Your company probably already “uses AI.” McKinsey’s 2025 State of AI report finds that nearly 80% of organizations run AI in at least one business function, but only around 20% have fundamentally redesigned the workflows that create real value.1
Studio, the visual workflow environment inside AgentOS, exists for the gap between those two numbers. It turns the workflows your team keeps talking about into AI‑driven flows you can design, test, and run in AgentOS — without requiring everyone to become an orchestration engineer.
The Automation Backlog Is Real
Every team has a list. Support ticket triage. Weekly competitive intel. Onboarding sequences. Meeting summaries. Workflows that eat time without being the actual work anyone was hired to do.
The ideas aren't the problem. The bottleneck is turning “we know exactly how this should work” into something that actually runs, can be trusted, and is easy to change.
You've probably already tried something. Zapier for the simple stuff. A few Python scripts someone half‑maintains. Maybe you evaluated three AI platforms and none of them were quite right.
You know what you need. You just can't build it fast enough, without pulling engineering resources, and in a way that's easy to hand off when requirements shift.
Why Building AI Workflows Is Still Hard
The problem isn't a shortage of tools. It's a structural gap between two kinds of tools that don't meet in the middle.
No‑code tools are fast to start, but hit a ceiling the moment your workflow needs real logic, branching, retries, or multi‑agent coordination. Diagrams get messy, debugging is painful, and it's hard to see what is actually happening when something goes wrong.
Code‑first frameworks are powerful, but building a production workflow means writing orchestration code, managing state, and debugging agent behavior before you've even validated the idea.
The result: engineers get pulled into workflow prototyping instead of product work. Non‑technical folks write Confluence pages about automation ideas that never ship. Everyone agrees the bottleneck is real. Nobody agrees on who should fix it.
Most teams end up with too little (a prompt and some if/else logic) or too much (a bespoke system nobody wants to maintain). The gap between "I can describe what I want" and "this is running in production" stays wide open.
AG2: The Framework That Makes This Possible
Studio is built on AG2, the open‑source multi‑agent framework behind AgentOS. That distinction matters.
Most visual workflow builders abstract away what's actually running. AG2 doesn't. Every workflow you build in Studio corresponds to real AG2 agent configuration: model selection, system prompts, tool bindings, conversation patterns. Nothing is hidden from you. It's surfaced visually and can be exported as code.
CaptainAgent is the orchestration engine at the core. When you describe a workflow in plain language, CaptainAgent turns that task into an initial agent graph: which agents to spin up, how they communicate, what logic gates and conditions are needed, and where human checkpoints belong. You start from a working foundation, not a blank canvas, and you can come back later and ask CaptainAgent to extend or refactor the same workflow from a new prompt.
Export and own your code. Once your workflow is built, export it as Python. Take it into your own infrastructure, modify it directly, version it like any other service, and run it however you want. You are never locked into Studio's runtime.
Studio: From Description to Working Workflow
Studio is a visual AI workflow builder inside the AgentOS web interface, where the building starts with a conversation.
You describe what you want in plain language. Studio uses CaptainAgent to interpret your task and generates an initial workflow on a visual canvas: agents, logic nodes, triggers, human review steps, and the connections between them.
You are not staring at an empty grid trying to remember every step. You are looking at a first draft of the actual system you had in mind — and you can run it immediately inside AgentOS.
Studio Visual AI Workflow BuilderDescribe It, See It Built
Open Studio and describe what you need:
"I need an agent that monitors our support tickets, categorizes them by urgency, and sends a Slack summary every morning."
Studio generates the workflow automatically. On the right side of your screen, a visual canvas populates in real time. Nodes appear, connections form, and your workflow goes from a sentence to a runnable agent graph in minutes.
You can hit Run right away and see how it behaves on real inputs. In AgentOS, you can step through each execution, inspect what every node saw and produced, and iterate: watch where it does the right thing, where it needs another branch or safeguard, and refine from there. The canvas becomes a shared artifact across engineering, ops, and business teams.
When you want to make bigger changes, you can either edit the graph visually or give CaptainAgent a new instruction (“add a weekly summary email,” “route anything over $5,000 through finance”) and let it update the workflow for you.
Configure Without Code
Every node on the canvas is independently configurable from a side panel: choose a model, write a prompt, bind tools, set output format, define conditions. You can see exactly what runs, in what order, and why. Change any of it by clicking.
- For agents, you choose the model, define the system prompt, and attach tools.
- For decisions, you specify routing rules, conditions, and fallbacks.
- For human steps, you decide what context to show and what kind of approval or input you need.
- For triggers, you control when and how the workflow starts — on a schedule, via webhook, or from product events.
When something breaks, you can see where. When requirements change, you add or swap a node rather than rewriting from scratch. Every change in the editor updates the underlying AG2 configuration, so the diagram is never just a sketch — it's the source of truth.
What You Can Wire Together
Studio focuses on a small set of primitives that compose into rich workflows:
| Node | What it does |
|---|---|
| Agent | An AI agent with a model, prompt, and tools. Chain multiple agents to research, analyze, and act in sequence. |
| Decision | Routes the workflow based on LLM reasoning or conditions. Define branches and a fallback: the agent decides which path to take. |
| Human Input | Pauses the workflow and asks a real person. Yes/No, free text, or single/multi‑select. The workflow waits, the human decides, then continues. |
| Trigger | Starts the workflow on a schedule, via webhook, or from a connected tool event (new ticket, new user, new order). |
| MCP Connector | Plugs into external services via Model Context Protocol. Your agents can search the web, read files, call APIs, and write to databases. |
With these pieces, teams are building:
📬 Support Ticket Router New ticket comes in → Agent reads and categorizes → Decision node routes to the right team → Human review for high-priority escalations → Sends Slack notification
📊 Weekly Business Intelligence Report Scheduled trigger every Monday → Agent pulls data from CRM → Another agent analyzes trends → Third agent writes the summary → Email sent automatically
🔍 Competitive Intelligence Monitor Webhook from your scraper → Agent reads new content → Decision node flags relevant items → Human reviews borderline cases → Updates shared Notion doc
🤝 Onboarding Automation New user event trigger → Agent sends personalized welcome → Schedules follow-up touchpoints → Routes to sales if high-intent signals detected
Start Building
Studio is available now at app.ag2.ai.
Describe what you want to build, and Studio will build the first version. You'll be looking at a working workflow in minutes, not weeks.
Studio is part of AgentOS: the platform for teams that want to move faster with AI without moving recklessly.
