Image 2.0 are effectively frozen.
Which subscription plans let a Grok meeting agent join Zoom, Microsoft Teams and Google Meet calls?
Following the Grok Imagine layoffs and the shift to enterprise tools, live meeting participation for Grok Bot 1.0 is available on self-serve Team plans ($40 per seat per month), custom Enterprise contracts and the individual $200-per-month Ultra tier of The X Subscription. Free, Plus ($8/mo), Premium ($30/mo) and individual Super ($100/mo) plans do not include meeting joins.
Why are Grok Bot 1.0 users hitting weekly usage limits so quickly after the Grok Imagine layoffs?
On 1 November 2026, just before the Grok Imagine layoffs broke, Elon Musk confirmed that the cost of serving Grok Bot was unsustainably high relative to revenue, prompting SpaceXAI to enforce stricter shared weekly pools under The X Subscription (where Plus is 1x, Premium is 4x, Super is 15x and Ultra is 50x). Persistent background tasks and automatic routing across premium sub-processors can drain a $30 Premium quota (~247M tokens) in two days.
How does the Grok 4.8 C and C++ rewrite help Grok Bot workplace agents after the Grok Imagine layoffs?
Alongside the Grok Imagine layoffs, SpaceXAI dumped its legacy Python, JAX and SGLang stacks and rewrote its reinforcement learning and inference engine from scratch in bare-metal C and C++, lifting Model FLOPs Utilisation (MFU) from 41.0% on SGLang to 62.5%. That efficiency gain lowers per-step agent cost, which is why Grok Bot users are receiving the Grok 4.8 soft rollout first.
What are the biggest security and GDPR risks of using Grok Bot in Slack, Teams and Zoom?
According to SpaceXAI’s Security FAQ, every Grok Bot built after the Grok Imagine layoffs pivot runs on US-only Firecracker microVMs with no EU or UK data residency, routes tasks across external sub-processors (Anthropic, Midjourney and Suno), ignores team model allowlists and reads every standard channel of any Microsoft Teams team it joins, while self-serve Team plans lack an admin kill-switch or SIEM audit logs.
Keep reading
Read more on this topic
Grok 4.8 reinforcement learning: 5 essential changes
How SpaceX Starlink engineers dumped JAX and SGLang for a bare-metal C and C++ stack hitting 62.5% MFU and powering the Grok Bot soft rollout.
Read the pieceOpen-weight frontierInkling 2 Small open weights: 594B Apache 2.0 under the bonnet
Why self-hosted Apache 2.0 Mixture-of-Experts weights are winning regulated European workloads that cannot accept US-only agent VMs.
Read the pieceHigh-speed inferenceGemini 4 Hydrogen: 5 essential facts on speed, cost and AI Overviews
Inside Google’s ultra-low-latency frontier tier and how hyperscale inference economics are reshaping enterprise search and coding.
Read the pieceWork with folkfox
Audit workplace agent economics and GDPR risk before rolling out seats
folkfox advises technical buyers on AI content systems, enterprise agent governance, technical positioning and model evaluation strategy across UK, European and global teams.
Want folkfox in your Google results? Set folkfox as a preferred source. Explore our newsroom background and latest dispatches.
Sources
Primary reporting dispatches, official docs.x.ai technical manuals, corporate filings and archived X records verified 14 November 2026.
- The Information: Elon Musk’s AI Group Guts Grok Video Generation Team (Aaron Tilley, Claudia Kim & Juro Osawa, 13 November 2026)
- AI Weekly Alert: SpaceX Super Intelligence cuts ~60 from Grok Imagine video team (13 November 2026)
- Startup Fortune: SpaceXAI cuts about 60 Grok Imagine staff to push coding tools and agents (November 2026)
- BigGo Finance / 36kr: SpaceX AI multimodal headcount and Grok Imagine restructuring breakdown (14 November 2026)
- TechCrunch: SpaceX officially closes its $60B Cursor acquisition (15 August 2026)
- Latent Space Wiki: xAI Grok Imagine technical history from Aurora MoE to Video 1.5 and Image 2.0
- xAI Product Portal: Grok Bot 1.0 overview and workspace onboarding
- xAI Newsroom: Introducing Grok Bot 1.0 for Slack, Microsoft Teams and live meetings
- xAI Newsroom: Shared Team Bots architecture and enterprise workflows
- xAI Technical Documentation: How Grok Bot Works (Firecracker Linux VMs, browser, shell & Vault)
- xAI Technical Documentation: Grok Bot Meetings (Zoom, Microsoft Teams & Google Meet integration)
- xAI Technical Documentation: Grok Bot for Microsoft Teams permissions and channel behaviour
- xAI Technical Documentation: Team Bots shared billing, triggers and seat rules
- xAI Technical Documentation: When Your Bot Takes Initiative in Slack and Teams threads
- xAI Technical Documentation: Tag @bot on X and UK/EEA/Australia regional restrictions
- xAI Technical Documentation: Grok Bot Security FAQ (US VMs, sub-processors, SIEM & network controls)
- xAI Technical Documentation: Grok Bot General FAQ and automatic model routing
- xAI Technical Documentation: The X Subscription tiers (Plus 1x, Premium 4x, Super 15x, Ultra 50x) and weekly limits
- DataNorth AI: Grok Bot 1.0 in Slack and Microsoft Teams enterprise governance analysis (November 2026)
- Business Insider: Grok Bot usage limits tightened as Elon Musk cites unsustainably high serving costs (November 2026)
Reporting: the folkfox newsroom, from 20 primary sources and 14 verified X records, checked 14 November 2026. Illustrations are conceptual watercolour plates of the folkfox vixen made for this special report. Nothing here is legal or financial advice.

