The increasing sophistication of Large Language Models (LLMs) has opened new avenues for enterprise teams to automate processes, particularly through AI agents. These agents, capable of autonomously completing complex tasks, require a robust and flexible framework to guide their actions. The Cutover platform's task model, used by enterprise technology operations teams to automate their runbooks or processes, is designed for orchestrating complex sets of tasks and presents itself as a promising foundation for building such agentic AI systems.
The need for a runbook task model
While LLMs can be impressive in their ability to generate human-quality text, their effectiveness in executing complex tasks over a series of connected steps is poor. A runbook task model, on the other hand, provides a structured representation of a process, defining the steps involved, their dependencies, and the data that needs to be passed between them. This enables the structured use of LLMs to undertake decision making within the network of tasks in the process or runbook.
How Cutover's runbook task model creates a strong foundation for agentic AI
Cutover's runbook task model, based on a directed graph, offers several key features that make it well-suited for building agentic AI solutions:
- Directed graph capability: The directed graph structure allows for the clear definition of task dependencies and sequencing. This ensures that tasks are executed in the correct order and that necessary prerequisites are met.
- Data passing: The task model facilitates the transfer of data between tasks, enabling a smooth flow of information throughout the process and offering huge support for specific decision making at specific steps by the agent.
- Visualization and audit logs: The ability to visualize the state of the process and access an audit trail is crucial for understanding the system's behavior and troubleshooting issues. It is also the source of very valuable and hard-won training tokens for the agentic system to get better and better.
- Iterative creation and execution: Cutover's runbook task model supports iterative development and execution, allowing for the continuous refinement and improvement of the process.
- Human tasks and non-AI automations: The model can accommodate both human tasks and non-AI automations, providing flexibility and enabling the use of existing tools and systems.
Building agentic AI solutions with Cutover
To illustrate how Cutover can be used to build agentic solutions, consider the example of a "general application recovery daemon" (GARD) used in the case of an incident in technology operations in the enterprise.
Step 1: Trigger and initialization: An external monitoring system detects a failure in Application “X” and sends a payload to Cutover. Cutover instantiates a general application recovery template and represents the payload as task 1.
Step 2: Intelligent task selection: The initial "thinking" task ingests the payload and, using its understanding of the problem and the available recovery template options, selects the appropriate recovery snippet (e.g., load balancer repair). It populates the snippet with the necessary data and triggers its execution.
Step 3: Recovery execution: The recovery snippet executes, restoring a new load balancer instance to a functional state that begins receiving application request traffic.
Step 4: Evaluation and iteration: A second "thinking" task evaluates the recovery outcome. If successful, the process concludes with notifications to the team. If not, the LLM can explore alternative strategies or escalate the issue.
In the example flow, all of the processing is visible in live execution. It also creates an audit trail for any forensics and builds the training tokens to fine-tune the next periodic update to the core LLM.
Cutover plus agentic AI benefits
Cutover’s strengths lie in its runbook task model architecture, making it a valuable platform for supporting agentic AI solutions. Here are some key benefits:
- Structured workflow for LLMs: Cutover provides a clear process roadmap for AI agents by defining tasks, dependencies, and data flow. This structure overcomes the limitations of LLMs that struggle with complex processes without clear guidance.
- Flexibility and keeping peoplehuman-in -the -loop: The model allows for incorporating human actors alongside AI tasks. This flexibility is crucial for real-world scenarios where human intervention or expertise might be necessary.
- Rich training data generation: Cutover's execution logs and audit trails provide valuable data for training LLMs. This data captures real-world process execution, failures, and successes, leading to continuous improvement of the AI agent's decision-making capabilities.
- Improved efficiency and scalability: By automating repetitive tasks and managing complex workflows, Cutover frees up human resources and allows AI agents to handle increased workloads efficiently.
- Transparency and auditability: Cutover's visual representation of the process flow and detailed audit trails provide clear visibility into the actions of agentic AI systems. This transparency builds trust and facilitates troubleshooting if issues arise.
- Iterative learning and improvement: Cutover's ability to handle the iterative creation and execution of tasks is ideal for agentic AI. The AI agent can learn from each process execution, refine its approach within Cutover's framework, and continuously improve its performance over time.
In summary, Cutover acts as a bridge between the power of LLMs and the complexities of real-world IT operations processes. By providing a structured approach to process definition, data management, and execution, Cutover empowers organizations to leverage the capabilities of LLMs while ensuring control and transparency. As AI continues to evolve, Cutover's role in enabling intelligent automation will become increasingly important.