Agent Marketplaces Are the New App Store: What Google's Gemini Enterprise Reveals About the SaaS Reset

Source: Unsplash

Google launched the Gemini Enterprise Agent Platform on April 22, 2026, and it is not just another developer release. It is an agent marketplace where Salesforce, Oracle, and ServiceNow agents now run inside the same governed runtime as Google's own agents. This article unpacks what that actually means for product designers, why the agent marketplace is closer to an App Store moment than a SaaS killing moment, the three new UX layers that did not exist last year, and what I am personally telling product teams to stop designing right now.

I have been building products for eight years, shipped 42 of them, and I cannot remember a six month stretch as disorienting as the one we are living through. On April 22, 2026, Google announced the new Gemini Enterprise Agent Platform, rebranded Vertex AI under it, and committed $750 million to push partners into agentic builds. Most coverage framed it as another Google vs OpenAI vs Anthropic round. From where I sit as a product designer, the bigger story is the partner directory inside Gemini Enterprise. Salesforce, Oracle, and ServiceNow are now shipping agents that run inside someone else's governance layer. That is the part that quietly broke the old SaaS shape.

"The Gemini Enterprise Agent Platform brings together model selection, model building, and agent building capabilities with new features for agent integration, DevOps, orchestration, and security. Customers can discover and deploy third party agents from leaders like Oracle, Salesforce, and ServiceNow within a secure and governed environment."
— Google Cloud Blog, April 22, 2026

What Google Actually Shipped on April 22

The platform consolidates three things into one runtime: a developer surface that replaces Vertex AI, a Gemini Enterprise app where teams can discover and run agents, and an open partner ecosystem with first class access to more than 200 models through Model Garden. That model count includes Gemini 3.1 Pro, Gemini 3.1 Flash Image, Lyria 3, and third party models like Anthropic's Claude Opus, Sonnet, and Haiku. For context, Gartner forecasts that 40 percent of enterprise applications will embed task specific AI agents by the end of 2026, up from less than 5 percent in 2025. The market for agentic AI is projected to grow from roughly $9.14 billion in 2026 to $139.19 billion by 2034, at a 40.5 percent CAGR, according to Fortune Business Insights.

So this launch is not arriving in a quiet quarter. It is landing in the middle of the steepest enterprise software repricing in fifteen years. After Anthropic's Cowork enterprise plugins shipped in February, Salesforce dropped 7 percent, ServiceNow dropped 7 percent, Intuit fell 11 percent, Thomson Reuters cratered 15.83 percent, and LegalZoom sank 19.68 percent in a single trading window. Now those same vendors are publishing agents inside Google's directory. That is the part the headlines are underweighting.

Why This Is the App Store Moment, Not the SaaS Killer Moment

I keep seeing two camps online. One says agents are killing SaaS. The other says SaaS will absorb agents and nothing changes. Both miss the actual move. The agent marketplace is the iPhone App Store moment. Apple did not kill phone software in 2008. It moved the value from the device manufacturer to the distribution layer, and changed who got rich. The Gemini Enterprise app, the Cowork agent directory, the OpenAI Frontier Alliances program, these are all racing to become the place where buying decisions happen. The vendor that owns the agent directory owns the next layer of enterprise software economics, regardless of who actually builds the underlying agents.

If you are a designer at Salesforce or ServiceNow right now, your product is already a tile in someone else's marketplace. The implications are uncomfortable. Your branding is reduced to an icon. Your onboarding flow is replaced by an agent install step that takes thirty seconds. Your dashboard, the thing your team spent two years polishing, is collapsed into a chat panel that asks the agent to summarize the same data. If you do not believe me, open the Gemini Enterprise app and run a test deployment of the Salesforce agent. The whole pre AI Salesforce UX is compressed into a card.

