Runtime as featured inForbesRead the article

AWS Bedrock AgentCore vs Google Gemini Enterprise Agent Platform

AWS Bedrock AgentCore vs Google Gemini Enterprise Agent Platform compared for 2026: runtimes, memory, identity, policy, models, evaluation, and pricing, plus where Runtime fits for payment teams.

Carlos Volante
Co-founder and CTO, Runtime
Updated October 8, 2026 · 7 min read

TL;DR

AgentCore and Gemini Enterprise Agent Platform are closely matched sets of agent primitives, and the right one is usually the cloud you already run on. AgentCore is more modular and has a mature policy engine; Google adds a low-code studio and Gemini first-party. Runtime is the finished harness for payment operations teams, and runs on either cloud.

Feature
AWS Bedrock AgentCoreAWS Bedrock AgentCore
Google Gemini Enterprise Agent PlatformGoogle Gemini Enterprise Agent Platform
RuntimeRuntime
What it isAWS Bedrock AgentCoreModular AWS agent servicesGoogle Gemini Enterprise Agent PlatformGoogle Cloud agent platform, ex-Vertex AIRuntimeAgent harness for payment teams
Build experienceAWS Bedrock AgentCoreHarness config, CLI, SDKsGoogle Gemini Enterprise Agent PlatformADK code or Agent Studio low-codeRuntimeOps teams build from SOPs
Agent runtimeAWS Bedrock AgentCoreMicroVM per sessionGoogle Gemini Enterprise Agent PlatformAgent Runtime with Agent SandboxRuntimeIsolated computer per run
MemoryAWS Bedrock AgentCoreShort- and long-term MemoryGoogle Gemini Enterprise Agent PlatformMemory BankRuntimeShared org memory across teams
Identity and tool accessAWS Bedrock AgentCoreIdentity, Gateway, PolicyGoogle Gemini Enterprise Agent PlatformAgent Identity, Agent GatewayRuntimeMasked credentials, RBAC
ModelsAWS Bedrock AgentCoreAny model, in or outside BedrockGoogle Gemini Enterprise Agent Platform200+ models in Model GardenRuntimeAny model, harness and model routing
Where it runsAWS Bedrock AgentCoreAWSGoogle Gemini Enterprise Agent PlatformGoogle CloudRuntimeYour AWS, GCP, Azure, or self-hosted
Approvals on money movementAWS Bedrock AgentCoreYou build themGoogle Gemini Enterprise Agent PlatformYou build themRuntimeBuilt in
PricingAWS Bedrock AgentCoreConsumption-based, publishedGoogle Gemini Enterprise Agent PlatformConsumption-based, publishedRuntimeBy request; free pilot for approved teams

AWS Bedrock AgentCore and Google Gemini Enterprise Agent Platform are closely matched sets of agent building blocks, and the better choice is usually the cloud you already run on: AgentCore for AWS shops that want modular services and a Cedar-compatible policy engine, Google for teams that want Gemini first-party and a low-code studio. Runtime is the AI agent harness for payment and fintech teams, a finished product that runs on either cloud, so you don't have to pick one vendor at all.

AWS and Google both rebuilt their agent offerings in 2026. AWS expanded Amazon Bedrock AgentCore with Policy, Evaluations, and a managed Harness, and Google launched Gemini Enterprise Agent Platform as the evolution of Vertex AI. Engineering teams choosing where to build production agents usually end up comparing the two directly.

AgentCore vs Gemini Enterprise Agent Platform at a glance

AWS Bedrock AgentCore is a set of modular services for building, deploying, and operating agents with any framework and any model. The core pieces are Harness (a managed agent loop defined by configuration), Runtime (a dedicated microVM per session), Memory, Gateway (APIs and Lambda functions exposed as MCP tools), Identity, Code Interpreter, Browser, Observability, Evaluations, and Policy, with Registry, Optimization, and agent Payments alongside. Each service can be used on its own.

Google Gemini Enterprise Agent Platform was announced at Google Cloud Next on April 22, 2026, and Google says all Vertex AI services will be delivered through it going forward. It is organized into Build (the open-source Agent Development Kit, low-code Agent Studio, Agent Garden templates, Model Garden), Scale (Agent Runtime, Agent Sandbox, Memory Bank, Sessions), Govern (Agent Identity, Agent Registry, Agent Gateway with Model Armor, threat detection), and Optimize (simulation, evaluation, observability, and an optimizer).

Building agents

AgentCore. Developers work through the AgentCore CLI, SDKs, and APIs. Harness, generally available since June 2026, lets a team define an agent with a model, instructions, and tools and run it without writing orchestration code. For custom logic, Runtime hosts agents built with LangGraph, CrewAI, LlamaIndex, Strands, OpenAI Agents SDK, Google ADK, or your own code.

Gemini Enterprise Agent Platform. ADK is open source in Python, TypeScript, Go, Java, and Kotlin, and Google says it can be containerized and run anywhere. Agent Studio adds a visual, low-code path that exports to ADK, and Agent Garden provides templates, including financial analysis and invoice processing. This gives Google a stronger story for less technical builders.

Runtime, memory, and security

AgentCore. Every session gets its own microVM with isolated compute, memory, and filesystem, terminated and sanitized when the session ends. Sessions run up to 8 hours on microVMs or up to 14 days on instances. Policy, generally available since March 2026, intercepts every tool call at the Gateway and checks it against rules written in natural language or a Cedar-compatible language. Identity works with Okta, Entra ID, Cognito, and other providers.

