Microsoft's Copilot Merger Adds a Paid Agent Tier

On July 29, Satya Nadella used Microsoft’s fiscal Q4 2026 earnings call to confirm what an internal memo had leaked four weeks earlier. Copilot Chat, GitHub Copilot, Copilot Cowork, and a new agent layer called AutoPilot are collapsing into a single app. His framing on the call, as reported by DevOps.com: “Copilot is evolving rapidly from chat to Cowork to Autopilots…we are bringing these Copilot experiences together, including code, in one super app.”

The consolidation is the headline. The billing line underneath it is the story.

AutoPilot is a new paid tier for always-on background agents, layered on top of seat pricing you already bought. It arrives roughly 60 days after GitHub Copilot moved to usage-based AI Credit billing on June 1, which I covered in Copilot’s Flat Rate Dies June 1, and roughly 45 days after Copilot Cowork went generally available on its own metered credit system. That’s three separate pricing changes to the same product family inside one fiscal year, and the third one lands before the second one’s promotional pricing expires.

Quick Verdict

Question	The Answer
Who confirmed it?	Satya Nadella, Microsoft FY26 Q4 earnings call, July 29, 2026.
What’s merging?	Copilot Chat, GitHub Copilot, Copilot Cowork, and the new AutoPilot agent layer.
When does it ship?	The July memo said August. Nadella said “this quarter,” which runs through October.
What is AutoPilot?	A paid tier for persistent background agents: cross-calendar scheduling, email summarization, recurring workflows.
What does AutoPilot cost?	Undisclosed. No public rate card as of August 7.
Is it included in your seat?	No. Reporting describes it as an additional paid subscription on top of existing seat pricing.
What’s getting cut?	Copilot Podcasts and Copilot Labs, per the internal memo.
Why now?	Fewer than 4.5% of 450 million commercial Microsoft 365 customers pay for Copilot features.
How many meters will you run?	Up to four: M365 seat, GitHub AI Credits, Copilot Credits for Cowork, and AutoPilot.
The immediate risk	Signing an AutoPilot commit before the September 1 GitHub credit cliff resets your real baseline.

What Nadella Actually Confirmed, and What He Didn’t

The earnings call was a good quarter to bury a pricing change in. Microsoft posted $90.0 billion in Q4 revenue, $331.8 billion for the fiscal year, Azure past $100 billion for the first time, and Microsoft 365 Copilot at more than 30 million paid seats. GitHub Copilot was cited at 50 million users on the same call.

Against that backdrop, “one super app” reads like a product simplification story. Read the memo that preceded it and the framing shifts.

The Decoder reported on July 3 that Copilot EVP Jacob Andreou told the roughly 11,000-person Copilot organization the product must “earn the right to exist,” and that the team had “stripped out what wasn’t working.” The Information broke the memo first. Inc. covered the internal messaging angle a few days later. That is not the language of a company simplifying a product surface. That’s the language of a company that has to monetize a distribution advantage it has not converted.

Two things Nadella did not say on the call, and both matter to a buyer.

He did not give a ship date. The memo pointed at August. On the call he said “this quarter,” which under Microsoft’s fiscal calendar means anything through the end of October. A two-month spread on a launch that changes how four products bill is a spread you have to plan against, not around.

He did not price AutoPilot. No rate card, no included allowance, no statement on whether it rides the existing Copilot Credit meter or gets its own. Every enterprise buyer sizing an FY27 budget right now is placing a line item with no number in it.

What is Microsoft’s AutoPilot tier?

AutoPilot is Microsoft’s new paid tier for persistent background agents inside the unified Copilot app. Unlike Copilot Chat, which responds when prompted, AutoPilot agents run continuously against your Microsoft Graph data to handle recurring work: scheduling across multiple attendees’ calendars, summarizing email threads, and maintaining document workflows. Microsoft has not published pricing. The first named agent is Microsoft Scout.

The Meter Stack Nobody Has Drawn Yet

Here’s what makes this different from a normal product consolidation. The four things merging into one app do not share one billing model. They arrive with three separate meters already running, and AutoPilot makes four.

