Committee document
The case for agent-ready pipelines, for the CIO
A one-page summary written to be forwarded as it stands. It is complete without Novuz, and the decision is left with the reader.
The decision
What is being decided?
Whether the warehouse migration should deliver pipelines that AI agents can operate under the company’s own controls, or leave that for a later project.
Why now
Why decide it now?
The pipelines are being rebuilt during the migration. Exposing their operations, permissions and logs is far easier during the rebuild than after cutover, when the code is untouched and the team has moved on.
Definition
What “agent-ready” means here
Each delivered pipeline can expose five operations as tools an agent can call: run, check status, retrieve lineage, inspect outputs and rerun a failed step. Each operation runs under the company’s identity provider, approvals and audit logging. Agents cannot change the converted logic.
What stays deterministic
What stays deterministic?
The conversion itself uses fixed, versioned rules: the same input gives the same output, and every object is reconciled against the original on the same inputs. A language model is used only for edge cases, on a model the company approves, and an engineer approves every proposal.
Risks and controls
The risks, and the controls
| Risk | Control |
|---|---|
| An agent reruns a regulated step wrongly | An approval before reruns on regulated pipelines; the audit log; reconciliation against the last good run |
| An agent reaches data it should not | Only the exposed operations are reachable, under the agent’s own identity |
| The agent layer changes the numbers | Agents operate pipelines; they do not edit converted code |
The ask
The ask
Include agent-ready delivery in the scope of the First Domain, test it there with read operations first, and decide on write operations after a month of logs.
Below the document
Where to read next
Print or save this page as the one-pager. The detail behind it is on the governance page; the assessment is the first step.
Last reviewed:
Find out what your logic debt is costing you.
Every object counted, lineage, dead code and a fixed price in three business days. $3,000, credited.
Not sure which door? The scorecard takes ten minutes and tells you.