01
Microsoft redesigns Copilot around Home, Code and Autopilot
Microsoft announced a redesigned Copilot that integrates Office work, natural-language app creation and a persistent personal agent. It also introduced Copilot Managed Runtime in preview for applications created through Cowork, Code and Copilot Studio.
- Plan limited pilots rather than a general deployment because Home and Code will start through Frontier and Autopilot is scheduled for private preview at September’s end.
- Define directory permissions, delegated access and human-approval rules before allowing Autopilot to act across enterprise applications.
- Verify Managed Runtime’s regional availability, processing locations and logging before placing Czech business data or production applications on it.
02
Copilot pricing combines subscriptions with usage-based credits
Microsoft described a Copilot pricing model that combines user subscription licences with usage-based Copilot Credits for workloads requiring higher quality or specific models.
- Build total-cost estimates from both licence counts and expected credit consumption rather than treating the subscription as the entire budget.
- Agree departmental spending limits and chargeback rules with finance before expanding access to metered workloads.
- Include model-access decisions in procurement reviews because the choice of model can affect operating costs.
03
Microsoft expands cost management for AI workloads
Microsoft expanded Agent 365 cost management to cover Code and Copilot Managed Runtime. It plans to add support for Copilot Studio agents in October.
- Include Code and Managed Runtime consumption in IT and finance reviews alongside Microsoft 365 subscription spending.
- Assign a budget owner to each departmental AI application before usage-based workloads expand.
- Treat October support for Copilot Studio as a roadmap item rather than an available control when approving agent deployments this month.
04
GitHub sets October date for global Copilot feature policy
GitHub announced a global default policy for generally available Copilot features and supported clients. The policy takes effect on 22 October 2026.
- Choose whether eligible features should default to enabled, disabled or organization-controlled before the October effective date.
- Use a disabled default where security teams must assess new capabilities before developers receive access.
- Document which decisions belong to central IT and which Czech subsidiaries or business units may make independently.
05
GitHub Copilot app adds OpenTelemetry monitoring
GitHub added enterprise-managed OpenTelemetry support to the Copilot app for exporting agent activity to compatible monitoring systems. Prompt and response content stays excluded by default.
- Assess whether existing observability platforms can ingest model-request and tool-usage traces before buying separate agent-monitoring tooling.
- Define incident-response procedures that use agent traces to investigate unexpected actions.
- Review telemetry fields, access rights and retention periods with security and privacy teams even when prompt content remains excluded.
06
GitHub expands automatic Copilot code-review settings
GitHub expanded personal settings and enterprise defaults for Copilot code review. Automatic review triggers now include pull-request creation, co-authorship, status changes and new pushes.
- Set enterprise review defaults to reduce inconsistent adoption across development teams.
- Pilot automatic reviews on selected repositories and measure useful findings against review noise before broad rollout.
- Keep required human approvals and existing security checks in place because automatic AI review does not replace release accountability.
07
GitHub broadens Copilot models and remote development support
GitHub’s weekly release roundup lists Claude Opus 5.5, GPT-6 Sol, GPT-6 Luna and Grok 4.7 across Copilot surfaces, with availability varying by model and plan. It also describes a gradual rollout of VS Code agent support for Dev Containers on SSH, Tunnel and WSL hosts.
- Test model quality, latency and consumption on representative Czech enterprise codebases before changing approved model policies.
- Check model availability against the company’s Copilot plan and developer clients rather than assuming identical access everywhere.
- Validate remote-container agents against managed development environments and do not equate remote execution with a guarantee that model processing stays on the host.
08
Microsoft Research announces RetroChimera for molecule synthesis
Microsoft Research and collaborators including GSK and Novartis announced RetroChimera, an AI model aimed at challenges in molecule synthesis and pharmaceutical research.
- Czech pharmaceutical and chemistry teams should assess the research against their synthesis-planning workflows rather than general office-productivity benchmarks.
- Require validation on relevant molecules and laboratory constraints before assigning operational value to the model.
- Keep commercial deployment assumptions out of near-term budgets because the announcement concerns research rather than an immediately deployable enterprise product.
09
Microsoft reports global generative-AI usage at 18.8 percent
Microsoft’s updated AI Diffusion Report estimates that 18.8% of the global working-age population used generative-AI tools in the second quarter of 2026, up from the previous quarter.
- Use the figure as a population-level adoption reference rather than a benchmark for paid enterprise Copilot usage.
- Check the report’s methodology and country coverage before using it in Czech board presentations or investment cases.
- Base workforce-enablement budgets on internal usage and task outcomes because global adoption does not establish local productivity gains.