Google Unveils Gemini Agent and New Medical AI Findings

By Suad Seferi · 9 October 2026

Google announced a new AI agent designed to carry out tasks across workplace applications on October 8, alongside findings from a clinical study in which its medical AI system interacted with real patients before their appointments. The announcements cover two areas where AI is moving into practical use: everyday business operations and medical consultations. Although the technologies serve different purposes, both raise questions about how much responsibility AI systems should be given and how their work should be supervised. Gemini moves beyond workplace assistance At its Gemini at Work 2026 event, Google Cloud introduced the Gemini agent, a system designed to complete tasks across business applications rather than simply respond to individual requests. According to Google Cloud's announcement, the agent can access organizational information, plan activities, use connected tools and deliver completed work through applications employees already use. The system is designed to operate across Gmail, Google Drive, Docs, Sheets, Slides, Chat and Calendar. It can also connect to external services, including Microsoft 365, Slack, Salesforce and development environments. For example, an employee could ask the agent to research a market, prepare a financial analysis and produce a presentation. Instead of managing each application separately, the agent would coordinate the different steps. Google also introduced the concept of coworker agents. These are AI agents with dedicated organizational identities, including email addresses, calendars and access permissions. Teams can assign them ongoing responsibilities and interact with them through shared documents or messaging channels. The company says organizations will be able to control what information these agents can access and review records of their actions. Security and accountability become central Giving AI systems permission to execute workplace tasks introduces challenges beyond those associated with conventional chatbots. An agent that prepares a document has limited operational consequences. An agent connected to corporate databases, financial systems or internal communications may be able to take actions affecting an organization's operations. Google says its architecture addresses these concerns through separate agent identities, role-based permissions, audit logs and administrative policies. The system can also select different AI models for different tasks, including models from Google's Gemini family and Anthropic's Claude family. This approach is intended to balance performance and operating costs. For businesses considering AI agents, the announcement shifts attention toward practical questions about access permissions, oversight and the reliability of automatically executed tasks. Google tests medical AI with real patients On the same day, Google announced results from a clinical study conducted with Beth Israel Deaconess Medical Center and published in The Lancet. The research involved 98 patients who interacted with Google's Articulate Medical Intelligence Explorer (AMIE), an experimental conversational AI system, before urgent primary care appointments. Unlike earlier evaluations using simulated consultations, the study examined interactions involving actual patients in a clinical environment. Physicians supervised the AI conversations in real time. According to Google's summary of the research, none required interruption under the study's predefined safety criteria. The company reported that AMIE's lists of possible diagnoses included the physicians' final diagnoses in 90% of cases. Clinicians also said the AI-generated summaries helped them prepare for patient visits in 75% of cases and influenced their approach to care in more than half of the consultations. These results suggest that conversational AI may help physicians collect information before appointments and organize relevant details. However, they do not establish that the system can independently diagnose patients or safely replace medical professionals. What the findings do and do not establish The study represents an early investigation of patient-facing AI in a supervised clinical setting. Its relatively small sample, the involvement of supervising physicians and the controlled research conditions limit how widely the findings can be applied. A diagnostic match rate also does not establish that an AI system makes correct clinical decisions in every situation. Medical assessments involve more than generating possible diagnoses, including examination, testing, treatment decisions and follow-up care. Google acknowledges that larger clinical trials are needed before the technology's safety and effectiveness can be assessed at scale. The potential value is therefore more specific than replacing a doctor. AI may be able to help gather patient histories, prepare clinical summaries and support communication between patients and healthcare professionals. What this means for AI adoption The workplace and healthcare announcements show Google developing AI systems for environments where accuracy, permissions and accountability matter. In business settings, organizations will need to decide which tasks agents can complete independently and which actions require human approval. Healthcare presents additional requirements involving patient safety, privacy, clinical validation and professional responsibility. For organizations in the Western Balkans, these developments provide a useful reference point for future AI adoption. The underlying questions are relevant to local businesses, hospitals and public institutions, even where resources and technical infrastructure differ substantially from those of Google's early enterprise customers and clinical research partners. The next tests will take place outside product demonstrations and limited research settings. For Gemini, organizations will need to establish whether autonomous workflows provide measurable benefits under real oper…