From the Microsoft edition of October 30, 2026
Microsoft introduces the Copilot harness in Foundry Agent Service (public preview) and GA long-running resilience for hosted agents; harness billing starts Dec. 1 at per-vCPU-hour and per-GiB-hour rates (Oct 28)
Microsoft (Foundry) · Cross-SLED (developers building agents) · Global (commercial Azure)
- Publisher
- Microsoft Foundry Blog
- Original publication
- 2026-10-28
- Source retrieved
- 2026-10-31
- Event date
- 2026-12-01
What happened
Microsoft's Foundry blog (Oct. 28) introduces the Microsoft Copilot harness in Foundry Agent Service in public preview: a managed execution system in which Microsoft runs the agent loop and sandbox while customers supply skills, MCP and Foundry tools (with configured authentication and approvals), files and subagents. It also announces general availability of long-running resilience for hosted agents, which restarts interrupted work with its original input while customer code handles checkpoints and duplicate-action prevention. Harness runtime is free until Dec. 1, 2026; from then sandbox compute and memory are billed at $0.0994 per vCPU-hour and $0.0118 per GiB-hour, plus model inference and tools. Customer quotes are from commercial firms, not the public sector.
Why it matters
Agencies and universities building agents on Azure get a managed agent runtime with a new consumption meter from Dec. 1. Government-cloud availability is not stated, so it pairs with the Azure Government feature page in this edition.
Evidence and measured results
Vendor announcement with feature description and pricing. No public-sector customers or independent performance data.
Limitations and uncertainty
Vendor claim; preview features can change and may lack compliance commitments. No statement of availability in Azure Government. Prices may vary by region and agreement.
Put this evidence to work
Lighthouse Advisory interpretation, grounded in this source. Enriched 2026-10-31; this does not change the original publication date. Each role follows the same checklist so takeaways compare across sources.
Sales
Role takeaway
- Customer problem
- Public-sector developers want agents for multi-step work but need predictable cost and a supported cloud.
- Stakeholders
- Enterprise architect
- Application development lead
- CISO
- Finance/FinOps
- Discovery questions
- Which multi-step tasks are candidates for agents?
- Is your workload in commercial Azure or Azure Government?
- How will agent compute costs be budgeted after Dec. 1?
- Who approves agent tool actions?
- Value hypothesis
- A managed harness may shorten agent build time, but only where cloud availability and cost controls fit.
- Potential engagement
- Agent pilot design with cost model and cloud-availability check.
- Unsupported claims
- Availability in Azure Government
- Public-sector customer results
- Total cost of ownership
Pre-sales engineering
Role takeaway
- Fit
- Teams building agents in commercial Azure; uncertain for Azure Government.
- Architecture and integration
- Declarative agents on Foundry Agent Service with MCP/Foundry tools, sandboxed execution, optional hosted agents.
- Prerequisites
- Foundry project and supported model
- Tool authentication setup
- Cost monitoring
- Constraints
- Public preview
- Switching harness after creation not supported
- Region/cloud availability unconfirmed
- Security
- Review MCP tool auth and approvals, sandbox isolation, and idempotency for system-changing actions.
- Proof of value
- Build one long-running agent with checkpoints and measure cost per run
- Test restart without duplicate actions
Delivery
Role takeaway
- Work
- Pilot agent build, cost tracking and evaluation.
- Dependencies
- Cloud availability
- Budget approval
- Evaluation dataset
- Ownership
- Application team with FinOps and security.
- Skills and adoption
- Agent evaluation and skill authoring.
- Governance checkpoints
- Cost alert set before Dec. 1
- Security review of tools
- Evaluation before production
- Acceptance criteria
- Proposed: cost per task measured
- Proposed: no duplicate actions in restart tests
- Risks
- Unexpected metered costs
- Preview changes
- Agent errors in connected systems
Implementation considerations
Lighthouse Advisory interpretation across the operating dimensions a public-sector buyer must settle before this evidence becomes a design. Each note answers the question under its heading for this specific source.
Architecture and integration
What must connect, and where does the AI sit in the workflow?
Choice between a Microsoft-managed harness and hosted agents with your own framework; long-running work needs application-level checkpointing and idempotency.
Governance
Who approves, reviews and stays accountable for outcomes?
Tool approvals and policy boundaries are configurable, but agent behaviour still needs evaluation and ownership by the customer.
Security and privacy
What data, permissions and controls need testing?
Sandboxed execution and MCP tool connections expand the attack surface; authentication and approval settings must be reviewed.
Accessibility and workforce
Who is affected, and what skills or accommodations follow?
Shifts developer effort from orchestration to skills and evaluation; teams need agent-evaluation skills.
Procurement
What should contracts, pricing and exit terms secure?
A new metered line item (sandbox compute and memory) beginning Dec. 1 on top of model inference.
Operating model
Which teams own the service once it runs?
Agents defined declaratively and source-controlled alongside application code, suggesting DevOps ownership of agents.
Publication history
- 2026-10-30Microsoft · Issue 0213 resources
Stable resource ID: microsoft-foundry-copilot-harness-preview-billing-dec-1-hosted-agents-ga-20261028