Skip to main content

The approach

There are three ways
to bring in agents.

Two of them rarely make it past the pilot.

The differences go deeper than vendor quality. Each approach starts somewhere different and ends somewhere different. The first digitizes the old process. The second adds a chat window on top of it. The third starts with one defined task, builds the agent around it, and grows only when the work calls for more.

DimensionTraditional ITAI Bolting✓ The Amralab Ecosystem
Starting pointExisting process, digitizedExisting process with a chat layerOne task, clearly defined
Data strategyData as a byproductData fed in, uncheckedOnly the data the task needs, by permission
ControlReviewed after go-liveLeft to the modelAccess and review agreed upfront
TimelineLong program, late feedbackFast demo, then stallsOne task first, then build on it
OutcomeDigitized chaosAn impressive demo nobody usesWork that gets done and reviewed

The mechanism

How one taskleads to the next.

01

A Defined Task

Every agent starts from a written task: the outcome, the tools, and the review points. That definition is what makes the work measurable.

02

Better Context

Doing the task shows what information it depends on. O’Brain keeps that context available to the agents that need it.

03

Measured Results

Each run is checked against the agreed criteria. Where a person stepped in, the reason is recorded.

04

The Next Task

What the first task teaches shapes the second. Teams add tasks, or bring in O’force, once the first one holds up.

Results depend on the task and the deployment.

Results depend on the task and the deployment, so each step is reviewed before the next one begins.

How it becomes real

Three phases.Each one reviewed with you.

Each phase builds on the one before. Defining the task shows which of the next phases you need, and when.

Phase 01 / 03

Phase 01 — Define

Agent O

We define the task before we build. Always.

Every engagement starts with one piece of work your team does today. Together we agree what the task is, what done means, which tools the agent may use, and where a person reviews or approves.

The output is a task definition: a short working document with the outcome, the success criteria, the tools in scope, and the review points. Agent O is built and measured against it.

One task · Clear criteria · Scoped to your use case

Discuss a use case

The Task

What needs doing, who does it today, how often it comes around, and what a finished result looks like.

The Context

What information the work depends on, where it is kept, and what an agent would need to see to do it well.

The Tools

Which systems the agent may use, with what permissions, and which it should never touch.

The Oversight

Where a person reviews, corrects, approves, or takes over, agreed before the first run.

Phase 02 / 03

Phase 02 — Build

Amralab & O’Brain

The agent, shaped around the work.

Amralab creates and trains the agent for the defined task, and O’Brain provides the reasoning and context behind its decisions. Each agent is shaped around your workflow instead of a generic template.

Every agent is tested on live examples of the task before your team relies on it, and results are checked against the criteria from Phase 01. You see the agent working on your material early in the build.

Trained on the task · Tested before handover · Reviewed against criteria

See example workflows

Agent Creation

An agent built for one defined role, task, or workflow, with only the tools it needs.

Agent Training

Trained on the task, the context, and the rules that should guide its decisions.

Evaluation

Did the task get done, did the handoffs hold, and where did a person need to step in?

Handover

Clear documentation of what the agent does, what it can access, and how to review it.

Phase 03 / 03

Phase 03 — Connect

O’force & Oxygen

From one agent to connected work.

When the work calls for different specialties, O’force coordinates specialized agents toward one agreed outcome. Oxygen brings agents, intelligence, and workflows into an operating environment shaped around how your business runs.

Most teams begin with one agent on one task. Connection comes once the first task is working and the next one is clear. Agent O will be revealed on 26 December 2026, and the waitlist is open now.

Grows with the work · Waitlist open · Meet O 26 DEC 2026

Meet O

Agent O

One agent on one defined task, with its own tools and review points.

O’force

Specialized agents passing work between them toward a shared outcome.

O’Brain

Shared reasoning and context across every agent in the deployment.

Oxygen

The environment that links agents to your approvals, systems, and teams.

Control & oversight

We decide accessbefore anything runs.

Agents act inside your systems, so control has to be planned from the start. These four principles apply to every engagement, and each one is agreed with you in writing.

Limited Access

Permissions

An agent can use only the systems and tools it is given. Permissions are set per task, reviewed with your team, and changed only when you decide. Nothing is connected by default.

Human Review

Oversight

Every engagement agrees where a person reviews, corrects, approves, or takes over. Those points are written into the task definition from day one.

Data Handling

Privacy

Agents see only the data the task needs. How that data is handled, stored, and returned is agreed for each engagement and set out in writing before work begins.

Recorded Runs

Accountability

Each run is recorded so it can be reviewed: what the agent was asked, what it did, and what it returned. If something goes wrong, you can see where it happened and why.

Common questions

The approach, explained.

In three phases. We define one task with you, create and train an agent for it, and then connect it to your workflows once the first task is working.

A single task can be defined, measured, and checked. It also shows what data, tools, and review points the work needs before anything larger is built.

A short task definition covering the outcome, success criteria, tools in scope, and review points. The agent is built and evaluated against it.

If the task is already well defined, the Define phase is short. We still write it down, because an agent cannot be measured against a goal nobody has agreed.

Each agent gets only the access its task needs, review points are agreed before the first run, and every run is recorded. Data handling is set out in writing for each engagement.

Agent O will be revealed on 26 December 2026. The waitlist is open now, and you can discuss a use case with us before then.

Start with one task.

Tell us about a piece of work your team repeats every week. We will tell you honestly whether an agent is a good fit.

How Amralab Works: From One Task to Connected Workflows | Amralab