Google announced the Gemini Enterprise Agent Platform at Cloud Next on April 22, 2026, and inside that platform, the panel most builders actually open first is the no-code Agent Designer. As of this writing — six days post-launch — most of the SERP is still occupied by Google’s own marketing and the press round-up. Real builder chatter is just starting; the most-engaged community post is a Japanese developer’s hands-on test of the schedule-execution feature, with one like and one repost. The next two months will produce thousands of “how-I-used-Gemini-Agent-Designer” threads. This guide is the one you can read before then, so when the threads start landing you already know what’s actually going on.
Five minutes if you skim, fifteen if you read carefully. By the end you’ll know what Agent Designer is, where it fits next to Vertex AI Agent Builder and the older Agentspace product, what panels you’ll see in the UI, the three settings Google’s own documentation glosses over, and when to keep it inside Agent Designer versus when to move down to the Agent Development Kit (ADK).
What Agent Designer Actually Is
Strip the marketing. Agent Designer is a no-code/low-code visual builder embedded inside the Gemini Enterprise app that lets a non-developer define an AI agent, give it tools, point it at data, and either trigger it on a schedule or expose it as a chat-style helper to colleagues. Google’s official docs call it “an interactive no-code, low-code platform for creating, managing, and launching single and multi-step agents in Gemini Enterprise.” Translated into plain English: it’s the same idea as Microsoft Copilot Studio, but inside Google Workspace, with first-class connectivity to Gmail, Drive, Calendar, and Docs.
The launch context matters because Google has been actively reshuffling its enterprise agent strategy for two years, and the names have moved around. Three products that were separate in 2025 are now consolidated:
- Vertex AI Agent Builder — the developer-centric, code-friendly cloud platform that’s been the production-grade option on Google Cloud since 2024. It still exists, but at Cloud Next ‘26 it was rebranded and folded under the broader Gemini Enterprise Agent Platform umbrella, with the developer surface now exposed through the Agent Development Kit (ADK) rather than a “Vertex AI Agent Builder” link.
- Agentspace — last year’s separately-branded employee-facing app for building and using agents tied to enterprise search. It’s been absorbed. Multiple commentators put it bluntly: “Google replaces last year’s Agentspace app and adds a visual Agent Designer.” If you opened a tab labeled “Agentspace” in 2025, that experience now lives inside the Gemini Enterprise app.
- Agent Designer — the no-code surface that sits on top of all of this. This is what you click on first, and it’s the focus of this guide.
The conceptual distinction worth holding in your head: Agent Designer is the business-facing surface. ADK is the developer-facing surface. They share the same underlying agent runtime. That means an agent prototyped in Agent Designer can be exported into ADK for code-level customization without rebuilding from scratch, and a developer-built agent can be wrapped to appear inside the Agent Designer gallery for non-developers to invoke.
The Pricing in Plain Numbers
Three numbers worth knowing.
Gemini Business: $21/user/month. This is the entry tier that includes the Gemini chat experience plus basic agent usage. It does not include the full Agent Designer feature set — you can use shared agents, but you can’t build new ones at this tier.
Gemini Enterprise Standard: $30/user/month. Includes Agent Designer access for simple single-step agents. Most teams that want to “try the no-code builder” land here.
Gemini Enterprise Plus: $30/user/month, but with the full feature set. This is the tier that the partner documentation specifically ties to “the ability to build and share custom agents via the no-code Agent Designer,” including multi-step orchestration, advanced governance, and Memory Bank for cross-session memory. Yes, it’s the same headline price as Standard — Plus is positioned as the “everything turned on” version of Enterprise rather than a price-step-up.
If your company is on Workspace already, Agent Designer slots in as an add-on. Workspace Business Standard ($14/user/month) plus Gemini Enterprise ($21-30) is the typical bundle for a mid-market deployment. Most analyst write-ups have noted promotional discounts running through April 2026 to drive Cloud Next adoption — worth asking your Google Cloud rep about, particularly if you’re committing more than 100 seats.
The 5-Minute First-Agent Walkthrough (Seven Panels)
Open the Gemini Enterprise app. The left navigation includes a panel called Agents. Click it; you land in the Agents gallery with three sections:
- Made by Google — the pre-built starter agents Google ships with the platform (we’ll come back to these)
- From your organization — agents that other people in your company have already built and shared
- Your agents — the agents you have built yourself
Click Create agent. That action launches Agent Designer. From here, the canonical 5-minute path runs through seven panels.
