Gemini at Work - Gemini Agent

• Unified agent: One agent handles chat, autonomous task execution, scheduled work, event-driven responses, and code generation, so users do not move between separate assistants for different jobs.
• Persistent execution: The agent runs in the cloud and keeps a single set of memories, context, and a personalization graph across devices and channels. Work that takes hours or days continues after a laptop closes, and no re-briefing is required when switching devices.
• Omnipresent access: Reachable from web, iOS, Android, Windows, and Mac clients, and through command line, Google Workspace, Microsoft 365, and Slack. It operates as a headless agent with no dedicated interface, so it can be embedded in third-party applications and surfaces.
• Multi-agent orchestration: Gemini dynamically creates a roster of temporary, job-specific sub-agents, each with its own task and scope, to divide and complete larger objectives.
• Coworker agents: Team-scoped agents carry their own identity, email address, calendar, Drive presence, and directory entry. Colleagues interact with them as with any coworker — adding them to a Chat space, @mentioning them, or tagging them in a document comment — and the agent acts under its own identity rather than the user's, seeing only what is shared with it.
• Domain skills: Purpose-built capabilities for data work let engineers describe an outcome in plain language and have Gemini generate PySpark code and production pipelines, while a Legal skill automates redaction of personally identifiable information and confidential material before filings go public.
• Model choice: Multi-model orchestration and smart routing select the appropriate model per task — Argon for frontier reasoning, Flash for speed and volume, Omni for generative media, and Gemma for lightweight open-weights edge workloads — to balance quality against cost.

• Audience: Enterprise knowledge workers, developers, data and ML engineers, legal teams, and security administrators deploying agents across an organization.
• Use Case: Automating multi-step operational workflows — credit memo preparation, document quality assurance, fiscal and contractual analysis, fraud and scam detection, customer-support escalation, and software planning. Organizations run deterministic chains of 70 to 80 specialized agents for complex financial and regulatory synthesis, and build no-code agents for knowledge extraction, executive reporting, market research, scheduling, and onboarding.

• Model families: Argon (frontier reasoning), Flash (speed and volume), Omni (generative media), Gemma (open-weights edge workloads).
• Infrastructure: AI Hypercomputer co-designing chips, network, data center, and software; the TPU 8i system delivers 80% better price-performance than the prior generation.
• Execution: Cloud-resident with persistent state; tasks run inside an Agent Sandbox with its own network boundary.
• Governance: Every agent receives a cryptographically attested identity governed with least-privilege permissions, propagated through standards such as OAuth; every action is written to an audit trail attributed to the agent rather than a person; Agent Gateway acts as an AI network firewall enforcing organization-wide policy on all traffic in, out, and between agents.
• Cost controls: Smart routing triages workloads to the lowest-cost capable model; real-time per-project spend caps in the Cloud Billing Console pause an agent's work when hit, with per-department chargeback.
• Access surfaces: Web, iOS, Android, Windows, Mac, command line, Google Workspace, Microsoft 365, Slack, and headless embedding.
• Deployment: Gemini Enterprise (formerly Gemini for Google Workspace), available through Google Cloud.

Gemini Spark was Gemini's first public step toward an autonomous agent — an early, consumer-facing attempt to go beyond a conversational assistant and hand the model a goal to pursue on its own. It proved the concept: an agent that could plan and act with limited supervision. Gemini at Work, built around the Gemini agent, is that same ambition delivered for the business use case — where "acting on its own" has to happen inside a governed, auditable environment with real identity, permissions, and cost controls. What Spark demonstrated in principle, Gemini at Work operationalizes at enterprise scale, turning the autonomous-agent promise into something a company can actually deploy across teams, applications, and systems-of-record.