Platform / Configurable workflows
02 — WORKFLOWS

AI agents work within a predictable workflow.

Every Altabric pathway — Design, UAT preparation, Change readiness, Analysis — is a declarative definition. It sets out who runs each step, what they produce, who reviews it, and where the human approval gate sits. Agents follow it. You see it.

WHAT IT DOES

Predictable orchestration. Configured for you.

The standard way of using AI today: open a chat, prompt the agent, copy the output, prompt again, do the review yourself. That works fine for one operator on one task.

It breaks the moment the project scales — more people, more workstreams, more delivery teams, more documents, more sequenced steps that depend on one another. Output fragments across chat windows. No one sees the whole. Quality drifts because review is ad-hoc. There is no record of who produced what, against what input, under whose approval.

Altabric replaces that with a defined workflow. The workflow names the agent who runs each step, the deliverable that step produces, the reviewer who assesses it, and the human approval gate. Agents follow the workflow. The workflow doesn't follow the agents.

A WORKFLOW, WALKED THROUGH

Progress along the pathway by completing predetermined workflows.

Below is the workflow that runs at Theme level on the Design pathway. You approve the Theme, an Owner agent produces the deliverable and decomposes the Theme into Epics, a Reviewer agent assesses, you approve the result, and child work items begin their own workflows.

HUMAN USER OWNER (agent) REVIEWER (agent) T1 — APPROVE THEME T2 — PRODUCE T2a — REVIEW T3 — YOUR APPROVAL Approve the Theme human gate Draft the PID decompose into Epics Review decomposition approve / redo / reject Approve the PID final human gate redo — capped at one CHILDREN CREATED Each Epic begins its own workflow

Producer, reviewer, human in the loop for approval. The same rhythm at every level — Theme, Epic, Story, Solution Design — and on every pathway.

UNDER THE HOOD

A workflow guided by a predetermined data structure.

Below is the step from the workflow above where the Owner agent produces the PID and decomposes the Theme into Epics. Nothing is hidden behind code.

# w1-theme.yaml — step T2 T2: name: Draft PID & decompose into Epics actor: OWNER archetype: producer_seed action: agent_execution output: type: document format: html creates_children: target_level: L2 target_tag: epic initial_status: new task_template: > Produce the Project Initiation Document for this Theme. Decompose into Epics — each a coherent unit of design work. next: T2a
ACTOR
The agent that runs this step. OWNER resolves at runtime to whichever role you've configured as accountable on this work item — CTO, Test Manager, whoever fits your delivery.
OUTPUT
What this step produces — a document, in HTML. The producer step always emits a real deliverable, not just a verdict.
CREATES CHILDREN
When you approve at T3, the engine creates the child work items the agent identified — Epics, in this case. Each one begins its own workflow.
NEXT
Where the workflow goes after this step — to the reviewer, T2a. The shape is the same at every step: a producer, a reviewer, your approval gate.

Every step in every workflow has the same shape. The agent gets a task, produces a deliverable, the reviewer assesses, you approve. Predictable in form. Configured in content.

TALK TO US

The workflow orchestrates the agents. Each agent focuses on its task.

A demo runs against your delivery shape, not a generic one. We read every message. You will hear back within two working days.

Request a demo