Google adds agentic AI to Gemini, shifts enterprise focus

Google turned Gemini into an AI‑driven agent that can plan, execute, and hand off work across business applications. The move marks the company’s first push to embed autonomous AI directly into enterprise workflows.

TechCrunch reports that the new Gemini agent can delegate tasks to sub‑agents, call multiple large‑language models, and even operate under its own workplace identity, complete with a dedicated email address. The feature launches today for business customers and is positioned as a way to reduce manual coordination across tools like calendars, document editors, and CRM platforms.

How the agent works across business tools

The Gemini agent follows a simple command‑and‑control model. A user issues a high‑level request—“schedule a product demo with the sales team and update the forecast”—and the agent breaks the request into discrete steps. Each step can be handled by a sub‑agent specialized for a particular domain, such as calendar management or spreadsheet updates. The system can switch between models on the fly, selecting the one best suited for reasoning, data extraction, or natural‑language generation.

A workplace identity gives the agent a persistent presence in an organization’s email system. The email address allows the agent to receive and send messages, file attachments, and maintain a thread history that other users can reference. According to the launch details, this identity is scoped to the organization’s domain, so the agent appears as an internal collaborator rather than an external bot.

The architecture also supports cross‑app orchestration. When a task requires data from a CRM, the agent can query the CRM API, transform the result into a spreadsheet row, and then trigger a follow‑up email—all without human intervention. The design leans on existing Google Cloud security primitives, ensuring that token handling and data access respect enterprise policies.

Industry context: AI slowdown and competition

Sundar Pichai’s recent remarks at the New York Times DealBook Summit echo a broader industry mood. He warned that “the low‑hanging fruit is gone” and that incremental improvements will dominate until a deeper breakthrough emerges. Pichai’s comment aligns with a growing consensus that generative AI’s rapid early gains are plateauing.

Competing firms are already packaging similar agentic capabilities. Meta’s Llama series and OpenAI’s ChatGPT plugins both enable multi‑step workflows, but Google’s integration with its own suite of productivity tools gives Gemini a distinct advantage in data locality and latency. The same summit also featured Microsoft’s Satya Nadella noting that AI growth will be “non‑linear,” a reminder that market dynamics could shift with a single technical advance.

Google’s internal rollout of AI Mode in Search illustrates the company’s broader strategy to embed generative AI across user‑facing products. AI Mode, described in a recent HN post, lets users ask complex, multi‑part questions and receive structured, AI‑generated answers. While AI Mode targets consumers, Gemini’s agent framework targets enterprises, suggesting a coordinated push to make AI the default interface for both search and work.

Enterprise implications and open questions

The Gemini agent promises to automate routine coordination, but it also raises questions about control, cost, and reliability. Enterprises will need to decide how much autonomy to grant the agent—whether it can schedule meetings without explicit approval or modify financial models on its own. The ability to delegate to sub‑agents adds flexibility but also expands the attack surface for misconfiguration.

Google has not disclosed pricing for the Gemini agent or the underlying compute resources. Early adopters will likely compare the offering against third‑party solutions like the open‑source AISheeter extension that lets users run LLMs directly in Google Sheets. AISheeter’s model‑agnostic design and transparent reasoning may appeal to cost‑sensitive teams, but it lacks the deep integration that Gemini’s native email identity provides.

Another practical concern is data residency. Companies handling regulated data will scrutinize where the agent stores intermediate results and whether it complies with industry standards such as ISO 27001 or GDPR. Google’s claim that the agent respects existing Cloud security controls is reassuring, yet enterprises will demand independent audits before granting it broad permissions.

Finally, the rollout coincides with Google’s decision to end free cellular access to Safety Signal on older Pixel Watch LTE models. While unrelated to Gemini, the move underscores Google’s willingness to monetize premium features after an initial free period. Enterprises may wonder whether similar tiered access could appear for Gemini’s advanced capabilities.

What to watch

In the coming months, watch for the first wave of enterprise pilots that test Gemini’s agentic workflow at scale. Key metrics will include task completion rates, error recovery times, and the frequency of human overrides. Google’s next public update—expected in early Q1 2025—should reveal pricing tiers and any limits on sub‑agent usage. Competitors’ responses, especially any new plugin ecosystems from OpenAI or Meta, will also shape whether Gemini becomes a de‑facto standard for AI‑driven business automation.