Human-in-the-Loop (HITL)

our agents currently do not operate with complete autonomy, which mandates the integration of human in the loop (hitl) processes during their creation the hitl feature in bridge is crucial for enhancing control and mitigating risks by requiring human approval for all critical ai actions this allows for the involvement of multiple participants in a single workflow if a user fails to respond to a hitl request within the established timeframe, the system will automatically deny it, ensuring a predictable timeout behavior that is thoroughly logged for transparency agents will pause for user review in the bridge ai assist ui, where decisions can be either approved or denied if there is no response within the specified timeframe, the decision will be automatically denied this procedure will continue until full (poa) is established agents involved in hitl steps will automatically defer requests if there is no action taken within a 30 minute window additionally, all processes must be completed within a two hour timeframe this aims to prevent system overload and ensures that workflows remain streamlined, even when multiple approval layers are involved cd process for ea review readiness bridge agentic ai offers templates for developers to build their agents using the enhanced bridge sdk, which includes features like knowledge bases and connectors while deploying an agent is straightforward, it's crucial to ensure that only valuable agents are launched to avoid clutter from unused agents, we've established checks in our continuous deployment (cd) process developers must submit a pull request (pr) with five key components deployment requires two layers of approval prior to advancement the first is a business approval, which is currently carried out manually a panel of two business executives reviews each request, accompanied by a succinct high level description (no more than five sentences) that conveys the asset's intent the second layer is a technical approval aimed at evaluating whether the code adheres to best practices, maintains security, and is free from issues such as hardcoding the technical team, under the leadership of architects, is responsible for endorsing the code technical / business analysis what is the value to kyndryl teams? describe how this agent helps address the business problem what is the value to the client? explain how to execute this automation/solution and the benefit of running this solution what kind of resource utilization will this solution have? resource utilization on the endpoint server does this solution make any changes to the target server? does the solution make changes to the system? does it reboot? is there any manual approval in play (human in the loop) before making such a change? is the agent deterministic or strictly rule ‑ based in its execution? describe the controls that are in place to prevent hallucination or unintended actions explain whether the solution ensures predictable, auditable, and safe behavior once both approvals are in place, the cd process can begin this means the code can move from development to production, ensuring that each deployed agent is necessary and beneficial for the business