Google Cloud Next '26: The agentic enterprise stack is now real

At Cloud Next '26, Google reframed Vertex AI as Gemini Enterprise and introduced an end-to-end stack for building, distributing, and governing AI agents. Here is what changed, who it serves, and what an SMB should do about it in the next two quarters.

Google Cloud Next '26 ran from April 22 to April 24 in Las Vegas, with an agenda built around a single thesis: the enterprise is becoming agentic, and Google now sells a full stack to support that transition. Across three keynotes, the company introduced or expanded twelve products that together constitute what Thomas Kurian, Google Cloud's CEO, described on Day 1 as "a unified stack with chips designed for models, models grounded in your data, and agents and applications built with those models."

The headline numbers framed the moment. Roughly 75 percent of Google Cloud customers are now actively using AI products. Three hundred and thirty customers processed more than one trillion tokens in the past twelve months. Direct API throughput now reaches 16 billion tokens per minute, up from 10 billion last quarter. Gemini Enterprise paid monthly active users grew 40 percent quarter over quarter. The data is consistent with what we observe across our own client base: agent adoption is no longer experimental, and the leaders are pulling away from the laggards faster than at any prior inflection point in enterprise software.

This analysis covers what we believe matters most for technology leaders at small and mid-sized businesses, organized by the three layers of the new Gemini Enterprise stack: the user surface, the developer surface, and the governance surface.

The user surface: Gemini Enterprise as the agent operating system

Gemini Enterprise is the renamed evolution of Vertex AI together with the Workspace AI surfaces, presented as a unified destination for end users to interact with agents. Six components are worth naming. Agent Designer is the no-code authoring environment. Inbox aggregates the status of every running agent the same way Gmail aggregates email. Projects are scoped containers that hold agent memory and files. Skills are agent capabilities invoked through @-mentions. Canvas is a real-time interactive editor for Docs and Slides. Deep Think is a multi-step reasoning mode for problems that exceed single-shot answers.

The most consequential addition is the Microsoft 365 export path. Canvas can now save directly to .docx and .pptx, removing the most cited objection from Microsoft-shop SMBs evaluating Workspace. Combined with what Google described as a 5x faster Microsoft 365 to Workspace migration tool, the licensing-portability case for moving off the Microsoft stack is now as strong as it has been at any point since Workspace launched.

For SMB technology leaders, the practical implication is that the "productivity AI" question, which was previously a vendor-by-vendor comparison between Copilot and Duet AI, has become a platform question. Gemini Enterprise is now the user-facing surface for Google's agent ambitions, and committing to it changes the trajectory of every downstream software decision.

The developer surface: ADK, Agent Studio, and the marketplace

Agent Development Kit architecture diagram. Source: Google Cloud.

The Agent Development Kit (ADK) is the code-first framework for engineering teams. Agents are declared as graphs of nodes, each node a callable function, with a Planner agent at the top deciding which Skills and Tools to invoke and in what sequence. A Memory Bank persists state across sessions. The framework integrates cleanly with remote Model Context Protocol servers, which means any tool already speaking MCP can be used by an ADK agent without rewriting.

For teams that prefer visual authoring, Agent Studio provides a low-code counterpart to ADK. Both surfaces target different audiences. Engineering teams write production agents in ADK. Line-of-business operators build internal-use agents in Studio. The realistic adoption pattern we recommend to clients is to start in Studio for the first three or four agents, then graduate the most complex one to ADK while the others remain in Studio.

Agent Marketplace inside Gemini Enterprise. Source: Google Cloud.

The Agent Marketplace is the announcement most likely to reshape SMB technology spending in the next eighteen months. Launch partners include Adobe, Asana, Atlassian, Box, Lovable, Mailchimp, Oracle, Salesforce, SAP, ServiceNow, UKG, and Workday. The economic shift is significant. A typical mid-market SaaS integration project costs between 40,000 and 120,000 dollars and takes three to six months to deliver. Installing a marketplace agent for the same SaaS platform takes minutes and runs roughly 120 dollars per month. For an SMB operations team of five, that is the difference between automating one workflow per year and automating ten.

