Research · Generative & Agentic AI

Copilots & Assistants

Copilots and AI assistants bring generative AI into the tools people already use: email, documents, meetings, code editors, CRM and service desks. Rollout success depends as much on data permissions and training as on the tool.

What they are

Copilots are AI assistants embedded in productivity suites, business applications and development environments. They draft, summarize, answer questions over company content and suggest next steps. Many business software vendors now include assistants in their products, sometimes enabled by default.

Where they deliver value

• Summarizing meetings, threads and long documents.
• Drafting emails, reports and proposals for human review.
• Answering questions over internal knowledge.
• Code completion and explanation for developers.
• Suggesting responses and next steps for sales and service staff.

The permissions problem

Assistants that search company content can surface anything a user technically has access to. Many organizations have years of overshared files and sites. Before broad rollout, review sharing settings and sensitivity labels so the assistant does not expose documents people were never meant to see.

Risks and pitfalls

• Oversharing of sensitive content through AI search.
• Licenses bought for everyone but used by few.
• Users trusting summaries without checking the source.
• Assistants enabled inside vendor products without review.

How to get started

• Clean up file sharing and permissions before enabling enterprise search.
• Pilot with groups whose work fits the tool, and measure usage and outcomes.
• Train users on checking outputs and on data rules.
• Expand licenses based on measured value, not by default.