Agentic AI System Designs · hub

Agentic Systems at a Glance

Every agent system covered by the series on one page: harnesses, frameworks, runtimes and builders, a comparison of the harnesses, and a quick chooser. Start with Four Ways to Build an Agent. Claude, made by Anthropic, helps write this site, and this page covers Anthropic, Google, OpenAI and Amazon products.

All systems by kind

🧱 Frameworks

you assemble the parts
OpenAI Agents SDKcoming in wave 2Google ADKcoming in wave 2LangChain / LangGraphcoming in wave 2CrewAIcoming in wave 2Claude Agent SDK & Microsoft Agent Frameworkcoming in wave 2

⚙️ Runtimes

you deploy and operate the agent
Amazon Bedrock AgentCorecoming in wave 3LangSmith Deploymentcoming in wave 3Temporalcoming in wave 3Vertex AI Agent Engine & Azure AI Foundrycoming in wave 3

🧩 Builders

you design an agent visually
n8ncoming in wave 4Zapier Agentscoming in wave 4Microsoft Copilot Studiocoming in wave 4Makecoming in wave 4

Which kind do you need?

Tick what is true for your project.

Harnesses, layer by layer

The eight-layer reference stack used in every teardown.

HarnessL1InterfacesL2Sessions & orchestrationL3Agent loopL4Context engineL5ToolsL6Policy & safetyL7ExecutionL8Model & telemetry
Claude CodeTerminal, IDE, desktop, web, Slack; Agent SDKSessions with resume; subagents; cloud sessions on the webOne general think → act → observe loopCLAUDE.md, auto memory, compactionBuilt-in file/shell/search tools, MCP, skillsPermission rules and modes, hooksYour machine (optional sandboxing ⚠️); cloud for web sessionsClaude models; transcripts, cost and usage telemetry
CodexTerminal, IDE, web (chatgpt.com/codex)Parallel cloud tasks; local sessionsAgent loop per taskAGENTS.md, repo contextShell, file edits, MCPApproval modes ⚠️OS sandbox locally ⚠️; container per cloud taskOpenAI models; task logs and diffs
Gemini CLI / JulesTerminal (CLI); browser + GitHub (Jules)Interactive session vs asynchronous taskAgent loop (CLI); plan-then-execute (Jules ⚠️)GEMINI.md context filesBuilt-in tools, MCPConfirmations in the CLI; plan approval in Jules ⚠️Your machine (CLI); cloud VM (Jules ⚠️)Gemini models (CLI "not optimized for Gemini 4+")
Deep AgentsYour own app (library)LangGraph threads; sub-agentsPlanning loop on LangGraphFilesystem as memory; offloadingYour tools, MCP, skillsHuman-in-the-loop interruptsPluggable backends: local, sandboxed, remoteAny chat model; LangGraph state and tracing
Hermes AgentTerminal + Telegram, Discord, Slack, WhatsApp, Signal, emailGateway, subagents, cron schedulerAgent loop (details on its docs site ⚠️)Curated memory, FTS5 session search, user modelTools, self-written skills (agentskills.io), Python RPC scriptsDepends on backend and configuration ⚠️Six backends: local, Docker, SSH, Modal, Daytona, Vercel SandboxModel you configure ⚠️

One line per layer; our reading of each project's documentation and code as of 7 Nov 2026.

Harnesses compared

HarnessRuns whereOpen sourceSubagentsPersistent memorySkillsMCPSandboxHuman approvalAsync / cloud tasks
Claude CodeTerminal, IDE, desktop, web—✓✓✓✓◐✓✓
CodexTerminal, IDE, cloud✓ (CLI)◐◐◐✓✓✓✓
Gemini CLI / JulesTerminal (CLI), cloud VM (Jules)✓ (CLI)◐◐◐✓◐✓✓
Deep AgentsYour Python / JS app✓✓◐✓✓◐✓—
Hermes AgentTerminal + chat apps✓✓✓✓◐✓◐✓

✓ built in · ◐ partial or via add-ons · — not a feature. Our reading of each project's documentation as of 7 Nov 2026; check before relying on it.

Framing adapted from an infographic by Denis Panjuta (third party). Feature marks are our reading of each project's documentation as of 7 Nov 2026. The infrastructure underneath all four kinds is covered in Agent Layer at a Glance.