06 / AI agent governance

Bring in AI agents.
Keep permissions, review and evidence.

Place AI-assisted work within project context. Manage providers, agents, tools and policies, then review attempts and outcomes before accepting them.

FUNCTIONAL SCOPE

What you can manage with Orkenty.

Context and executions

Distinguish jobs from execution attempts. Preserve the context used and relate results to their source work and project.

Tools, skills and MCP

Organise tools and skills in governed registries. MCP integrates tools and sessions according to configuration and effective permissions.

Review and QA

Relate AI outcomes to review and tests. A completed result does not automatically mean accepted code or a released delivery.

Consumption and controls

Read reported consumption, provenance and available estimates. Package, pause and concurrency controls depend on installation and enablement.

From a prompt to work with context

A job defines work; each run represents an attempt. Project sources and snapshots preserve the context used for that attempt. Results can be reviewed alongside their provider and state.

Persistent memory by scope supplies context where configured. Its usage and provenance must be distinguished from a current interpretation of the project. A new attempt does not remove the need to review evidence from earlier attempts.

Capabilities with usage rules

Tool and skill registries describe capabilities, contracts, scope and governance. Policies and permissions determine authorised use in the corresponding environment. MCP integrations connect tools within that framework.

Agent teams can distribute phases and responsibilities across profiles. Coordination and automatic advancement are opt-in capabilities subject to configuration and review. They do not provide universal authorisation to execute tools or publish changes.

Accept outcomes with evidence

Review and QA compare the result with the request. AI packages group stories and context and apply acceptance controls according to installed capabilities. Pausing or resuming a package is different from accepting a story result.

Reported consumption and configured rates support usage and estimate review. Missing metrics or incompatible currencies must not be treated as zero cost. Budgets are informational, not invoices or automatic spending caps. Autonomous execution with Codex remains under laboratory validation.

A PRACTICAL JOURNEY

From setup to team review.

01

Define work and authorised context.

02

Select a provider, agent and permitted capabilities.

03

Review each attempt, result and reported consumption.

04

Check review and QA before accepting or advancing.

ILLUSTRATIVE USE CASE

Put the capability in context.

A story includes acceptance criteria and project sources. An agent produces a result in an identifiable attempt. The team reviews it and checks QA before acceptance; uncertainty is not converted into success merely because information is missing.

Available actions depend on installed version, enabled integrations and current permissions. Review your specific journey in a demonstration. The example is illustrative and is not a customer testimonial.

COMMON QUESTIONS

Understand the scope.

Does AI act without supervision?

Configuration determines enabled capabilities. Permissions, review and controls remain part of the process; universal autonomy is not promised.

Does MCP remove the need for permissions?

No. MCP is an integration mechanism; tools and sessions remain subject to configured governance and scope.

Is reported consumption an invoice?

No. It describes usage and, where applicable, an estimate with its provenance and rate. It differs from the provider’s actual charges.

A CONVERSATION ABOUT YOUR ENGINEERING

See Orkenty in the context
of your team.

Tell us what you need to connect. We will prepare a demonstration focused on your requirements, quality processes and delivery workflow.

01 · Review your process and priorities.
02 · Explore a functional workflow.
03 · Define the scope you need to evaluate.

Business contact:
info@orkenty.com

Privacy inquiries: jorge.bernal@hova-it.com · Hova-IT