Copilot becomes a superapp, Gemini gains face, and AI giants self-regulate

The race for artificial intelligence is no longer measured solely by benchmarks. This September 25, 2026, marks three developments that paint the same picture: assistants are becoming work environments, gaining a face and a phone, and at the same time, their creators are trying to establish regulations before governments do. Microsoft is unveiling Copilot, which aims to be as central as Office. Google is giving Gemini a face—and a voice in 97 languages—and letting it make calls to businesses on your behalf. And OpenAI, Anthropic, and Google are making progress on an industry security authority called SAFA.

Microsoft turns Copilot into the "super app" of work

Microsoft officially unveiled today the redesigned Copilot app, which had been leaking for weeks. The app is now structured around three pillars: Home, Code, and Autopilot. Jared Spataro, head of AI in the workplace, summed it up with a phrase clearly intended for presentations: “Just as Office defined work in the PC era, the new Copilot is designed to define it in the AI ​​era.” Satya Nadella went even further at Microsoft X, describing Copilot as “a new operating system for work.”

Home combines Chat and Cowork. Chat remains for quick conversations; Cowork is for end-to-end delegated tasks—an RFP response, a launch kit, a financial close—that return as iterable-ready results. Office also arrives within Copilot: embedded Word, Excel, and PowerPoint, with real, editable documents, not screenshots or corrupted copies. The Today feature, entering private preview in October, promises a command center that not only lists emails and meetings but also generates drafts and prepares follow-ups.

Code is the real surprise. Anyone can describe a dashboard, tracker, or automation in natural language, and Copilot builds it in an isolated environment, using the same underlying technology as GitHub Copilot, and hosts it in the tenant with Managed Runtime. It doesn't replace developers; it gives them another channel to the platform while the rest of the team moves away from relying solely on files.

Autopilot—formerly Scout—is the persistent agent. It has its own identity, memory, computer, and workspace within the tenant, understands the business via Microsoft IQ, and appears in Teams, Outlook, and Docs as a colleague mentioned with @. It can monitor channels, manage a vendor review process, or pick up a project days later, even while you sleep. Home and Code will begin rolling out to the Frontier program in the coming weeks; Autopilot enters private preview at the end of the month.

The price also changes. The per-user subscription license covers "everyday" use (chat and Office). Cowork, Code, Autopilot, and front-end models like Astra and Fable will be billed on a pay-as-you-go basis, with FinOps for AI to ensure IT isn't caught off guard. The Verge sums up the tone: Microsoft is no longer competing for consumers against ChatGPT or Muse and is positioning itself as "AI made for work." Sources: Microsoft official blog, The Verge, and Reuters (September 25, 2026).

Google puts a face to Gemini and lets her make calls for you

While Microsoft is reinforcing the office, Google is focusing on presence. Gemini 3.8 Live with Live Avatar, available since September 24th in Gemini Enterprise, combines live conversation with low-latency video. The avatar listens, sees, and responds with lip-syncing and coherent expressions. It can switch between 97 languages ​​within the same conversation without breaking lip-sync, and makes asynchronous calls to background tools—in a demo, it assists with hotel check-in while continuing the conversation.

Companies choose from pre-designed avatars or create one from a reference image (currently restricted to authorized accounts). Google watermarks audio and video with the SynthID. There is no release date yet for the Gemini app for individual users, nor a specific public price for Live Avatar: it's an enterprise tool, not a mobile filter.

Meanwhile, TechCrunch details "Call for Me": Gemini calls businesses on behalf of the user. It's launching as an experiment for Pixel 11 owners with a Gemini subscription in the US, using the Phone app beta. It calls using the user's personal number, can share data authorized by the user, navigates automated menus, holds calls on hold, and allows users to follow the call with live transcription and take control at any time. Typical uses include checking store stock, making a restaurant reservation, rescheduling an appointment, or leaving something on hold. It's an evolution of Ask for Me, Hold for Me, and Direct My Call, now with more complex conversations.

Google DeepMind's new head, Koray Kavukcuoglu, told The Information that Gemini 4 is in its final stages of refinement and will arrive "well before" the end of 2026, with a strategy of releasing early and iterating. Sources: El Nacional, TechCrunch, and Europa Press.

SAFA: OpenAI, Anthropic and Google want to set the rules

The same group of labs competing for frontier AI models is, according to The Information, setting up an industry standards authority: the Standards Authority for Frontier AI (SAFA). The Verge and Bloomberg Law place its launch between late 2026 and early 2027. The body would be independent of the government and would be responsible for supporting third-party testing before deployment, defining how to report security incidents, and establishing risk assessment baselines.

The announcement aligns with the message that Sam Altman (OpenAI), Dario Amodei (Anthropic), and Clément Delangue (Hugging Face) delivered to the UN this week: common standards for measuring capabilities, safeguards, human oversight, and rapid incident reporting protocols. Amodei emphasized that, if mismanaged, AI “could be a risk to humanity as a whole.” OpenAI stated that international standards “can be as important to the pace of the frontier as alignment research.”

The obvious question is whether an authority created by the very players in the race acts as a brake or a shield. Companies will still need their own controls over business model risk, data exposure, and agent behavior. SAFA doesn't replace state regulation; it attempts to anticipate it with a shared language for evidence and incident reporting. Sources: The Verge, Bloomberg Law, BBC, and CIO.

What does this mean for those who use AI today?

Three threads, one pattern. First, the assistant is no longer just a chat box: Microsoft is transforming it into a workspace with documents, lightweight code, and agents that never shut down. Second, the interface is becoming more human: Google is selling visual presence and real calls, not just text. Third, governance is becoming industrialized: the same labs that launch cutting-edge models want to define how failures are measured and reported.

For teams in Mexico and Latin America, the practical message is clear. If your company uses Microsoft 365, it's worth looking at Frontier, FinOps, and which tasks will be billed on a pay-as-you-go basis before indiscriminately deploying Autopilot. If you serve customers, Live Avatar and Call for Me foreshadow a world where the first point of contact might be a face-to-face agent or a phone call—with a SynthID watermark and the reputational risk of sounding "too human." And if you build or purchase frontier models, SAFA doesn't exempt you from internal auditing: at best, it provides a common vocabulary for when something goes wrong.

The AI ​​of September 25th isn't just more powerful. It's more pervasive, more expensive to operate when running autonomously, and more aware that the world demands rules. Anyone who falls asleep in the chat in 2023 will wake up to a digital companion in Teams, an avatar at reception, and an incident form that someone—government or SAFA—will want filled out.

Sources

• Microsoft Blog, Sep 25, 2026 — Introducing the new Copilot with Home, Code and Autopilot
• The Verge — Microsoft Copilot super app; SAFA
• Reuters — Microsoft revamps Copilot
• The National — Gemini Live Avatar
• TechCrunch — Gemini Call for Me
• Europa Press — Gemini 4
• Bloomberg Law / The Information — SAFA
• BBC — OpenAI and Anthropic at the UN
• CIO — AI frontier governance