Today's issue: Colin says "All screens in use" can be misleading when other Bots still work, a correction on the 0.70.2 idle fix, and last night's short Grok Bot outage.

Full issue: Oct 13, 2026

What's new

No new version is on the changelog. It still ends at 0.68.1 from Oct 7. Staff continue to name 0.70.2 on the forum, but it is not in the changelog. Cursor's status page is All Systems Operational this morning after a short overnight incident. (confirmed)

All screens in use can mean one Bot's screen failed to start (confirmed)

On Oct 13 at 4:14 AM ET, Colin said a shared-computer "All screens on the shared computer are in use" error "isn't a capacity problem" when other Bots on that computer still have screens. The message "is misleading in this case." Staff are tracking it as a known issue.

Who is affected: Anyone running several Bots that need browser screens on one Bot computer, especially overnight unattended browser jobs.

What to do now: 1. See whether other Bots on the same computer can still screenshot. 2. Retry once, then try Update Grok Bot's Computer if it keeps failing. 3. If it returns, file times and timezone in Bug Reports rather than Resetting first.

Correction check on the 0.70.2 idle fix (confirmed staff conflict)

Our Oct 12 fix quoted mohitjain that 0.70.2 keeps the local helper running when the window is unfocused. Later the same day at 4:49 PM ET, deanrie still described the Mac as away after about 10 minutes out of focus or when the screen is locked, and said the helper "doesn't start again automatically when the bot wakes up for a scheduled job." We reopened the known issue for re-test. Screen lock and scheduled-job restart stay open. 0.70.2 is a build staff named on the forum, not in the changelog, and whether it fixes this issue is now disputed. (confirmed)

Overnight Cursor status incident hit Grok Bot (confirmed)

From 8:22 PM to 11:07 PM ET on Oct 12, status.cursor.com logged "Degraded performance and errors across Cursor services," including the IDE, Cloud Agents, Automations, Grok Bot, and Origin. Grok Bot was marked recovered at about 9:49 PM ET. The incident resolved at 11:07 PM ET. This morning the page shows All Systems Operational. (confirmed)

The X Subscription follow-up (confirmed)

The Oct 12 unify story still holds. New this morning: bundled-subscription-terms (Last Updated Oct 7, 2026) name X Subscription Corp as merchant of record, and Anysphere as seller for Cursor Top-Ups. Absolute pool sizes are still unpublished. At 3:43 AM ET Colin answered that included third-party usage stays with your existing Cursor plan for as long as you keep it, while on The X Subscription third-party models are "billed separately at API list prices through top-ups," with an optional spending cap. At 4:07 AM ET he added that "All 3rd party model usage on the X Subscription is from top-ups."

PersonalAgentBench (confirmed)

micro1's PersonalAgentBench now ranks six personal agents. Grok Bot is tied fourth at 65% task and trusted completion, behind OpenAI's Dots and Hark. Details: Hark / PersonalAgentBench.

Beyond Grok Bot

• GPT-6.1 Sol Codex and Work degraded; mitigation monitoring
• GPT-6.1 Sol default context jumps to 500k in Codex and Work
• Managed Agents dynamic workflows fan out hundreds of tasks
• Early evidence: Anthropic says distillation harvests fell 90 to 99 percent

How people are using it

Setup of the day: Check every draft invoice was actually sent. Sunnyvale's morning pass marks the tracker Sent only after QuickBooks shows a real send (a ChatGPT stack, per their notes). Same author also runs a Monday collections draft queue that leaves Gmail drafts for a human to send.

• A generic Bot failed reply can mean your trial usage is gone
• Keep the method in a skill, and put run data in a git repo
• Use Tailscale on the machine when the local helper keeps dropping

The digest

Screens and overnight work. Devon_Lybbert's Oct 12 report of overnight browser stalls with a stuck "all screens in use" error got a same-morning staff confirmation that the message can be wrong when other Bots still work. See the standalone known-issue.

Idle helper still needs re-test on 0.70.2. mohitjain's early Oct 12 posts named an unfocused-window fix. deanrie's 4:49 PM ET reply the same day still described away after ~10 minutes unfocused, screen lock, helper stop, and no restart for scheduled jobs. 0.70.2 is a forum-named build, not in the changelog, and whether it fixes this is disputed. We reopened the known issue.

X Subscription reaction. Colin answered the biggest question overnight: on The X Subscription, third-party models in Cursor are paid from top-ups at API list prices, not from the shared pool, while existing Cursor plans keep their included third-party usage. He also said nothing happens automatically to existing Cursor Pro plans, and that users who don't migrate will be moved later "with advance notice." Premium ($30) is the comparable plan to Cursor Pro, and Pro+ ($60) has no direct equivalent. Absolute numbers for the Plus-multiple pools are still not on x.ai/pricing.

No changelog bump. 0.68.1 remains the latest listed build. Linux users may still need the staff-posted 0.70.2 .deb if apt has not caught up.

Workflow: a Managed Agents morning brief. Anthropic's Oct 8 reference automation runs a scheduled Claude Managed Agents session that reads bookmarked Slack channels and GitHub PRs and posts one dated brief to Slack. We have not run it ourselves; it needs a Slack app, a read-only GitHub token and the Managed Agents beta.

Read the full issue and more: thedailybot.vercel.app