Technology & governance
Humans review and approve. Agents do the rest.
Agents prepare the work. Your team keeps authority over the outcome. Our governance approach starts with clear responsibilities, defined boundaries and evidence you can review.
Prepare · Review · Approve · Record
Human in the loop
- Agent
- Human
- System of record
Bank reconciliation · Month-end
- Agent
Prepare the work end to end, with supporting evidence and exceptions ready for review.
- Human
Your subject-matter experts check the work against the workflow, rules and business context.
- Human
An authorised person approves the result before the workflow proceeds. Unresolved exceptions return for correction.
- Platform
Keep a full audit trail of agent actions, human decisions and resulting changes inside the platform.
Full audit trail
Step 4 of 4
Audit trail
- Step 1AgentPrepare the work end to end, with supporting evidence and exceptions ready for review.
- Step 2Human · SME teamYour subject-matter experts check the work against the workflow, rules and business context.
- Step 3Human · SME teamAn authorised person approves the result before the workflow proceeds. Unresolved exceptions return for correction.
- Step 4Platform · system of recordKeep a full audit trail of agent actions, human decisions and resulting changes inside the platform.
Agents prepare the work; the client's subject-matter experts review and approve it; every action is recorded in the platform with a full audit trail.
The audit trail should identify who did what, when, which version ran and what was approved. Protect logs against alteration, restrict access and agree retention periods. Capture the evidence needed to reconstruct decisions without copying unnecessary sensitive data.
Who does what
Five actors around every agent
- AClient or consultant
Subject expert
Defines the workflow, rules and edge cases.
- BDaVero AI pod
Forward Deployed Consultant
Builds the agent around those requirements, with the Solutions Engineer.
- CClient or consultant
Validator
Checks controls and approves results in the platform.
- DDaVero AI
Operator
Runs the agent, monitors exceptions and coordinates resolution.
- EDaVero AI
Owner
Owns the agent and its ongoing improvement.
The client defines and approves. DaVero AI owns and runs the agents.
Constitution
Hardened with a constitution
Rules, controls and edge cases that make the skill reliable.
The constitution defines permitted actions, required approvals and when to stop and escalate.
Frameworks we align with: This approach is designed to align with the NIST Artificial Intelligence Risk Management Framework (AI RMF)1, ISO/IEC 420012 and SOC 2 practices3.
- 01
Access control
Apply least privilege: give each person and agent only the access needed for its task. Separate preparation from approval; review and revoke permissions as responsibilities change.
- 02
Monitoring
Check model and agent performance against agreed acceptance criteria. Track errors, unexpected behaviour and changes in output quality; route exceptions to a named person.
- 03
Change control
Version prompts, models, tools and workflow rules. Test changes, document review and approval, and keep a rollback path before release.
- 04
Incident response
Define how to pause an agent, contain an issue, preserve evidence and notify responsible people. Investigate the cause and validate corrections before restarting.
Platform-native
Work inside your platform. Keep data within agreed boundaries.
The design aim is for agents to run inside your platform and for business data to stay there. Confirm any model-provider or integration data flows before deployment. Limit collection to the task, define approved uses, protect sensitive fields and agree retention and deletion rules.
Start a conversation
Bring a workflow. Define the approvals, data boundaries and evidence it needs.