Surface	Billing model	Status as of Aug 7
Microsoft 365 Copilot seat	Per-user subscription	Existing, 30M+ paid seats
GitHub Copilot	Usage-based GitHub AI Credits, 1 credit = $0.01	Live since June 1, 2026
Copilot Cowork	Usage-based Copilot Credits, $0.01/credit PayGo	GA June 16, 2026
AutoPilot	Additional paid subscription, structure undisclosed	Announced, unpriced

Copilot Cowork is the one most buyers have not looked at closely, and it’s the best available proxy for how AutoPilot will bill. Microsoft’s GA announcement is specific: Cowork bills on usage-based Copilot Credits calculated from four inputs, which Microsoft lists as “model use, context retrieval, tool calls, and runtime.” Pay-as-you-go pricing is $0.01 per credit, with a P3 commitment option for discounts. It’s off by default and requires a Microsoft 365 Copilot license before the meter can even start.

Read that fourth input again. Runtime. For a Cowork task that a human kicks off and watches, runtime is bounded by the task. For an always-on background agent, runtime is the product. That’s the whole pitch.

Cowork also came out of Microsoft’s Anthropic partnership. Microsoft’s March announcement describes bringing the technology behind Claude Cowork into Microsoft 365, and Microsoft’s own testing put Copilot Cowork at 30 to 40 percent cheaper per prompt than Claude Cowork. Useful benchmark. Also a reminder that the cost floor on this tier is set by a model vendor Microsoft doesn’t fully control, which is exactly why running AI agents keeps repricing upward.

AINave’s breakdown estimates an AutoPilot background task at 10 to 20 model calls, against one call for a standard chat query. That’s an estimate, not a Microsoft figure, and I’d treat it as directional. The direction is the point.

What should enterprise buyers do before the Copilot merger ships?

Five moves, in order, all doable inside two weeks:

1. Inventory every Copilot meter you’re already running. M365 Copilot seats, GitHub Copilot AI Credits, Cowork Copilot Credits. Most orgs have at least two live and have never seen them on one page.
2. Model Q4 against the standard GitHub credit allowance, not the promo. Enterprise promotional credits drop from 7,000 to 3,900 per user on September 1. That reversion is a contract term, not a forecast.
3. Confirm Cowork’s default state in your tenant. It ships off by default and needs admin activation. If someone enabled it in June for a pilot, the meter has been running for seven weeks.
4. Set tenant, group, and user spend limits before AutoPilot ships. Microsoft already exposes these controls for Cowork credits. Configure them now so the ceiling exists before the new tier lands inside it.
5. Put an unpriced placeholder line in the FY27 budget and defend it. Not a guess at the number. A named line with an owner, so the first invoice lands against a decision instead of a surprise.

Why Microsoft Is Doing This

The conversion math explains the merger better than any product rationale.

Reporting on the Andreou memo puts paid Copilot conversion at under 4.5% of Microsoft’s 450 million commercial Microsoft 365 customers, with only 20 to 30 percent of those payers using it weekly. Microsoft has the largest distribution footprint in enterprise software and, by its own internal accounting, has not turned it into engaged paying users.

So the strategy makes sense on Microsoft’s side. Consolidate four half-adopted surfaces into one app so the good product carries the weak ones. Cut what didn’t land, which is why Copilot Podcasts and Copilot Labs are on the list. Then attach a new paid tier to the one capability enterprises are actually asking about, which is autonomous agents.

The Cowork detail is the tell. Cowork shipped as a standalone agentic tool in March, hit GA in June with its own meter, and by early July was being folded into a merger framed partly as a response to weak paid conversion. A product does not go from GA to absorbed in six weeks because the strategy worked.

I’ve made the structural version of this argument before in Why Your AI Tool Stack Is About to Get Smaller (and More Expensive) and again in Your AI Add-Ons Are Being Cut. Standalone AI add-ons keep collapsing back into platforms, and the price goes up on the way in. Microsoft just ran the cleanest version of that play anyone has run this year.

The Lock-In Decision You’re Actually Making

Consolidation is genuinely good for some buyers. One login across GitHub and Microsoft 365, one governance surface, one admin console, one identity graph. Mitch Ashley of the Futurum Group put it well to DevOps.com: “Microsoft is claiming the agent control plane at the work surface.” If you’re fighting agent sprawl across a dozen tools, a single governed control plane has real value.

