Every major AI model today can generate text.
Many can call tools.
Some can execute functions.
But very few can produce something you truly own.
That’s the problem I’ve been thinking about while building OpenVerb.
⸻
AI conversations disappear. Workflows shouldn’t.
Imagine spending an hour building an automation with an AI assistant.
Tomorrow you want to:
- run it again,
- edit it,
- review it,
- send it to a coworker,
- store it in Git,
- or execute it from another application.
Most AI platforms make this surprisingly difficult because the automation is trapped inside a conversation.
The workflow isn’t portable.
⸻
What if AI generated documents instead?
Instead of generating another block of text, what if AI generated a structured execution document?
That’s the idea behind .ov files.
A .ov file is a portable workflow that describes what should happen, not just what the AI said.
Think of it as the equivalent of:
- PDF for documents
- Markdown for notes
- Docker for containers
…but for AI execution.
⸻
Example
version: "1.0"
name: New Client Onboarding
actions:
- verb: crm.create_contact params: name: "John Doe" email: "john@example.com"
- verb: email.send params: to: "john@example.com" subject: "Welcome!" template: "welcome"
- verb: file.create_folder params: path: "/Clients/John Doe"
The AI doesn’t execute these actions.
It simply generates the workflow.
An OpenVerb-compatible runtime can then:
- validate it,
- collect missing inputs,
- enforce policies,
- execute each verb,
- produce receipts.
⸻
Why not just use function calling?
Function calling is useful, but it’s usually tied to a particular model, SDK, or provider.
A portable execution document separates:
- reasoning
- execution
The model decides what should happen.
The runtime decides how it happens.
That makes workflows reusable across applications instead of being locked inside one AI ecosystem.
⸻
Why this matters
Portable workflow files unlock possibilities that conversations can’t.
Imagine:
- Version-controlling automations with Git.
- Reviewing AI-generated workflows before execution.
- Sharing workflows with teammates.
- Publishing workflow libraries.
- Executing the same automation across multiple applications.
- Building marketplaces of reusable AI workflows.
Once workflows become files, they become assets.
⸻
The OpenVerb vision
OpenVerb isn’t trying to build another chatbot.
It’s an attempt to build an open execution layer for AI.
The long-term vision is an ecosystem where applications speak a common language of verbs and portable execution documents, allowing AI-generated workflows to move freely between compatible tools instead of remaining locked inside proprietary platforms.
The web became powerful because of open standards.
I believe AI automation needs them too.
⸻
I’m curious:
If AI could export every automation as a portable workflow file, what would you build with it?
Top comments (0)