> For the complete documentation index, see [llms.txt](https://developer.harness.io/llms.txt). Markdown versions of documentation pages are available by appending `.md` to page URLs; this page is available as [Markdown](https://developer.harness.io/release-orchestration/3.0/processes/ai-based-process-creation.md).

# AI-Based Process Creation

AI-based process creation allows you to generate release processes by describing them in natural language. The AI analyzes your description and creates a structured process with phases, activities, and dependencies.

### How It Works <a href="#how-it-works" id="how-it-works"></a>

Harness supports creation of processes using Harness AI, which is an AI-based approach. Most of the time, release processes are available in a textual fashion documented in different sources. Release Orchestration enables users to provide that process documentation and create the process as an entity in Harness Release Orchestration.

#### 1. Provide Process Documentation <a href="#id-1-provide-process-documentation" id="id-1-provide-process-documentation"></a>

Provide a natural language description or documentation of your release process. This can be:

* Textual documentation from existing sources
* Process descriptions from organizational documentation
* Multi-service release processes with planning, building, validation, deployment, and monitoring phases

**Example:**

```
"Multi-service release process starting from planning of the release up until building 
and doing other functions, different heterogeneous functions under one umbrella, using 
a process up until releasing and monitoring in production. The process includes:
- Release planning and coordination (Owner: Release Manager)
- Build and artifact creation
- Testing and validation
- Feature flag enablement
- Production deployment
- Monitoring and rollback"
```

#### 2. AI Analysis and Generation <a href="#id-2-ai-analysis-and-generation" id="id-2-ai-analysis-and-generation"></a>

Once you provide the prompt and ask the AI agent to create the process, it automatically:

#### 3. Process Visualization <a href="#id-3-process-visualization" id="id-3-process-visualization"></a>

The process is visualized in a graphical view, showing:

* All phases that have been created (from release planning and coordination up until rollback and documentation)
* Activities within each phase
* Dependencies between phases and activities
* Owner assignments

#### 4. Review and Save <a href="#id-4-review-and-save" id="id-4-review-and-save"></a>

After AI generation, you can:

* Review the generated phases and activities
* Verify owner assignments
* See a summary of what the process is enabling (modeling the entire process as an entity and enabling orchestration using activities)
* Save the process

### Best Practices for AI Process Creation <a href="#best-practices-for-ai-process-creation" id="best-practices-for-ai-process-creation"></a>

#### Be Specific <a href="#be-specific" id="be-specific"></a>

Provide detailed descriptions:

* **Recommended:** "Deploy to staging, run smoke tests, wait for QA approval"
* **Avoid:** "Deploy and test"

#### Include Dependencies <a href="#include-dependencies" id="include-dependencies"></a>

Mention what must happen before other steps:

* **Recommended:** "After deployment completes, run integration tests"
* **Avoid:** "Deploy and test"

#### Specify Activity Types <a href="#specify-activity-types" id="specify-activity-types"></a>

Indicate what should be automated vs manual:

* **Recommended:** "Automatically run unit tests, manually review security scan results"
* **Avoid:** "Run tests and review"

#### Include Approval Points <a href="#include-approval-points" id="include-approval-points"></a>

Mention where approvals are needed:

* **Recommended:** "Require production deployment approval from release manager"
* **Avoid:** "Deploy to production"

### Refining AI-Generated Processes <a href="#refining-ai-generated-processes" id="refining-ai-generated-processes"></a>

After the AI generates a process:

1. **Review the structure**: Ensure phases and activities make sense
2. **Check dependencies**: Verify execution order is correct
3. **Validate activities**: Confirm activity types and configurations
4. **Add details**: Enhance with specific configurations
5. **Save**: Save the process and start adding reusable activities

### Example Workflow <a href="#example-workflow" id="example-workflow"></a>

#### Input Description <a href="#input-description" id="input-description"></a>

```
"Multi-service release process:
1. Code freeze and branch creation
2. Build all services in parallel
3. Deploy to integration environment
4. Run integration tests
5. Deploy to staging
6. User acceptance testing with sign-off
7. Production deployment with approval
8. Post-deployment validation"
```

#### AI-Generated Process <a href="#ai-generated-process" id="ai-generated-process"></a>

The AI creates:

* **Phase 1: Preparation**
  * Code freeze activity
  * Branch creation activity
* **Phase 2: Build**
  * Build Service A (automated)
  * Build Service B (automated)
  * Build Service C (automated)
* **Phase 3: Integration Testing**
  * Deploy to integration (automated)
  * Run integration tests (automated)
* **Phase 4: Staging**
  * Deploy to staging (automated)
  * UAT (manual)
  * UAT sign-off (approval)
* **Phase 5: Production**
  * Production approval (approval)
  * Deploy to production (automated)
  * Post-deployment validation (automated)

### Limitations <a href="#limitations" id="limitations"></a>

AI-generated processes are a starting point:

* May require refinement for complex scenarios
* May not capture all organizational nuances
* Should be reviewed by subject matter experts
* May need customization for specific tools and integrations

### Related Topics <a href="#related-topics" id="related-topics"></a>

* [Process Modeling](/release-orchestration/use-release-orchestration/release-orchestration-processes/process-modeling.md)
* [Modeling complex processes](/release-orchestration/use-release-orchestration/release-orchestration-processes/process-modeling.md#modeling-complex-release-processes)