The cost is that a merged app has one exit door.

Today you can move GitHub Copilot spend to Cursor without touching your M365 tenant. You can run Claude for research while Copilot handles Office. After the merge, the chat session hands off to the coding agent hands off to the background agent inside one identity graph, and every handoff you use is a seam you can no longer cut cleanly. That’s the actual product. It’s also the actual lock-in.

For enterprises already deep in Microsoft 365 E5 or the E7 Frontier tier I broke down in Microsoft 365 E7: What SMBs Need to Know, the incremental lock-in is small because the lock-in already exists. Take the consolidation. Negotiate the AutoPilot rate as part of the next enterprise agreement, while Microsoft still needs the attach-rate story.

For SMBs and mid-market buyers running a mixed stack, the calculus flips. You’d be trading the portability that’s been your main negotiating position for a governance layer sized for a 5,000-seat org. Wait a quarter. Let enterprise buyers price-discover AutoPilot for you. Nothing about a background scheduling agent gets worse if you adopt it in January instead of September.

The Anti-Hype Read

Three things worth keeping in proportion.

AutoPilot might be cheap. Nobody has seen the number. Microsoft’s own benchmark had Cowork running 30 to 40 percent under Claude Cowork per prompt, and a company with a 4.5% conversion problem has an obvious incentive to price the new tier to drive attach rate rather than margin. Budget for a real line item. Don’t assume it’s a large one.

The merger may slip. Memo said August. Nadella said “this quarter.” No Message Center notice has posted a firm retirement date for Podcasts or Labs. Microsoft has pulled Copilot changes back before. Plan for the cost change, not for the calendar.

Consolidation solves a real problem. The four-tool, four-login, four-audit-log version of Copilot is genuinely worse than one app. The critique here is about the billing structure arriving alongside it, not about the product decision. Both can be true.

What doesn’t get proportioned away is the sequencing. Three pricing changes to one product family in one fiscal year, with the third announced before the second’s promotional period ends, is a pattern. It deepens the Copilot trust deficit that was already showing up in procurement conversations this spring.

My Read

Microsoft is doing the right strategic thing in a way that costs its buyers predictability, and predictability is what enterprise procurement actually buys.

The merged app is the correct product. Chat handing off to a coding agent handing off to a persistent background agent, under one identity and one audit trail, is where this whole category is going. Microsoft getting there first at 450 million seats of distribution is a real advantage.

But every buyer running Copilot has now absorbed three billing model changes since June 1 with no stable baseline in between. The June 1 GitHub AI Credit shift reset dev tool costs. The June 16 Cowork GA added a second meter most admins have not audited. AutoPilot adds a third before the September 1 promo cliff even resets the first one’s real number. A finance team cannot forecast against that, and a finance team that cannot forecast starts saying no to the whole category.

That’s the part I’d push back on if I were in the room. Not the consolidation. The cadence.

The teams that come out of this fine are the ones treating the merger as a procurement event with a known date range instead of a product update that shows up in the Message Center. Inventory the meters. Set the caps. Model Q4 against the post-promo allowance. Put the unpriced AutoPilot line in the budget with a name next to it.

The teams that get hurt are the ones that find out they’re running four meters when the October invoice tells them.

Your Next Step: Pull one page this week showing every active Copilot meter in your tenant: M365 Copilot seats, GitHub AI Credits per developer, and whether Copilot Cowork is switched on. Set tenant and per-user spend limits on the credit-based surfaces before the merged app ships. If Cowork has been running since the June 16 GA and nobody has looked at the credit consumption, that number is your best available preview of what an always-on AutoPilot agent will do to the invoice.

---

Related Reading:

• Copilot’s Flat Rate Dies June 1: What to Do Now
• GitHub Copilot’s New Bill Will Shock Your Dev Team
• Why Your AI Tool Stack Is About to Get Smaller (and More Expensive)
• Microsoft 365 E7: What SMBs Need to Know
• Agent Sprawl Is Killing SMB Productivity
• 90% Don’t Trust AI With Their Data. Copilot Shows Why.
• What Running AI Agents Actually Costs in 2026