Build · Process owners and finance systems

Build a Governed Agent

Learn how to turn a suitable governed workflow into a scheduled, logged agent: a connection instead of an upload, deterministic rules as code, bounded AI assistance, a review gate, and a reproducible operating pattern. The group builds a working prototype in the workshop and learns how to evaluate what would be required before organizational use.

AI handles defined preparation. A person reviews. Your controls remain in place.

2.5-hour workshop · up to 8 · your office or online · $4,500

The course

About this course

This is the technical follow-on to The Governed Workflow. A small group selects one suitable process—often a reconciliation, load validation, missing-mapping proposal, or close-calendar digest—and uses it to learn the architecture of a governed agent.

In a guided build, participants connect a sanitized source, put deterministic rules into code with tests, use AI for bounded exception handling or drafting, configure a review gate, create a log, add a schedule, and document the result in a README. A second participant follows the documentation to expose missing assumptions and unclear operating steps.

The workshop artifact is a working prototype created for learning and evaluation. The transferable outcome is the team’s ability to apply the architecture to other suitable processes. This course runs on Claude Code.

Who it is for
Process owners and finance systems, reporting, and data people: the people who would own a governed process
Format
Hands-on 2.5-hour workshop for a small group. Participants build and test one guided prototype, document it, and learn the pattern they can apply to additional suitable processes.

What the team learns and practises

Outcomes

  • Draw the operating boundary for a governed agent: permitted inputs, outputs, review, logging, and prohibited actions
  • Connect a sanitized source without pasting data manually
  • Put deterministic calculations, joins, thresholds, and validation rules into code with tests
  • Use AI selectively for exception classification or narrative drafting
  • Configure and exercise a review gate, run log, and schedule in a workshop environment
  • Write a README and rules file that expose assumptions and allow a colleague to reproduce the exercise
  • Evaluate gaps that must be addressed before organizational or production use
  • Reuse the learned architecture when evaluating additional suitable processes

Outline · 2.5 hours

How the session runs.

Usually 9:30 to 12:00 or 2:00 to 4:30. Other times are fine.

  1. 0115 min

    Pick the process and draw the boundary

    • One suitable process, or the practice process provided. Inputs, outputs, who approves, where it logs.
    • The read-only boundary: what the agent may touch and what it may only propose.
  2. 0230 min

    A connection instead of an upload

    • Pointing the tool at a folder, an export, or a query instead of pasting.
    • The project rules file and skills, saved instructions the agent reuses, that make it behave like a colleague who knows your platform.
    • Where it is reliable and where it is confidently wrong.
  3. 0335 min

    Rules as code

    • The deterministic core: sums, joins, as-of logic, thresholds, written as a script with tests.
    • The AI layer on top: read the exceptions, classify timing versus error, draft the note.
    • Generating the test cases nobody had time to write.
  4. 0440 min

    The approval gate, the log, the schedule

    • In the workshop pattern, output lands in a review folder or draft email rather than a system of record.
    • Configure and inspect a run log that records the actions and results of the workshop prototype.
    • Scheduling it, and the morning digest that says what ran and what failed.
  5. 0530 min

    Can someone else reproduce the exercise?

    • Swap seats. A second person follows the README and runs the prototype in the workshop environment.
    • What to fix so they can. Conventions: where the rules live, what gets committed.
    • Scoping a second process with the same pattern.

Bring

What to bring

  • A laptop with a terminal and a Claude login your organization allows for work; this course runs on Claude Code. Setup instructions come ahead of the course.
  • Sanitized copies of the inputs for one suitable process, or use the practice data provided.

If IT blocks the tool, tell us on the call and we plan around it.

About your data

About your data

Courses use practice files that resemble real Finance work, so confidential data is not required to learn the method. A participant may use a sanitized file—or a company file only when the organization’s policy and approved AI environment permit it.

The method makes the input, transformation, expected output, and human review explicit. It does not require an unrestricted agent inside company systems, and it does not replace the organization’s technology, privacy, security, records-management, data-residency, or approval requirements.

Where a workflow could change a system of record, nothing posts without authorized human approval.

Instructor

Neil Perejma

Founder, 1113 Technologies and EPM ONE Consulting

Neil Perejma has advised Finance and operations leaders for more than 20 years on the processes, reporting, and controls their organizations depend on, and every engagement has left the team able to carry the work forward.

Today his clients are asking how to bring AI into Finance work without weakening governance. His courses give Finance teams a practical way to do it, keeping AI inside the controls they already have.

More about Neil

Workflow examples

Recurring Finance work this course is built around.

The others

Other courses

Questions about training your team?

Book a 20-minute call with Neil Perejma, on Zoom or by phone.

Book a call