Regisseur · Product Walkthrough

From a blank canvas to a running case — built by the person who owns the process.

Follow Renata Kessler, an Operations Lead at Woodgrove Life, as she draws a new workflow, builds and tests an AI agent for it, connects the pieces together, and then watches a real case run through it — start to finish, in one sitting.

Woodgrove Life — a Regisseur demo workspace 19 screens, all captured live from a real build-and-run session

Workflow steps, drawn from scratch

New AI agent, built and simulated

Real case, run end to end

Manager approval, by a real person

Screens captured live

Renata Kessler

Operations Lead at Woodgrove Life. Draws the workflow, builds the agent, and turns the design into a running case — no engineering team required.

Operations Lead / Builder

Owen Marsh

Team Manager. Reviews the AI's risk assessment and gives the final approval before a beneficiary change takes effect.

Team Manager

The builder's journey — design to a real case, in one sitting

Draw the graph

4 nodes, verified clean

→

Build the agent

AI-assisted prompt, tested in simulation

→

Wire it together

Agent + form bound to the graph

→

Run a real case

Agent runs, manager approves, case completes

About this walkthrough

Every frame in this walkthrough is a real screenshot from one continuous session against a live Regisseur environment — the workflow, agent, form, and case shown were all created during the session you're watching, and the case advanced because the platform ran it, not because anything was staged. The AI prompt drafting, the simulation, and the agent's assessment of the submitted request are live model calls. The intake form was onboarded the way it works in production: a PDF form document uploaded, its fields extracted, and mapped — the document itself was written for the fictional Woodgrove Life. The walkthrough narrates a small cast (Renata, Owen); on screen you'll see the single demo operator account that genuinely performed every step. Saving a workflow makes it immediately available to run — there is no separate publish step for workflows — while agents are published as versioned snapshots, and the running case stays pinned to the versions it started with (final chapter).

Chapter 1

A blank canvas

Before any AI or automation exists, there is an empty template with a name.

1 Start

Where every workflow starts

Renata Kessler, an Operations Lead at Woodgrove Life, opens the workflow library — every process the company runs lives here side by side, ready to reuse, adjust, or build from nothing.

2 Start

Naming the new process

She starts a brand-new workflow: "Beneficiary Change Request" — the process for when a policyholder asks Woodgrove Life to change who receives their benefit.

3 Start

A blank canvas

A blank canvas. Nothing is pre-built for her — Renata will draw the whole process herself, one step at a time.

Chapter 2

Drawing the workflow

The whole shape of the process, drawn node by node and checked before it can run.

4 Graph

The first step: intake

The first node goes down: Beneficiary Change Intake, the point where a policyholder’s request enters the workflow.

5 Graph

Four steps, drawn end to end

The whole process now has a shape: Intake → AI Risk Assessment → Manager Approval → Change Effective. Renata drew all four steps and the connections between them before writing a line of logic.

6 Graph

The platform checks its own work

Before anything can run, Regisseur verifies the graph on its own — every node reachable, every path resolved, nothing left dangling. A clean pass, with no separate manual audit needed to catch a broken workflow.

Chapter 3

Building the agent

An AI agent, drafted with AI assistance and proven safe in simulation before it touches a real case.

7 Agent

A new AI agent, from scratch

Renata creates "Beneficiary Change Assessor" — the AI agent that will handle the judgment work at the Risk Assessment step.

8 Agent

Drafting the agent’s judgment, with AI help

Rather than write the agent’s instructions from a blank page, Renata describes what she needs in plain English and lets AI draft the first version — she reviews and refines it before it goes near a real request.

9 Agent

Defining what goes in, and what comes out

The agent’s input (the requested change) and output (a risk rating and its reasoning) are defined as structured schemas, so its judgment always comes back in a shape the rest of the workflow can rely on.

10 Agent

A deterministic, inspectable pipeline

The agent’s actual behavior is a sequence of concrete steps — read the request, call the AI model for a risk judgment, write the result to the case. Nothing hidden inside a black box; every step can be reviewed and tested on its own.

11 Agent

Testing it before it ever touches a real case

Renata runs the agent in simulation with sample input — the AI step executes for real, but no case is touched and nothing is saved. She can see exactly how it behaves before trusting it with a real request.

Chapter 4

Wiring it together

The graph, the agent, and the intake form connect into one workflow — each stays an independent, reusable piece.

12 Wiring

Assigning the agent to its step

Back on the graph, the Risk Assessment node is wired to the Beneficiary Change Assessor agent she just built — the shape of the process and the AI judgment come together.

13 Wiring

Connecting the intake form

The Intake node is bound to "Woodgrove Life — Beneficiary Change Intake" — a real form Renata onboarded minutes earlier by uploading the PDF Woodgrove already uses and mapping its fields to the case record. The form and the node stayed independent pieces right up to this moment, joined here at design time — not hard-coded into each other.

Chapter 5

From design to running case

The moment a design becomes a real case — and the platform runs it, end to end.

14 Run

From design to a real case

With the workflow saved, Renata creates a real case from it — the moment a template stops being a design and starts being work in progress.

15 Run

The case, running

The case is live. Intake is underway, and the rest of the workflow — exactly as she drew it — is already waiting behind it.

16 Run

The agent’s judgment, held for human eyes

The agent has read the submitted request and produced its real assessment — a risk level, a recommended action, and its reasoning. Because Renata set this agent to always be reviewed, its work waits here for a person before the case moves on.

17 Run

The agent does its job, for real

The Beneficiary Change Assessor runs its live risk assessment. This is the same model call and the same pipeline Renata built minutes earlier — now executing for real, against a real case.

18 Run

A person makes the final call

The AI’s assessment reaches the Manager Approval step. It informs the decision — Owen Marsh, the team manager, reviews it and approves before the change is allowed to take effect.

19 Run

The change takes effect

With the manager’s approval, Change Effective completes and the case closes. The whole process Renata drew, built, and tested minutes earlier just ran a real request, start to finish.