Panel 1 — Name and Instructions. The agent gets a name and a natural-language description of what it should do. Google’s official prompt example is something like: “Summarize new support tickets in our ‘Priority’ label and propose responses.” The Instructions field is the agent’s role definition — write it the way you’d brief a new junior employee on their first task. Be specific about what to do, what not to do, and what voice or tone to use.
Panel 2 — Tools. This is where you connect what the agent can actually act on. Workspace tools (Gmail, Drive, Calendar, Docs, Sheets) are presented as first-class options with one-click authorization. Beyond that, Google ships connectors for common third-party SaaS — Jira, Salesforce, ServiceNow, Slack, Notion. For tools not in the gallery, you can add an HTTP function tool by pasting an API spec. The pattern most builders settle into in week one: start with two or three Workspace tools, get the agent working, then add third-party tools incrementally.
Panel 3 — Knowledge sources. Distinct from Tools. Tools are actions the agent can take (send an email, create a calendar event); Knowledge sources are data the agent can read for context (a Drive folder, a shared Drive, a Google Cloud data connector pointing at BigQuery). The architecture documents emphasize that this separation is deliberate — it lets you give an agent broad read access to a knowledge base while keeping tight control on what it can write or send. Configure this panel narrowly. An agent that has access to “all of Drive” reads, summarizes, and (depending on Tools configuration) potentially shares anything any user might have access to. Don’t do that on Day 1.
Panel 4 — Triggers. Three trigger types are documented: manual (the agent waits for someone to invoke it through chat), schedule-based (cron-style: every weekday at 8 a.m., for example), and event-based (fires when an email matching a filter arrives, or when a Drive document is updated). Most first agents stay on manual triggers for the first week, then graduate to schedules once the maintainer trusts the output.
Panel 5 — Permissions. Who in your organization can invoke this agent? The default is just you. You can extend to specific Google Groups, to entire org units, or to “all users in this Workspace.” This is the panel that determines blast radius. Practical guidance: leave at just you until the agent has run cleanly for a week, then promote to a small test group, then expand.
Panel 6 — Test. Built into Agent Designer’s right-hand pane. Type a sample input, watch the agent execute, see which Tools it called and which Knowledge sources it pulled from. Read its draft output. Iterate. The Test panel is the single most underrated feature in the product — most builders rush past it, then ship an agent that reasons differently in production than in their head.
Panel 7 — Deploy. Three deployment surfaces: the Gemini Enterprise app’s chat surface, the Agents gallery (where colleagues with appropriate Permissions will see it), and an embed link you can drop into a SharePoint page, an internal portal, or a Workspace site. Pick at least one; you can add others later.
That’s the canonical path. From “Create agent” to a working draft typically takes 20-30 minutes for a first build, less for builds two and beyond once you’ve internalized the panel layout.
The Three Settings Google’s Docs Don’t Emphasize Enough
This is the part of the SERP that does not yet exist — what’s missing from Google’s own documentation, observed from the small but useful community pulse already published.
Setting 1 — Memory Bank, off by default. Memory Bank is the feature that lets an agent remember a user’s preferences, history, and habits across separate sessions. By default, agents in Agent Designer treat each invocation as stateless — every conversation starts from zero. If you’re building a recurring workflow agent (one that helps the same person every morning, for example), turn Memory Bank on explicitly in the Plus-tier configuration. The feature is buried inside an advanced tab; most first-time builders never find it.
Setting 2 — Tool execution mode: “auto” vs “ask first.” Each Tool you add can run in two modes. Auto means the agent invokes it without confirmation when it decides the tool is needed. Ask first means the agent shows the user what it’s about to do and waits for approval. Workspace tools default to auto in some configurations, which is fine for read operations like “fetch my calendar,” but dangerous for write operations like “send this email to the customer.” The 30-second fix on every agent you build: walk through the Tools list and explicitly set ask first for any Tool that creates, sends, or modifies anything externally. This is the setting that prevents “the agent emailed the wrong client” stories.
Setting 3 — Knowledge source freshness. Knowledge sources cache. Drive folder content, BigQuery extracts, all of it. The default cache lifetime is 24 hours for most source types, which means an agent that’s “always up to date with the latest pricing memo” might actually be working off yesterday’s version of the memo. The freshness control is in an advanced tab on each Knowledge source. For agents that depend on time-sensitive data — pricing, inventory, calendar availability — set the freshness to “always fresh” or 1-hour cache. The trade-off is latency on each agent run, but for the workflows where freshness matters, the trade is worth it.
