Microsoft gives Copilot eyes, a voice, and deeper thinking

For the past year, Microsoft Copilot has felt less like a companion and more like an eager corporate intern glued to your Word doc. That is finally changing.

Under the direction of Mustafa Suleyman, the DeepMind cofounder who took over as CEO of Microsoft AI earlier this year, Copilot just received its biggest makeover yet. The stark prompt box is gone, replaced by a warm, card-based interface heavily inspired by Pi, the empathetic chatbot Suleyman built at Inflection AI. The goal is simple: make AI feel less like software and more like a supportive daily partner.

A voice that listens, a brain that pauses

The most noticeable immediate upgrade is Copilot Voice. You can now choose between four voices and talk naturally to the assistant. Just like having a conversation with a human, you can interrupt it at any point to redirect the brainstorming or simply vent after a long workday.

Inside a new experimental playground called Copilot Labs, Microsoft is introducing two far more ambitious tools: Copilot Vision and Think Deeper.

Vision essentially lets Copilot look over your shoulder at what you are browsing. It analyzes both text and images on your active webpage in real time. If you are furniture shopping, for example, it can study a room photo, recommend rugs, and debate color palettes with you as you click through items.

Think Deeper takes the opposite approach of quick chatbots. Likely powered by OpenAI's o1 reasoning model, it pauses before answering, spending extra seconds to work through step-by-step logic. It is built for genuine dilemmas, like weighing the pros and cons of moving to a new city or comparing car models side by side.

Paid morning briefs and guardrails on your screen

There is also Copilot Daily, a spoken morning brief covering news and weather read aloud by your preferred Copilot voice. Instead of scraping the web indiscriminately, Microsoft has signed paid licensing deals with major publishers including Reuters, The Financial Times, Axel Springer, and Hearst. With media companies bleeding jobs and Microsoft already facing lawsuits over training data, paying publishers directly is a calculated move to secure reliable sources.

Microsoft is also walking on eggshells with Copilot Vision. To avoid another privacy backlash, the company made sessions entirely opt-in and ephemeral. The moment you close the session, the data is wiped completely—nothing is stored or used for model training. During this initial preview, Vision will only work on a whitelist of popular websites and flatly refuses to interact with paywalled or sensitive pages.

Voice is launching initially in English across the US, UK, Canada, Australia, and New Zealand, while Daily kicks off in the US and UK. Vision and Think Deeper are rolling out first to select Copilot Pro subscribers in English-speaking markets.

Why it matters

This redesign marks a pivotal shift for Microsoft. For months, Copilot was merely repackaging OpenAI's raw technical breakthroughs into enterprise software. By bringing in Suleyman and the Inflection team, Microsoft is finally investing in product personality, trying to turn an administrative utility into an indispensable daily interface.

For users, Vision is easily the most compelling feature here. An assistant that can actually see the messy, multimodal web with you has far more utility than an isolated chat window. But its strict limitations—whitelisted domains and no paywalled content—mean it might feel crippled in day-to-day use until Microsoft loosens the reins.

For the broader industry, the battleground has clearly moved from model benchmarks to sensory interfaces. Microsoft is preempting Google's Gemini Live and Apple Intelligence by planting a voice-and-vision assistant directly on desktops and mobile devices. Whether people actually want an AI peering at their browser tabs is another question entirely, but Microsoft is betting that a calmer, friendlier tone will be enough to earn that trust.