About
Agents move with speed. Your team stays in control. Steps holds the workflows your team wrote, so the AI you already use doesn’t start from scratch.
How it works
People guide. Your AI follows.
One shared workflow for your team and the AI agents you already use. When judgment is needed, people carry on from the same step.
Same workflow
See the next step.
People and agents share the workflow. If an agent stalls, continue from the same step.
Incident response
Choose your path
Opportunity
Agents can take on critical work
Agents can carry out critical work, not only drafts, when they follow a workflow your team wrote.
People still decide the steps, and everyone can see them.
Constraint
No shared workflow, no real control
If the next step is hidden, people can’t take over when an agent stalls.
A shared workflow is how you continue.
Where Steps fits
AI-native workflows your team writes and your AI follows
Workflows in one place, with context at every step and the next step visible. People take over when judgment is needed, then hand back.
Write the workflow so it can be followed and checked.
Branches, discussions and edits keep the same workflow current.
One current workflow for your team and its AI.
One place
Past experience lives in the discussion
The team’s steps stay in one place, and the discussion on a step holds what happened last time: the gotchas and the exceptions. Your AI reads that discussion with the published step. People, and your AI with Can edit, add to the discussion when something is worth noting, not on every run.
Everyone contributes
The whole team can write the workflow
Not only a few people writing prompts. Anyone can record how the work actually runs, on video, in a voice note via your AI, or in the discussion on a step.
Your stack
Works with the AI you already use
Keep Claude Code, Cursor, Codex or your own setup. Steps doesn’t run your AI. It’s where your team says what should happen.
Developers
Find the workflow in the context flood
Notes, skills, and plugins compete for attention. A Steps workflow is the one you want followed.
Why a workflow
Keep agents moving when automations break
A brittle script or a one-shot prompt falls over when something changes, or an edge case appears.
It is an orchestration of small, reusable skills, and the way your team processes the work: the order, the checks, and the decisions.
Write the steps for how code is written and structured in this repo, and the order of things to check before changing an API or opening a pull request. Your AI reads the published workflow and follows it.
Agents
Don’t make your AI invent the process.
Cursor, Claude Code, Codex, VS Code, any MCP client or your own orchestrator can read the workflows your team already wrote, and, with your permission, change them, including the current published version.
They follow that workflow instead of inventing one.
What you get
Your AI starts from the process your team defined.
When a decision changes, update the workflow so everyone keeps following the same one.
You still write, publish and share.
Hook in
A token, then find, confirm, and follow
Steps doesn’t run your AI. It guides it. Create a token in Settings, and your assistant looks up the workflow, checks it’s the right one, and follows the steps.
Connect your AI
Hook up the agent you already use, or open a workflow to see what it will follow.
FAQ
Questions
Short answers. Open a question for the detail.
Is Steps an AI?
No. Steps doesn’t run a model or act on its own. It’s where your team writes and versions the workflows your AI follows. You bring the AI you already use, such as Claude Code, Cursor or Codex, and connect it to Steps. Your AI does the work, Steps gives it the agreed workflow, and your team decides what that workflow is.
Which AI tools does Steps work with?
Any MCP client that can connect to a remote server with an Authorization header, plus anything that can call our HTTP Agent API. There are setup guides for Claude Code, Cursor, Codex, VS Code, Claude Desktop, Gemini CLI, Windsurf, Zed, OpenCode and Goose on the Connect page.
How do I get our workflows into Steps without writing them from scratch?
Let your AI write the first draft. Record a voice note, film a quick video, or point your AI at a doc or chat thread, and ask it to write the workflow into Steps, branches and all. Your team reviews the draft and publishes it. Steps doesn’t transcribe audio or run a model itself. The AI you already use does the conversion. On paid plans you can also attach videos and PDFs to the steps they explain.
What can my AI change?
Only what your token allows. Can read is enough to find and follow workflows. Can edit can change the current published version. Those edits are live immediately. Edits to a draft are not live until someone makes it current. A new version is created only when someone explicitly starts one. Someone with publishing permission can publish, unpublish, or make a version current. Changing who can see a workflow needs that permission. Leaving Can publish off does not stop Can edit from changing the current version. Deleting a workflow or suite needs its exact name, typed by you, and your AI asks before it deletes anything.
How do agents with Steps compare with agents given the same documents?
Compared with agents given all the same documents in the prompt. 108 tasks each, same model, same grader.
They finished about as often (103 of 108 tasks, against 102 of 108). We don’t claim they pass more tasks — they spend less effort getting there.
About 43% lower median time. Under half as many tool calls. About half as many input tokens.
My AI’s video upload failed. What should I do?
Large uploads through your AI can time out, especially on slower connections. Ask your AI to split the video into short clips, one for each step, and attach each clip to its step. Short clips are easier for your team to follow anyway. If you’d rather keep a long video in one piece, upload it in the Steps web app instead. On the Free plan, uploads are JPEG images only, up to 25 MB.
Example prompt: "Split this video into one short clip per step, then attach each clip to its matching step in the workflow."