That is not a niche pricing footnote. It is a FinOps problem for growing teams that already live in Word, Excel, Teams and Outlook, and who are about to discover that "turning on the clever bits" can move cost from headcount planning into a consumption meter.
What changed this month (Home / Code / Autopilot in plain English)
On 25 September 2026, Microsoft's Jared Spataro announced a rebuilt Copilot app with three capabilities rolled into one experience:
- Home — the new starting point. Chat (quick questions, drafts, lookups) and Cowork (work you delegate end-to-end — an RFP pack, a launch kit, a customer briefing) sit together, with Office in Copilot so Word, Excel and PowerPoint are built into the same surface. Real editable files, not throwaway chat paste.
- Code — natural-language apps, dashboards, trackers and automations for knowledge workers, powered by the same underlying technology as GitHub Copilot. Results can run in a sandboxed Copilot Managed Runtime hosted inside your Microsoft 365 tenant.
- Autopilot — previously Scout. A persistent cloud agent with its own identity, memory and workspace. It keeps working when nobody is at the keyboard: watching channels, chasing threads, running recurring processes such as a supplier review.
Microsoft says Home and Code are rolling out via the Frontier programme; Autopilot expands to private preview at the end of September. Under the hood, Fabric IQ grounds Chat and Cowork in Power BI semantic models (generally available), with Dynamics 365 and Power Platform grounding rolling out — so Copilot can pull deal history, tickets and workflow context without a separate connector project for every question. National Technology's Rory Bathgate notes the same UK trade take: one unified app, FinOps for agentic spend, Auto model routing, and grounding without bolting on separate pipelines for every data source.
The bill splits in two (USL vs Copilot Credits)
Microsoft's own framing is the one finance and IT should pin on the wall:
| Layer | What it covers | How you pay |
|---|---|---|
| User subscription licence (USL) | Everyday Chat + Copilot in Word, Excel, PowerPoint, Outlook and Teams; Auto routes models by accuracy, speed and cost | Fixed per-user seat |
| Usage-based billing (UBB) / Copilot Credits | Cowork, Code, Autopilot, long-running agents, frontier models (Microsoft cites examples such as Astra and Fable) | Consumption — credits |
TechMarketView's Simon Baxter put the UK angle bluntly: the seat licence stays for everyday use; the agentic tier sits on Copilot Credits. Partner commentary from Universal.cloud (28 September 2026) reports that for Copilot Business, the variable layer is on by default from 2 November 2026 for new CSP licences — treat that as partner reporting and confirm with Microsoft or your reseller before you lock a budget line.
Microsoft has not, in the Spataro announcement, published a simple public rate card for every capability. Third-party blogs sometimes quote US list cents-per-credit figures from mid-2026 commentary — treat those as indicative only and confirm live pricing with Microsoft, Microsoft Learn's usage-based billing docs, and your reseller.
What is clear: FinOps tooling is expanding. Cost management in Agent 365 is growing beyond Cowork and Work IQ APIs to cover Code and Copilot Managed Runtime, with Copilot Studio agents planned for October. Admins get spending policies, model-family controls (which also constrain what Auto may choose), credit-request routing into approval workflows, and end-user visibility of credit balances inside Copilot.
Why always-on agents change the cost curve
A seat licence is predictable: N users × licence. Agentic work is not. TechMarketView's point matters for UK growing firms as much as for the NHS England example they cite: Code and Autopilot drive spend because they keep running without a person at the keyboard. An agent watching channels overnight or looping on a supplier process is no longer tied to someone in a meeting.
That is excellent for throughput — and dangerous for a budget built only on headcount allocations. Public-sector buyers feel this first; private SMEs hit the same curve the month a team "tries Autopilot on everything" without caps. TechMarketView notes SIs may help with outcome framing or consumption management; for most growing firms the practical translation is simpler: treat Copilot Credits like cloud spend — owner, alerts, monthly review, kill-switch.
A FinOps checklist before you open the floodgates
Steal this before you invite the whole company into Frontier features:
- Map who is already on a Copilot USL — and who actually needs Cowork, Code or Autopilot. Everyday Chat/Office is not the same product as an always-on agent.
- Turn on cost management first. Spending policies, credit approvals and model-family limits — configure monthly and per-group caps before broad rollout (Universal.cloud's checklist lands in the same place).
- Decide who gets expensive models. Frontier models and long-running agents are where UBB bites. Route access by role, not curiosity.
- Govern Autopilot like a digital hire. Own identity, permissions, Purview, audit. An agent that can @mention colleagues needs the same scrutiny as a contractor with a laptop.
- Ground value in real data. Fabric IQ / Dynamics / Power Platform grounding only pays if semantic models and CRM data are trustworthy.
- Pilot one workflow with a kill-switch. Measure hours saved and credits burned for two weeks before you scale.
- Brief finance on variable AI lines. If the FY budget assumes "Copilot = seats," update it before November CSP defaults (where applicable) and before Autopilot leaves private preview.
- Confirm pricing with Microsoft / your reseller. Do not hard-code third-party dollar-per-credit figures into a board paper — use Microsoft Learn and Cost Management in the M365 admin centre for prepaid vs pay-as-you-go mechanics.
Why this matters for Clarity Growth
Clarity Growth helps UK SMEs and scale-ups select tools, govern cost, and connect AI into real operations — websites, ecommerce, CRM, finance and the messy middle where work actually happens. Microsoft's Copilot rebuild is a useful case study: keep the fixed seat for everyday productivity, treat Credits as a governed consumption layer, and only wire Code or Autopilot into processes where data quality, permissions and a named owner already exist. The win is not "more AI seats." It is M365 AI that shows up in ops without surprising the CFO.
Sources
- Introducing the new Copilot with Home, Code and Autopilot — Microsoft Blog (Jared Spataro) (25 September 2026)
- Microsoft shifts Copilot's agentic tier to usage-based billing — TechMarketView (Simon Baxter) (28 September 2026)
- Microsoft unveils new unified Copilot app — National Technology (Rory Bathgate) (28 September 2026)
- The new Copilot: Home, Code and Autopilot in one app — Universal.cloud (28 September 2026)
- Usage-based billing overview / Copilot Credits — Microsoft Learn (admin mechanics; confirm live rates with Microsoft)