Conversations are where work happens. Give them a process.
Every day, people and AI agents do real work through conversations. They draft, debug, analyse, plan. The conversation is the execution environment; it already is. But without structure, conversations are invisible, inconsistent, and ungovernable. This is the problem ConvOps exists to solve.
ConvOps (Conversational Operations) is the practice of driving work through conversations that follow a defined workflow. The orchestrator drives the conversation. The conversation is where the work happens. The workflow defines how it should be done.
chatbots structure conversations to answer questions. convops structures them to get work done.
the nine principles
The orchestrator drives, the workflow guides
The AI agent the human talks to drives the conversation. It reads the current step, executes or delegates, and advances. The workflow does not take over: it surfaces the current step's instructions inside the conversation.
Conversations without process are chaos
Unstructured conversations produce inconsistent results. A conversation that follows a defined workflow, step by step, produces output that is consistent, repeatable, and auditable. Every time.
The step surfaces inside the conversation
Each step appears inside the conversation as a briefing: what to do, what standards to follow, what to produce. Only the current step. Not the whole process at once.
Gates define where humans are required
A gate is a point where the process defines whether a human must be involved: approval, review, sign-off. When a gate exists, the conversation pauses until the human acts.
Any agent can connect
ConvOps is not tied to one provider, model, or tool. Any AI client can connect, read the current step, and execute. The workflow is the contract. The agent is interchangeable.
Conversations are resumable
Every step records what was done and what comes next, so anyone, human or agent, picks up where the last one left off. Context is never lost. Conversations never start from zero.
The workflow is the audit trail
Every step records who did the work, what was produced, and what was decided. The conversation that executed the work is the record of the work. Not reconstructed after the fact.
Process knowledge is encoded, not tribal
Company standards and best practices live in workflow step instructions, not in someone's head. The process is documented, executable, and transferable. People leave. Process stays.
The workflow replaces the prompt dump
Entire processes are crammed into system prompts today: pages of instructions, no sequencing, no checkpoints. A workflow delivers only the instructions needed right now.
what it replaces
| before | with convops |
|---|---|
| Employee opens ChatGPT, asks random questions | A conversation the workflow guides |
| Agent gets a prompt dump with the entire process | The current step surfaces with clear instructions |
| Every person does it differently | Every conversation follows the same process |
| No audit trail of AI-assisted work | Steps record who did what, and when |
| Work cannot be resumed by someone else | Anyone picks up at the current step |
| AI usage is ungoverned | Gates define where humans are required |
convops is not
- Not a chatbot framework.Chatbots structure conversations to answer questions. ConvOps structures conversations to follow a process.
- Not another agentic workflow.Agentic workflows are pipelines where agents pass output forward. ConvOps embeds the workflow inside a living conversation with full context.
- Not a workflow automation tool.No visual canvas, no drag-and-drop pipeline. The conversation is the execution environment.
- Not an agent framework.ConvOps does not build agents. It gives any agent, or human, a process to follow.
- Not project management with AI bolted on.Conversation-first, built for how work actually happens now.
The canonical manifesto, v1.0, is maintained in the open and welcomes signatories.