The Three New UX Layers Designers Now Have to Solve

Working with three different enterprise teams this quarter, I have started naming the new design layers out loud because nobody seems to agree on what to call them. Here is the working vocabulary I have settled on:

• The agent discovery layer. How does a user know an agent exists, what it does, what data it touches, and whether it is trustworthy? This is product page design for agents, and most teams are still using model cards that look like academic papers. They do not work for enterprise buyers.
• The handoff layer. When an agent finishes a task, what does the human see? When it gets stuck, who picks up? Most teams are still designing this as a chat transcript. That fails the moment the agent runs for fifteen minutes and produces a forty step plan. The handoff layer needs its own pattern language, with structured artifacts, diff views, and approval prompts.
• The cross agent orchestration layer. When the Salesforce agent and the ServiceNow agent both need to act on the same record, who wins? What does a user see when two agents disagree about what should happen? This is the hardest one, and almost nobody is shipping good UI for it yet.

If your design system does not have components for at least the first two layers, your product will feel a generation behind by Q3 of this year. I am not exaggerating. The teams I work with that already have agent discovery cards, audit trails, and a clear human in the loop pattern are shipping at roughly twice the velocity of teams still polishing dashboards.

What I Am Telling Teams to Stop Designing

Three things I am asking product designers to stop doing this month:

First, stop adding new dashboard views. If your roadmap has more than two new charts in the next quarter, kill them. Replace them with a single agent that can answer whatever question the chart was trying to answer, plus an audit panel that shows how it got there. The audit panel is the new dashboard. I wrote about this in more depth on Medium last month, and the response from in house design leaders was the most heated comment thread I have ever had on that platform.

Second, stop designing settings pages as forms. Settings inside an agent native product are policy. Permissions are policy. Tone is policy. Data access is policy. Designing them as a flat list of toggles is a 2018 pattern. Try a policy editor instead, the way IAM consoles handle role boundaries. Users get one place to express intent across all the agents they have installed, with constraints and exceptions stated as rules, not checkboxes.

Third, stop letting engineering own agent error states. When an agent fails, the user does not see a stack trace. They see a confused product. Designers should be writing the failure copy and the recovery flow themselves. If your product's agent says "I encountered an error processing your request", that is a UX failure pretending to be a backend failure. Own it, write the right copy, and design the retry loop.

The Honest Take From Inside the Cycle

I will be transparent. I think Google's Gemini Enterprise launch is more important than the OpenAI and Anthropic moves of the last six months, because it normalizes the multi vendor agent runtime. OpenAI's Frontier and Anthropic's Cowork are still primarily single ecosystem stories. Google walked into the room with Salesforce, Oracle, and ServiceNow agents already inside, and said the runtime is the product. That is a different bet, and it is the one I think wins for the kinds of regulated enterprises I have spent most of my career designing for.

The risk for designers is real though. 96 percent of organizations are increasing their agentic AI investments, which tells you the spend is locked in. 79 percent of enterprises report at least some agent adoption already. The window to learn this craft is narrower than the window we had with mobile or with cloud. If you are still treating agents as a feature inside an existing SaaS product instead of a product surface in their own right, you are building the wrong thing. I do not say that to be alarmist. I say it because I have watched designers refuse to learn responsive design in 2012, refuse to learn design systems in 2018, and they were right that nothing dramatic happened that quarter. They were also out of the senior conversations within two years.

I will keep tracking this on reloadux.com and on my Medium under Bootcamp. If you want the deeper version of these patterns with worked examples, that is where the long form lives.

What patterns are you actually seeing in your own product? Are you already designing agent discovery cards, or still polishing dashboards? Drop a comment with what you are working on. I read every one and I am collecting the most interesting ones for a longer piece on reloadux.com next month.

Sources: Google Cloud Blog (Gemini Enterprise Agent Platform announcement, April 22, 2026), Google Cloud Press Corner ($750 million partner fund, April 22, 2026), Bloomberg (Google Releases New AI Agents to Challenge OpenAI and Anthropic, April 22, 2026), Fortune (AI Agents from Anthropic and OpenAI Aren't Killing SaaS, February 10, 2026), Fortune Business Insights (Agentic AI Market Forecast 2026 to 2034), Gartner (enterprise application AI agent forecasts), Constellation Research (Enterprise Technology 2026 Trends).