Microsoft’s annual developer conference, Build, opened in Seattle on Tuesday, a day after the company unveiled its “Copilot+ PC” laptops. If Monday was about hardware, Tuesday was about putting the Copilot AI assistant into every product it sells. The conference runs until Thursday; these were the announcements from the opening keynote that will actually reach ordinary users.
Team Copilot
Copilot has so far been a personal assistant; Team Copilot makes it a participant. Added to a Teams meeting it keeps the agenda, takes notes visible to everyone, writes the summary and produces a transcript of every speaker — so someone who joins late, or misses the meeting, gets the whole thing without asking. The same assistant works in Teams chats and in Loop documents. It arrives in preview later this year for Microsoft 365 Copilot subscribers.
Edge translates videos as they play
The Edge browser gains real-time AI translation of video: play a lecture on YouTube, LinkedIn or Coursera and Edge produces translated captions — and, later, dubbed audio — live. The launch languages are a short list that includes English, German, Hindi and Italian, with more promised.
Agents that do the job
The word of the year is agent: a Copilot configured, in Copilot Studio, to work like a virtual employee — reading incoming email and acting on it, walking new staff through onboarding, chasing a process from start to finish without a human prompting each step. Microsoft framed it carefully as “under your supervision”, but the direction is clear: the assistant that answers questions is becoming the assistant that finishes tasks.
Smaller things
Teams will let users create custom emoji (with the group admin’s permission). Windows File Explorer will understand coding projects, showing Git status and letting developers search across a repository from the file manager. And the Copilot+ PCs from Monday get their software story: on-device AI for the Recall feature, live captions and image generation, running on the new Snapdragon X chips.
The gap to notice
Hindi is in Edge’s first translation batch; Bangla is not. It is the pattern of every AI launch this year — Bangla, the sixth most-spoken language on Earth, arrives a year or two behind. Until it does, the video translation is a feature for watching English lectures in Hindi, not for the 170 million people who would rather have them in Bangla.
What “agent” means in this announcement
The word was used all week and explained nowhere, so: an agent is a model that has been given tools and a goal instead of a question. A chatbot answers; an agent is told to do something, decides which steps to take, calls software to take them, and reports back.
The difference is not intelligence but permission. An agent that processes supplier invoices needs read access to a mailbox, write access to a ledger and the authority to flag a payment — and the interesting failure is not a wrong answer, it is a wrong action already taken. That is why every serious deployment of this spends its effort on what the agent is allowed to touch rather than on the model, and why a demo is a poor guide to whether it works.
The language gap is a commercial decision
The article notes that Edge’s live translation launched with Hindi and not Bangla, and it is worth being precise about why. It is not model capability — Bangla is the sixth or seventh most-spoken language in the world and there is no technical reason a system that handles Hindi cannot handle it.
It is that enterprise software is sold to enterprises, and Microsoft’s Indian customers buy seats in numbers that justify engineering work in Hindi. Bangla arrives when somebody with a purchase order asks for it. The practical consequence for a Bangladeshi institution is that asking, in writing, through a licensing reseller, is not a futile gesture — it is the only mechanism by which these lists get longer.
Source: The Verge, via Prothom Alo.
