Cursor Is Part of SpaceX Now — What to Change in Your Agent Stack (and What Not To)

One-line takeaway: 2026-08-14 is an org close. Update model pickers, privacy checks, and Copilot admin gates from product notices already shipped — do not rewrite your stack from the $60B headline alone.

On 14 August 2026, Cursor said the SpaceX acquisition is officially complete, finishing the path that started with the April SpaceXAI partnership (Cursor Blog). The Verge reported the same-day close of a $60 billion deal (The Verge), and an SEC filing marks the merger effective 14 August 2026 with Cursor equity converting into SpaceX Class A consideration at an implied Cursor equity value of $60.0 billion (SEC filing).

What engineers actually touched that week landed on different days. Grok 4.6 shipped in Cursor and Grok Build on 12 August, AIUC-1 certification landed on 13 August, and on close day GitHub Copilot started rolling Grok 4.6 out across multiple surfaces. This post is a settings-and-governance read, not an M&A hot take.

Why it matters

“Who owns the IDE vendor?” sounds like Slack drama. In an agent stack it splits into three concrete jobs:

1. Model supply and economics — Cursor’s close note ties mega-scale GPU access to stronger, more economical models and points at Grok 4.6 (released Wednesday) as an early sample (joining-spacex, grok-4-6).
2. Trust evidence for buyers — AIUC-1 pairs an organizational controls audit with adversarial testing of agent behavior; Cursor cites Schellman’s review and recurring evaluation (at least quarterly, plus a full annual audit) (aiuc-1).
3. Dual-surface policy — When the same model name appears in Cursor and Copilot, team allowlists must cover both product admin surfaces (Copilot changelog).

The failure mode is jumping from the headline to “turn off every non-Grok model.” That is not what the close post says.

How it compares

Read the week as four event types, not one acquisition story.

Compare along announcement kind:

• Org event: ownership and filings. Settings panels do not auto-rewrite.
• Model event: list price and availability. Grok 4.6 starts at $2 / $6 per million input/output tokens; fast variant at 2×; Cursor and Grok Build get 2× included usage for the first week (Cursor, SpaceXAI).
• Trust event: independent certification you can hand to a CISO — still not a substitute for repo hooks and secret hygiene.
• Distribution event: on Copilot Business/Enterprise, the Grok 4.6 policy is off by default until an admin enables it (GitHub Changelog).

Config / checklist example

A minimal team checklist (no invented toggle names):

# Post-close agent stack checklist (2026-08-14+)

## Cursor
- [ ] Model picker: document Grok 4.6 trial scope (which repos / agents)
- [ ] Re-verify Privacy Mode / retention still match team policy
      (do not assume close-day changed them — check product + terms)
- [ ] Note first-week 2× included usage for Grok 4.6 in Cursor/Grok Build
      (primary wording: "first week" after 2026-08-12)
- [ ] Keep a written multi-model fallback if Grok is rate-limited

## GitHub Copilot (if you also run it)
- [ ] Business/Enterprise: admin enables Grok 4.6 policy (off by default)
- [ ] Confirm which surfaces see the model (VS Code / CLI / cloud agent / …)
- [ ] Do not assume Copilot usage pricing equals SpaceXAI API $2/$6 list

## Portable extensions
- [ ] Prefer Agent Plugins 1.0 (`plugin.json` + skills/ + mcp.json) for
      skills/MCP that must travel across Cursor and Copilot-compatible clients

Cursor’s docs support Agent Plugins (root plugin.json) alongside Cursor Plugins (.cursor-plugin/plugin.json) (Cursor docs/plugins). Ownership can consolidate; portable skill packages still lower exit cost.

{
  "$schema": "https://agent-plugins.org/schemas/1.0.0/plugin.schema.json",
  "name": "team.deploy-guards",
  "version": "1.0.0",
  "description": "Portable skills + MCP for deploy guardrails"
}

Practical use

1. Split the Slack memo. One line for “SpaceX close complete (14 Aug)” and one line for “Grok 4.6 in Cursor since 12 Aug; Copilot rollout 14 Aug (Biz/Ent admin gate).”
2. Do not crown a promo default. First-week 2× included usage is experiment budget, not a quarterly architecture decision.
3. File a ticket for the Copilot admin gate. Off-by-default is itself a governance choice — enable or explicitly refuse.
4. File AIUC-1 materials for procurement, and note the recurring-evaluation premise so nobody treats it as a one-shot stamp.

I almost blocked a weekend to “rewrite our model policy after the acquisition,” then lined the primary posts by date and realized half the checklist was already product news and half was still “unknown until terms update.” The weekend shrank to verification, not mythology.

Senior-engineer perspective

The real risk is concentration, not the press cycle. If Cursor’s roadmap narrative is now tightly coupled to SpaceXAI compute and Grok product lines, incentives can tilt the default model menu over time. The counterweight in the same week is portability: Agent Plugins 1.0 and multi-client distribution give you a way to keep skills/MCP from living only inside one vendor’s sidebar.

If Privacy Mode, training terms, or third-party model availability did not change in a primary notice, do not encode rumors into the runbook. Seniors earn their keep by refusing to promote speculation to policy.

Using it in Cursor

1. Enable Grok 4.6 on a bounded agent surface and run one long-horizon task as your baseline comparison (the positioning Cursor/SpaceXAI emphasize).
2. Keep rules/hooks that ban secret commits and require human publish/deploy — AIUC-1’s secure-defaults story complements that; it does not replace it.
3. Prefer Agent Plugins shape for new shared skills; add .cursor-plugin only for Cursor-specific hooks/rules.
4. Do not conflate close day with Cloud Agent Builds performance work — that landed as its own product thread.

FAQ

Q. Does the close remove Claude/GPT from Cursor?
A. The close post does not say that. Wait for an explicit product or terms update.

Q. Why don’t I see Grok 4.6 in Copilot yet?
A. Rollout is gradual. Business/Enterprise needs the admin policy enabled (changelog).

Q. Does AIUC-1 mean we can skip our security review?
A. No. It is procurement evidence, not a substitute for repo secrets, hooks, and least privilege. Recurring evaluation is part of the story.

Q. Should $60B alone change our default model?
A. No. Defaults should follow measured agent success rate, latency, and the invoice — not valuation headlines.

Sources

• Cursor is now a part of SpaceX
• The Verge — Cursor is officially part of SpaceX
• SEC filing — Merger effective 2026-08-14
• Introducing Grok 4.6 · Cursor
• Introducing Grok 4.6 | SpaceXAI
• Grok 4.6 is now available in GitHub Copilot
• Grok 4.6 in GitHub Copilot | SpaceXAI
• Cursor earns AIUC-1 certification
• Cursor docs — Plugins
• Agent Plugins

Closing

This week’s useful work is dating the switches: close for governance, Grok 4.6 for pickers and promos, AIUC-1 for trust packets, Copilot’s off-by-default policy for admin tickets. Looking ahead, team defaults will keep drifting toward whatever combination of agent success rate and token bill survives a quiet month — not toward whoever won the Friday headline.