01
Microsoft starts unifying Copilot into one app
Microsoft began rolling out a unified Microsoft Copilot app that merges consumer Copilot with the Microsoft 365 Copilot experience. Microsoft will retire several consumer features on 2026-08-18 as part of the consolidation.
- Update internal training and comms for users who currently distinguish between “Copilot” and “Microsoft 365 Copilot,” because Microsoft is converging the entry point and branding.
- Review which Copilot features your organization relies on and whether users depend on consumer-only capabilities, because Microsoft is removing Podcasts, Group Chat, and free Deep Research on 2026-08-18.
- Validate where Copilot artifacts and outputs land and how they are governed, because the consolidation includes changes to how content is organized across Microsoft services.
02
Microsoft plans major Maia 300 chip production boost
Microsoft discussed securing foundry capacity to significantly increase production of its Maia 300 AI chips, with deliveries reported for 2027. The plan would expand Microsoft-controlled accelerator supply for Azure AI infrastructure.
- Track how Microsoft-priced AI services evolve on Azure, because more in-house accelerator capacity can affect availability and unit economics versus third-party GPU supply.
- Ask Microsoft account teams which Azure services will run on Maia hardware and what performance isolation you can expect, because platform-level changes can affect latency and predictability for enterprise workloads.
- Plan procurement and architecture with a multi-year horizon, because Microsoft’s reported commitments suggest sustained AI capacity build-out through 2027 rather than short-term pilots.
Source — Reuters AzureAI infrastructure 03
GitHub Copilot adds Grok 4.6 model option
GitHub started rolling out Grok 4.6 as a selectable model in GitHub Copilot across major IDEs and Copilot plans. Enterprise admins can control availability via policy.
- Run a controlled evaluation across your approved repos and languages, because model choice can materially change code quality and review workload for enterprise teams.
- Align editor support with your developer fleet, because the rollout targets multiple IDEs used in mixed Windows and cross-platform Czech enterprise environments.
- Update governance and allowed-model policies, because enabling a new model option changes compliance posture and can affect audit evidence for AI-assisted development.
04
GitHub ships Agent Plugins 1.0 for Copilot agents
GitHub made Agent Plugins 1.0 generally available across VS Code, Copilot CLI, the GitHub Copilot SDK, and the Copilot app. GitHub positions plugins as reusable extensions across compatible agent clients.
- Standardize internal tooling by building one plugin for multiple Copilot surfaces, because reuse reduces maintenance overhead for platform engineering teams.
- Define an approval and distribution workflow for plugins, because agent extensions can introduce data egress and access risks similar to other integrations.
- Prioritize a small set of high-value enterprise plugins (ticketing, knowledge base, CI/CD actions), because agent portability can accelerate adoption once governance is in place.
05
GitHub adds per-model token breakdown to Copilot reports
GitHub introduced per-model token reporting that breaks down input, output, cache read, and cache write tokens for Copilot usage. The change improves visibility for both admins and individual users.
- Set up chargeback or showback by team and model, because per-model token accounting supports more accurate allocation of AI operating costs.
- Use the breakdown to compare models on real workload profiles, because cache behavior and output volume can drive materially different bills.
- Add usage thresholds and alerts to governance processes, because better reporting enables proactive cost control before expanding rollout.