The protocol layer: A2A, Agent Registry, and A2UI

Beneath the user and developer surfaces sits a protocol layer that received less press attention but may prove the most durable contribution of the conference. The Agent-to-Agent (A2A) protocol defines how two agents communicate. Agent Registry is the directory of every deployed agent within an organization, described from stage by Casey West, Google's architecture advocate, as "the DNS of your internet of agents." A2UI is an open-source standard for agents to generate their own user interfaces in a single shot, removing the front-end engineering bottleneck.

The combination matters because it makes multi-agent systems tractable for organizations without large platform teams. A sales agent that closes a deal can hand off to a finance agent for invoicing, which hands off to a customer-success agent for onboarding, without any of those agents requiring prior knowledge of one another. They discover capabilities through the registry, the same way devices discover services through DNS. For a small business, this is the architectural unlock that makes a twenty-agent network feasible with a five-person operations team.

The governance surface: Identity, Gateway, and the Wiz integration

The Gemini Enterprise Agent Platform stack. Source: Google Cloud.

For the past year, every CISO conversation about agents has eventually reached the same blocker: the lack of a defensible answer to an auditor asking how the organization can prove an agent did not exceed its authority. The new agent governance components address that gap directly.

Agent Identity issues every agent a cryptographic credential the same way human users receive identities. Agent Gateway provides a policy enforcement layer that scopes what each agent can touch. Anomaly Detection flags agents behaving outside their established baseline. Integration with Security Command Center captures the full audit trail. The Wiz acquisition, which closed earlier in the year, also produced three new defensive agents introduced at the conference: Red Agent for offensive testing, Blue Agent for incident investigation, and Green Agent for autonomous remediation.

For SMBs without a dedicated security analyst, the framing matters more than any single product. Pre-Next-26, agent governance had to be hand-rolled. Post-Next-26, the controls exist as platform primitives, which means SOC 2 and HIPAA evidence collection becomes a configuration exercise rather than a multi-quarter engineering project.

What this means for SMB technology leaders

Three takeaways stand out from the announcement set, each with a clear operational implication.

First, the build-versus-buy calculus has inverted. A year ago, building custom agents was the only realistic path because the ecosystem was thin. With the Agent Marketplace launch and 12 named partners, the default for the next twelve months should be to install existing agents and only build custom ones for genuinely differentiated workflows. The teams that internalize this first will achieve operating leverage faster than teams that default to engineering every solution.

Second, the governance model finally exists. Organizations that have been waiting for agent identity, scoped permissions, and audit trails before deploying production agents no longer have an excuse. The controls are present, the integration with Security Command Center is documented, and the cost of waiting another six months is no longer defensibility but lost competitive ground.

Third, the Microsoft 365 escape hatch is real. For SMBs already evaluating a Workspace move, the combination of accelerated migration tooling and Canvas-to-Office export changes the licensing math materially. The reasonable next step is a 90-day proof of concept with one team, not a full enterprise rollout decision.

How TekNinjas helps SMBs adopt the new stack

Most TekNinjas client engagements that start with "we want to use AI" spend the first four weeks discovering that the actual question is "which operating model do we want for the next decade." The new Gemini Enterprise stack does not eliminate that question, but it shortens the path to a defensible answer. Our recommended first 90 days for an SMB engagement follows this sequence:

• Weeks 1 to 2: identify three workflows where agents pay back inside the quarter. Lead qualification, accounts payable invoice routing, and tier-one IT helpdesk are usually on the list.
• Weeks 3 to 4: install one Agent Marketplace agent that covers one of those workflows, with the goal of teaching the team the Gemini Enterprise interface while shipping a measurable result.
• Weeks 5 to 8: build one custom agent in ADK or Agent Studio for the workflow the marketplace does not cover, with audit logging configured from day one.
• Weeks 9 to 12: stand up the governance baseline (cryptographic identity, scoped IAM policies, anomaly alerts to the existing on-call rotation) before scaling to additional agents.

The window in which this approach yields a meaningful competitive advantage is, in our estimation, the next 18 months. After that, the same operating leverage will be table stakes across the SMB segment, and the differentiation will move to which organizations have learned the patterns in production versus which are still starting fresh.

Sources: Google Cloud Next '26 Welcome post, Day 1 keynote recap, Day 2 developer keynote recap, and Cloud CISO Perspectives. All product details, customer references, and stage quotes are drawn from official Google Cloud publications. Photos hosted on Google's Cloud Blog CDN. Analysis and recommendations are TekNinjas's own.