Google's Gemini Agent Has No Price. Its Gems Deadline Does: March 2027

Google announced a universal work agent on October 8. The more urgent news is the deadline it carries.

At its Gemini at Work 2026 event on October 8, 2026, Google Cloud introduced the Gemini agent — described by CEO Thomas Kurian as "a single, universal agent for work." The line the company built the announcement around is blunt: work now starts in the prompt window. You hand over an objective rather than instructions, and you come back to finished work.

The same week, Google locked in the schedule that replaces Gems with skills. For any team that actually uses Gems in production, that second item is the one with a hard date attached.

What the Gemini agent actually is

Google lists six architectural principles, and they are worth reading as a spec rather than as marketing.

Unified agent — chat, autonomous work on assigned objectives, and code generation through one interface
Omnipresent access — web, iOS and Android, Windows and Mac desktops, command line, Google Workspace, Microsoft 365, and Slack. It can be embedded in third-party apps and run as a headless agent with no interface of its own
Persistent execution — the work lives in the cloud. Tasks lasting hours or days keep running after you close your laptop, and a single set of memories, context, and one personalization graph follows you across devices
Multi-agent orchestration — the agent can spin up a roster of temporary, job-specific sub-agents, each with their own identity. Coworker agents are the persistent version: a team member with its own @agents.company.com email address, its own storage, and a place in the company directory. They see only what has been shared with them
Deeply contextual — it arrives knowing your tools, data, and work history, and keeps learning
Model choice flexibility — Gemini is the agent; the model underneath is a separate decision. Google says it orchestrates across its own model family and Claude models from Anthropic today, with other private and open models to follow

Google's stated rationale for the last point is economic rather than technical: the best model for a job is not always the largest one. Matching model to task raises accuracy on hard work and lowers cost on simple work.

Pricing and availability: no price, no GA date

This is where the announcement gets thin, and it matters.

There is no published price for the Gemini agent.
There is no general availability date. Google's wording is "soon."
SMBs are in early access now, according to Sharon Prosser, VP of SMB and Scaled at Google Cloud
Financial Services and Legal packages are in preview; Government, Healthcare, and Retail are listed as coming soon

The financial services package carries more than 50 foundational skills and draws on FactSet, LSEG, S&P Global, and SEC filings, displaying confidence scores, methodologies, data lineage, and source citations. CME Group and Deutsche Bank are named users. The legal package inherits matter-level permissions and ethical walls from NetDocuments and iManage.

On the infrastructure side, Google says its TPU 8i system delivers 80% better price-performance than the prior generation, and it presents four model families: Argon for frontier reasoning, Flash for speed and volume, Omni for generative media, and Gemma for lightweight open-weights edge work.

Cost control: spend caps that pause the agent

Kurian reduced the enterprise agent question to two clauses. What decides whether an agent program "succeeds or stalls" is whether it can be governed and whether it can be afforded.

The number Google uses to frame the problem is striking: per-token prices have fallen 98% since 2024, according to Google Cloud, and enterprise AI volume has still "exploded." Kurian's phrasing: "Running every simple loop through a premium model quickly breaks corporate budgets."

Three controls follow.

Multi-model orchestration picks the model per job

Smart Routing triages workloads to the model delivering the best performance at the lowest cost

Real-time spend caps set a hard limit on a project's AI spend in the Cloud Billing Console. Gemini monitors token usage and sandbox costs; when a cap triggers, that project's agent pauses, and work resumes with a single click

Because tracking is per project, companies can charge AI costs back to specific departments. That is the detail finance teams will care about most.

There is precedent inside Google. Project Spend Caps arrived in the Gemini API on March 16, 2026 — a monthly dollar limit per project that took about ten minutes to take effect, after developers reported billing errors in August 2025. Notably, neither Google document states an enforcement delay or any dollar thresholds for the new Cloud Billing version. Treat those as open questions.

The pressure behind this is measurable. A KPMG survey of 2,145 senior leaders found that 49% had narrowed, delayed, or paused AI agent deployments, with operating costs outrunning value as the main driver. Only 35% said their AI operating costs were fully visible and actively monitored.

On governance, Kurian reduced the problem to four questions: who the agent is, what it is permitted to do, what it did and where that can be seen, and what it is barred from touching. Every agent gets its own cryptographically attested identity with least-privilege permissions, and every action lands in an audit trail attributed to the agent rather than a person. All traffic passes through Agent Gateway, which Google describes as an AI network firewall enforcing policy in real time — write a rule once, such as "agents may not open documents classified Need to Know," and it applies to every agent in the company.

The deadline nobody is talking about: Gems retire in March 2027

While the agent announcement got the attention, Google confirmed the Gems → skills transition dates in a Workspace Updates post, and they are not negotiable.

Skills are reusable custom instructions built on the open, Markdown-native SKILL.md standard, which means you can copy a skill created on another platform into the Gemini app. The functional difference from Gems is inline stacking: you can apply a "weekly newsletter" skill and an "institutional brand guidelines" skill simultaneously in one prompt. Gems could not do that.

Two catches matter. First, skills do not sync between the Gemini app and Workspace — if you need a skill in both, you build it twice. Second, Workspace Studio flows using the "Ask a Gem" step stop working in March 2027, and you already cannot add new "Ask a Gem" steps.

The dates:

October 5, 2026 — skills begin rolling out in Workspace (Rapid Release completes by October 12; Scheduled Release starts October 19)
October 13, 2026 — skills begin rolling out in the Gemini app, completing by mid-November
November 17, 2026 — Gems move to the Settings panel of the Gemini app; you can still create, edit, and use them
No sooner than March 1, 2027 (business and enterprise) — Gems can no longer be created, edited, or used. "Ask a Gem" flows stop. Remaining Gems auto-migrate to draft skills
No sooner than June 1, 2027 (education) — same, plus Google Classroom and Gemini LTI-supported LMS

Google says it will share auto-migration details with admins at least 30 days in advance.

What to do with this

You cannot buy the Gemini agent today. No price, no GA date. SMBs are in early access; financial services and legal are in preview
You can and should audit Gems now. List what your team uses, assign an owner to each, and rebuild the ones that matter as skills. March 2027 is roughly five months away
Before any rollout, settle who sponsors each coworker agent account, what it can see, and what happens to its work when a spend cap pauses it. Whether a coworker agent consumes a Workspace licence is still unanswered
Three things remain unpublished: the Gemini agent's price, the enforcement delay on the new spend caps, and which Claude models the agent can actually route to. None of Google's two posts name them

The interesting shift here is not that Gemini became an agent. It is that an agent having its own email address, its own audit trail, and its own budget cap that can stop it mid-task is now the default enterprise operating model — and the vendors are shipping the guardrails at the same time as the capability. Google is unusually explicit that it expects cost governance, not capability, to be what determines whether this works.