Gemini Enterprise Agent Platform. Agent Runtime advertises sub-second cold starts and supports long-running agents that work for days. Memory Bank generates long-term memories from conversations. Agent Identity gives each agent a cryptographic ID with auditable authorization policies, and Agent Gateway applies consistent security policies and Model Armor protections against prompt injection and data leakage. Security Command Center powers an agent security dashboard.

Models and ecosystem

AgentCore. AWS says Runtime works with any foundation model in or outside Bedrock, including OpenAI, Gemini, Claude, Nova, Llama, and Mistral, and Harness supports Bedrock, OpenAI, Gemini, and OpenAI-compatible providers. Observability emits OpenTelemetry into CloudWatch.

Gemini Enterprise Agent Platform. Model Garden lists more than 200 models, with Gemini first-party and Claude Opus, Sonnet, and Haiku among third-party options. Event-driven and batch agents connect natively to BigQuery and Pub/Sub. Google names PayPal and Payhawk among its agent customers in financial services.

Pricing

AgentCore is consumption-based with no upfront commitment. Runtime microVMs start at $0.0895 per vCPU-hour and $0.00945 per GB-hour, billed per second on active use. Gateway is $0.005 per 1,000 invocations, and Memory, Policy, Evaluations, Browser, and Code Interpreter are metered separately.

Gemini Enterprise Agent Platform is also consumption-based. Agent Runtime and sandbox compute are $0.085 per vCPU-hour plus memory per GiB-hour, idle time between turns is not billed, and Memory Bank, Sessions, and Gateway usage are metered. On both platforms, model usage is billed separately.

Where Runtime enters the picture

Runtime is the AI agent harness for payment and fintech teams. Both platforms give an engineering team excellent parts, and both tie the result to one cloud. Building on either still means deciding which payment actions need a human, building the approval flow and the interface, wiring in your ledger, processor, and bank portals, and producing an audit export a sponsor bank will accept. An engineering team can spend two quarters on that before the first agent touches real data.

Consider a partner-bank RFI about a flagged wire. Compliance needs the transaction, the KYC file, the alert history, and the analyst notes, pulled from four systems and answered within the bank's deadline. On Runtime a compliance agent gathers the evidence, drafts the response, and waits for an approver before anything goes to the bank, and the full record of that run stays with you. That works the same whether your data sits in AWS, Google Cloud, or both.

  • No single vendor to depend on. Agent computers run in your AWS, GCP, or Azure account or fully self-hosted, with your own sandbox provider and fallbacks. Models route across Claude, GPT, Gemini, and open-weight models, and harnesses across Claude Code, Codex, and OpenCode.
  • Guardrails you set. Read-only by default, approvals before money moves, RBAC, masked credentials, PII and PCI stripped from prompts and logs, and every run recorded.
  • Built for teams. Support, payment ops, finance, risk, and compliance share agents, templates, and one memory, called from Slack, Teams, email, or the dashboard.
  • An engineer, not an account executive. A forward-deployed AI engineer with a payments background (Hulu's payments team at Disney, Finix, Modern Treasury, BlackRock) builds the first agents with you.
AgentCoreGemini Enterprise Agent PlatformRuntime
What you getModular AWS servicesGoogle Cloud agent platformFinished harness for payment teams
CloudsAWSGoogle CloudAWS, GCP, Azure, self-hosted
Approvals on money movementYou build themYou build themBuilt in
Who builds agentsDevelopersDevelopers, low-code buildersOps, risk, finance, with an FDE
New agent

Create a banking ops agent that handles RFIs, hold-harmless requests, and returns of funds from our partner bank, verifies against our ledger, and drafts replies for my approval.

Describe the agent you want
Agent ready

bank-ops-agent

Tools picked from your prompt

EmailLedgerSFTPMastercard
Describe the work in plain English, attach the SOP, and Runtime builds the agent.

Which should you choose

  • Choose AgentCore if you run on AWS, want to adopt agent services one at a time, and value a deterministic policy engine on every tool call.
  • Choose Gemini Enterprise Agent Platform if you run on Google Cloud, want Gemini first-party, or want a low-code studio next to a code-first framework.
  • Choose Runtime if the goal is to put payment ops, risk, and finance teams on agents with approvals and an audit trail built for money movement, across whichever clouds you already use.

Frequently asked questions

Is AgentCore the AWS equivalent of Gemini Enterprise Agent Platform?

Broadly, yes. Both provide a managed agent runtime, memory, agent identity, a gateway for tool access, observability, and evaluation. Google packages them as one platform that replaced Vertex AI, with a low-code Agent Studio. AWS offers them as independent services you can adopt one at a time.

Which has better model choice?

Both are open. AgentCore Runtime works with any foundation model in or outside Amazon Bedrock. Google's Model Garden offers more than 200 models, including Gemini, Gemma, and Anthropic Claude, and ADK can connect to other providers.

Which is cheaper, AgentCore or Gemini Enterprise Agent Platform?

Both publish consumption pricing and their compute rates are close: AgentCore Runtime microVMs start at $0.0895 per vCPU-hour, and Google's Agent Runtime compute is $0.085 per vCPU-hour. Total cost depends on memory, gateway, evaluation usage, and models.

Can I use both clouds for agents?

Each platform's managed services run in its own cloud. Runtime runs agent computers in your AWS, GCP, or Azure account, so one harness can cover agents across clouds.

Where does Runtime fit next to AgentCore and Gemini Enterprise Agent Platform?

Both give engineers agent infrastructure on one cloud. Runtime is the finished harness payment ops, risk, and finance teams use, with approvals before money moves and a full audit trail. It runs agent computers in your AWS, GCP, or Azure account or self-hosted, so it works with either platform rather than replacing your cloud.

See Runtime on your most manual workflow

Bring one SOP. A forward-deployed AI engineer builds the first agent with your team, inside your cloud.