Agent Designer vs ADK: When to Drop Down to Code
A practical heuristic. Stay in Agent Designer when:
- The agent’s logic is “do these three things in sequence,” not “make a complex decision based on twelve interacting factors”
- You’re connecting standard Workspace and SaaS tools, not internal company APIs that need authentication tricks
- The agent’s audience is non-developers, and the maintenance has to live with the team that uses it
Drop to ADK when:
- You need multi-agent orchestration where one agent dispatches to specialized sub-agents (the developer-side calls these “graphs” — networks of agents handing work between each other)
- You need fine-grained logic that’s hard to specify in a single Instructions box (“if the customer is in Tier 1, call this API first; if they’re in Tier 2, do A or B based on contract status”)
- You’re building a production system that needs deployment pipelines, version control on agent definitions, and CI/CD integration
The good news: an agent built in Agent Designer can be exported into ADK without losing work. The migration path is one-directional but clean — you don’t need to rebuild from scratch when complexity demands a code-level treatment.
How It Compares to Copilot Studio, Salesforce Einstein, AWS Bedrock AgentCore
Three rough takes, kept short.
Microsoft Copilot Studio. Same conceptual product (no-code agent builder for the enterprise productivity stack), built around Microsoft 365 instead of Google Workspace. Strengths: deepest enterprise integration if your company is already on M365 Copilot. Weaknesses: the studio surface is more developer-leaning than Agent Designer; non-technical builders find Copilot Studio’s flow editor harder to start in. Pricing: $30/user/month for Microsoft 365 Copilot, with Agent Studio access included; the Copilot Studio standalone is $200/month per tenant plus per-message metering.
Salesforce Einstein Agent Builder. Distinct positioning: Salesforce-data-native. If your company runs sales, service, or marketing operations inside Salesforce, Einstein Agent Builder is the product with the deepest CRM-context awareness. Pricing is bundled into Salesforce per-user costs, but the agent layer’s premium add-on can run $50-100/user/month depending on edition.
AWS Bedrock AgentCore CLI. Different audience entirely. AgentCore is a developer command-line tool for building agents on AWS Bedrock — there’s no no-code UI in the AgentCore CLI itself; the no-code surface lives in different AWS products. If you’re an AWS-first shop and the agent workflow lives close to your data (Redshift, S3, Lambda triggers), AgentCore is the right match. If you want a non-developer to build the agent, you’re not in AgentCore’s target audience.
The honest read. For a typical mid-market enterprise on Google Workspace, Agent Designer is the lowest-friction starting point for building first agents in 2026. For an M365 shop, Copilot Studio. For a Salesforce-heavy operations team, Einstein. For an AWS-native developer-led organization, AgentCore. The platform decisions are mostly inherited from where your data already lives — pick the agent builder that matches your company’s existing cloud and SaaS posture, not the one that has the prettiest demo video.
The Early-Mover Gap (and Why That Matters for You)
As of late April 2026, real builder/developer hands-on threads with meaningful engagement are scarce. The most-engaged community post is a Japanese developer’s hands-on test of schedule execution; the next is a builder’s seven-step Agent Studio orientation; the third is a frustration reply from a tester who couldn’t find the visual builder button at all. That’s the entire pulse. Google Cloud Next was less than a week ago.
What this means practically: the SERP for gemini enterprise agent designer tutorial is still wide open. If you’re a builder at a company on Workspace, the agent you ship in week one and write up plainly afterward — what you built, what you learned, what tripped you up — has roughly 90 days of clean ranking real estate before the bigger platforms (Reddit deep threads, YouTube walkthrough videos, Lenny’s Newsletter posts, Ben’s Bites round-ups) saturate the SERP. That’s a real first-mover window, but only if you actually build something and document it honestly while the field is empty.
Bottom Line
Agent Designer is six days into being available, and if your company’s on Workspace, it’s the easiest way to put a working agent in front of users this week. The 5-minute walkthrough is genuinely 5 minutes once you know the panels. The traps are the three default-behavior settings — Memory Bank off, Tool auto-execute, Knowledge freshness — that the docs don’t flag clearly enough. The migration path to ADK is clean if you outgrow the no-code surface.
Build something this week. The SERP is empty, the platform is fresh, and the first round of “I built a real agent in Gemini Enterprise” posts is going to age very well in search.
