Claude Doubled Your Limits, Moved Into Office, and Started Scanning Your Sessions
Claude Pro limits doubled, Word/Excel/PowerPoint add-ins GA, and dreaming scans past sessions by default. Three changes this week — what you need to know.
Published: May 12, 2026 Impact: High — capacity upgrade, new Office integration, background memory feature
Three Changes in Seven Days
Between May 6 and May 7, Anthropic shipped three meaningful Claude updates. Two are unambiguously good for paying users. One needs scrutiny.
The rare headline here is that Claude gave something back — doubled rate limits with no price increase — while quietly introducing a background feature that reads your past sessions without prominent disclosure.
On May 6, at Anthropic's Code with Claude developer event in San Francisco, Claude Code's five-hour rate limit was doubled for Pro, Max, Team, and seat-based Enterprise plans. Peak-hour throttling dropped entirely for Pro and Max. On May 7, Claude launched as a generally available add-in for Word, Excel, and PowerPoint — with Outlook entering public beta. The same day, Anthropic announced "dreaming" for Claude Managed Agents: a scheduled background process that scans your agent's past sessions and reorganizes its memory store between tasks.
The Limits Doubled — Here's the Math
Claude Code now gets twice the five-hour window it had before. Peak-hour restrictions on Pro and Max are gone. Anthropic credits a new compute partnership with SpaceX — over 300 megawatts of new capacity, including more than 220,000 NVIDIA GPUs coming online within weeks.
The free tier did not get an upgrade.
This is the inverse of the typical watchdog story. Anthropic added capacity and passed it directly to paying users. No price increase, no new tier required. If you're on Pro, Max, Team, or Enterprise: your Claude Code limits doubled automatically on May 6.
The Office Integration Changes the Competitive Math
Claude is now available as add-ins for Word, Excel, and PowerPoint (generally available) and Outlook (public beta). All four install from the Microsoft Marketplace and work on Windows, Mac, and the web version of Office. Cross-app context continuity is the key feature: Claude carries your full conversation context as you move between apps.
No additional charge. Any paid Claude plan — Pro ($20/month), Team, or Enterprise — includes the Office add-ins.
Here is why this matters competitively:
| Option | Monthly Cost | Office AI |
|---|---|---|
| Claude Pro + Microsoft 365 Personal | ~$30 | Claude in Word, Excel, PPT, Outlook |
| Microsoft Copilot add-on | $30 | Copilot in Word, Excel, PPT, Outlook |
At roughly equivalent cost, Claude Pro subscribers who already pay for Microsoft 365 now get Claude embedded in Office at no extra charge. The bundling math may favor Claude over Copilot for users who already subscribe to both.
What This Costs You: The Dreaming Feature
This is the section that needs careful reading.
Anthropic launched dreaming on May 6 as part of Claude Managed Agents (research preview). Dreaming is a scheduled background process — default cadence: daily — that reviews your agent's past session transcripts and existing memory store, extracts patterns, merges duplicates, replaces stale entries, and produces a reorganized memory store for future sessions.
Precision matters here: this is not model training. Dreaming does not update Claude's weights. It updates a structured memory store that future agent sessions read from. The distinction is real and important.
What you are trading:
- Past session transcripts — the dreaming process reads them. These contain your agent's full work history: prompts, outputs, tool calls, decisions.
- Memory stores — reorganized and rewritten automatically, by default, without per-change approval.
- Opt-out status — users can switch to "review before apply" mode, meaning they approve memory changes before they land. But the background scan of past sessions still runs. There is no documented opt-out of the scan itself.
Anthropic's documentation gives developers control over two things: the cadence (hourly, daily, weekly, or manual) and whether memory updates apply automatically or wait for review. What it does not document is a way to prevent the scan from happening at all while keeping the feature enabled.
Verdict: For developers building autonomous agents, this trade-off is likely acceptable — the whole point of managed agents is persistent improvement. For users handling sensitive client data inside agent sessions, review the "review before apply" setting and set it immediately. Default-on background scans of past sessions deserve more prominent disclosure than a blog post.
What to Do Right Now
If you use Claude Code (Pro, Max, Team, or Enterprise): Your limits doubled on May 6. No action required.
If you use Microsoft 365: Install the Claude add-in from the Microsoft Marketplace. It works on your existing paid Claude plan. Test it in Word or Excel first — cross-app context is the differentiating feature.
If you use Claude Managed Agents with dreaming enabled:
- Go to Claude Console → your agent → Dreaming settings
- Under "Dreaming cadence," verify the schedule matches your tolerance
- Under memory update control, switch from "automatic" to "review before apply" if your sessions contain sensitive data
- Check Anthropic's documentation for any opt-out of the scan itself as the feature moves from research preview to general availability
What Happens Next
30 days: Outlook integration exits beta (or Anthropic communicates a timeline). Watch whether dreaming's documentation clarifies opt-out scope before GA.
90 days: Claude hit #1 on the Apple App Store in February 2026. Every prior vendor that hit #1 eventually compressed the free tier. Claude hasn't yet — but the compute deal that enabled today's limit doubling is also the capacity base that makes free-tier restrictions easier to impose selectively.
6 months: The SpaceX compute deal's 300+ MW comes fully online. If Anthropic passes additional capacity to users rather than banking it for enterprise, the limits-doubling pattern could repeat.
Bottom Line
Claude gave paying users more for the same price. That earns a positive mark on the watchdog scorecard.
The Office integration is genuinely useful and the competitive math is real: Claude Pro plus a Microsoft 365 subscription you already pay for now includes full Office AI for no additional cost.
The dreaming feature is the one to watch. The technology is legitimate — it is not training on your data. But default-on background scans of past sessions, with a "review before apply" option rather than a true opt-out, is a disclosure pattern that warrants scrutiny. Anthropic should make this setting prominent before dreaming exits research preview.
Claude's watchdog score holds. The limits upgrade and Office integration are positive. The dreaming disclosure gap is a flag, not a deduction — but only if it gets fixed before GA.
Sources
- Code with Claude 2026 — Anthropic's Developer Conference — Anthropic (May 6, 2026)
- New in Claude Managed Agents: dreaming, outcomes, and multiagent orchestration — Anthropic (May 6–7, 2026)
- Dreams — Claude API Docs — Anthropic
- Claude for Microsoft 365 — Anthropic
- Anthropic doubles Claude Code limits, thanks to a deal with SpaceX — PCWorld (May 2026)
- Anthropic is doubling Claude Code rate limits after deal with SpaceX — Engadget (May 2026)
- Anthropic updates Claude Managed Agents with three new features — 9to5Mac (May 7, 2026)
Share This Article
"Claude just doubled your coding limits AND moved into Word, Excel, and PowerPoint — for no extra charge beyond your $20/month. Meanwhile Microsoft charges $30/month for Copilot to do the same thing."
"Claude's 'dreaming' feature scans your past agent sessions every day by default. Not model training — but a background read of your history with no opt-out documented anywhere."
"Every vendor that hit #1 on the App Store eventually restricted the free tier. Claude hit #1 in February. The clock is ticking. Lock in Pro now if you're considering it."
"Claude doubled rate limits because of a SpaceX compute deal — 220,000+ GPUs. This is the rare watchdog story where the vendor gives something back instead of taking it away."