The Stack Decision: Bundled Copilot vs. Your Standalone AI Sub
The Departure
Last week I told you to wait a week before adding another AI subscription. The week is up. Microsoft's new bundled Microsoft 365 SKUs went live on July 1 with Copilot built in, and the numbers are public now.
So you can run the math instead of guessing. You got a number two weeks ago. This is the week you make the call with it.
The Co-Pilot
Tool: Claude Opus 4.8 via claude.ai. Still the stable workhorse. Fable 5 is back this week but on a throttled, short-term window (more on that below), so run your analysis on Opus, not on a model you might lose again in five days.
The Use Case: A paste-and-go stack decision. Drop in your current Microsoft 365 plan and cost, plus your standalone AI subscription and cost, and you get back a switch, stay, or test-for-30-days call with the monthly and annual dollar difference attached.
The Prompt (paste two things first: your M365 plan and its monthly cost, and your standalone AI subscription(s) with the monthly cost. Then add this):
"You are a fractional CFO helping me decide whether to switch to Microsoft's new bundled Microsoft 365 Business Standard with Copilot ($23.50/user/month) or Business Premium with Copilot ($32/user/month), replacing what I pay now. Here's my current setup: [paste my M365 plan and cost, plus my standalone AI subscription(s) and cost]. Do this: (1) add up what I pay per user per month across my M365 plan and my standalone AI sub(s); (2) compare that total to the bundled Copilot tier that matches my current feature level; (3) list what I'd gain and what I'd lose in the switch; (4) flag if any of my AI use sounds like Copilot Cowork's heavier automation (multi-file work, long agent tasks), since that may still bill separately by credit even inside the bundle; (5) give me one call: switch, stay, or test 30 days first, with the monthly and annual dollar difference."
New here, or did this get forwarded to you? You don't need any prior dispatch. Name your M365 plan and cost, add your standalone AI sub and cost, run it cold.
And it runs on your phone. Type the two lines into the Claude app from a gate or a back seat, run it, and screenshot the call.
The Upgrade
Topic: The math to sanity-check, and why this week's news proves last month's point
One number to know before you run the prompt. Business Basic is $7/user/month, but that's a light AI chat tier, not full in-app Copilot. Business Standard with Copilot is $23.50, and that's the real one: work-grounded Copilot inside Word, Excel, PowerPoint, Outlook, and Teams. Business Premium with Copilot is $32, the same Copilot set plus device security. The standalone Copilot add-on is $21 list, promoted at $18. So a team on Basic ($7) plus a standalone AI sub (around $20) is already near $27/user/month, versus $23.50 bundled, which also includes full desktop Office. That's roughly $3.50 per seat per month, about $42 per year. For some setups, not all. And the real question isn't the $3.50. It's whether bundled Copilot does the job your standalone tool does today. Run the prompt and answer that first.
One thing I can't confirm, so I won't pretend I can. The bundle covers in-app Copilot. But Microsoft's Cowork page still says usage-based billing applies, with no visible carve-out for the new bundles. So the heavier multi-tool automation still appears to bill separately by credit. Test it before you assume it's in. (Same for Microsoft's "30-40% cheaper than Claude" line: their own number, untested. Don't bank on it.)
Now the news, because it proved the point from three weeks ago. Fable 5 and Mythos came back online on July 1 after 18 days dark, with a stronger safety classifier, and were included in subscriptions only through July 7 at up to 50% of weekly limits. After that, it moves to metered credits. Don't rebuild your workflows around it this week. Use the window, keep the artifact, let the model stay borrowed.
Then look at what happened to OpenAI. Days before Fable returned, the White House restricted OpenAI's new flagship top tier, "Sol," to roughly 20 vetted government partners. Regulators consider it the same capability threshold as Mythos (per The Rundown's reporting). Two labs, two models, same pattern in three weeks. That's how frontier AI works in the US right now. Plan the stack around the pattern, not around any one model.
And one more for the cost audit. Sonnet 5 shipped July 1 with lower API pricing ($2/$10 per million tokens through August 31, then $3/$15). That's API pricing, not a change to your Claude subscription. But early reports say it can cost about the same per task as Opus because it works longer to finish the job (per Ben's Bites). A lower sticker price isn't automatically a lower bill.
Skipping this week: OpenAI's Jalapeno chip (no new details) and Sakana's Fugu (still unverified benchmarks).
The Landing
Fifteen minutes today. Open Claude Opus 4.8. Paste your current M365 plan and cost, add your standalone AI sub and cost, and run the stack-decision prompt. Get your switch, stay, or test-for-30-days call with the dollar difference attached.
That's the follow-through on the audit you ran two weeks ago. It's the first Thursday of Q3. Make the call instead of carrying the question into another month.
Safe Travels,
Marcellus