Copilot adoption, Build 2026 aftermath, super app controversy, in-house models, agent push
Key Questions
What new AI models and Copilot features were announced at Build 2026?
Microsoft unveiled MAI-Thinking-1, MAI-Code-1, Copilot Agent Mode, and Windows agent OS along with eight new models. Additional tools like Agent Optimizer, ASSERT, and RAMPART were introduced to support enterprise governance and agent deployment.
How is enterprise adoption of Copilot progressing according to recent data?
Paid seats reached 25M with major wins including NHS (505k seats) and KPMG (276k), yet overall adoption sits at 3.3% with 80% of pilots failing. Reports indicate penetration around 5% and E5 usage below 20% among some cohorts.
What steps is Microsoft taking to reduce dependence on OpenAI for Copilot?
The company is overhauling Copilot's design interface using its in-house MAI-Image-2.5 model and evaluating DeepSeek for Copilot Cowork. It has also restricted internal use of Claude Fable 5 due to data retention issues.
What changes were made to Copilot pricing and model support?
Copilot Cowork GA introduced consumption-based pricing that is 30-40% cheaper per prompt with multi-model options including Anthropic, GPT-5.5, and in-house models. Usage is managed via Copilot Credits under PAYG and P3 tiers.
What governance and security issues affect Copilot deployments?
Two outages occurred in June, critical vulnerabilities like SearchLeak and EchoLeak were patched, and new tools address agent sprawl. ShareGate MCP now integrates M365 governance into Claude, ChatGPT, and Copilot.
Build 2026 unveiled MAI-Thinking-1, MAI-Code-1, Copilot Agent Mode, Windows agent OS, security tools. Copilot adoption mixed: 25M paid seats (up from 20M) but 3.3% adoption, 80% pilot failure. Major enterprise wins (NHS 505k, KPMG 276k, Lloyds 97% active, Atos 56k, Marlabs HR agents) counter narrative. New tools: Agent Optimizer, ASSERT, RAMPART address enterprise governance. Microsoft restricts internal use of Claude Fable 5 over data retention concerns. Two Copilot outages in June, including cascading authentication failure; no SLA for Copilot yet; Fortune 500 firms pausing rollouts. AvePoint CEO reports Copilot penetration at 5%, E5 below 20%. Microsoft released 8 new models at developer conference. Microsoft quietly overhauling Copilot's design interface with its own MAI-Image-2.5 model to reduce OpenAI reliance; mixed reviews. Unverified rumor that Microsoft is rolling back Copilot AI features from built-in apps. Another critical security vulnerability (SearchLeak) in M365 Copilot discovered and remediated. Microsoft patched another critical Copilot vulnerability (EchoLeak, 9.3 CVSS, zero-click). Nadella warns AI could absorb corporate knowledge. Brad Smith reassures on AI jobs timeline (25 years, not 2.5). Microsoft weighs DeepSeek for Copilot Cowork to manage costs—decision within weeks; cost analysis shows 30x chatbot cost. Copilot Cowork GA worldwide is a major milestone with consumption-based pricing (30-40% cheaper per prompt), multi-model support (Anthropic, GPT-5.5, in-house Cowork 1), and Fortune 500 adoption. Model choice and usage-based billing via Copilot Credits with PAYG and P3 tiers. An article shifts narrative from demand proof to ROI proof. Microsoft confirmed shift to usage-based Copilot pricing, with executive admission that power users drive costs too high. Microsoft launched Microsoft IQ, a new enterprise data unification layer for AI agents and Copilot. Microsoft Agent 365 launched as governance platform for AI agent sprawl. Build 2026 technical announcements on Fabric and Power BI Agent Skills are incremental. New securities fraud lawsuits related to Copilot disclosures add legal overhang. ShareGate MCP integrates M365 governance into Claude, ChatGPT, and Copilot—addressing governance sprawl from security incidents.