01 · AI Agent Factory

Strategy

Decide on evidence, not assumption.

We interview the whole operation, map the AS-IS process, design the TO-BE, and return three to five initiatives with business cases and sequence. The deliverable is fixed regardless of the surface examined.

What it is

We interview the whole operation, map the AS-IS process, design the TO-BE, and return three to five initiatives with business cases and sequence. The deliverable is fixed regardless of the surface examined.

What it is not

It is not a deck of recommendations with no evidence. If an existing strategy is sound, we say so and do not sell the engagement.

The four layers

Every layer is a legitimate endpoint.

You can stop at any layer with no obligation to contract the next.

EntryValidation$18–28k · 3–4 weeks

Independent verification of an existing strategy, architecture, or data foundation, with the commitment to say it is sound if it is. Obligation of means.

ScopedDiagnostic$40–100k · 6–8 weeks

The whole operation interviewed, AS-IS mapped, TO-BE designed, and three to five initiatives returned with business cases and sequence.

RecurringAdvisoryCommitted monthly hours

The portfolio is kept live and re-scored as conditions change, rather than re-commissioned each time the business moves.

ProductAI Agent Factory$150/mo per module

Consulting output becomes working agents, with tooling schemas and guardrails configured and packaged, which the client pushes to their own production environments.

Reference prices in US dollars, before tax. Confirmed in the proposal.

Layer by layer

What the work looks like, step by step.

Entry

Validation

We verify what you already decided, and we say it straight.

$18–28k · 3–4 weeks
  1. 01

    We read what already exists, without interviewing anyone.

    We receive the material and read all of it before speaking to anyone, so we arrive with questions rather than a questionnaire.

    Days 1 to 3
  2. 02

    We test every assumption against evidence.

    Every claim in the strategy is set against the data that would support it. Anything unsupported is marked as an assumption, not a fact.

    Week 2 · obligation of means
  3. 03

    We issue a verdict, even when it leaves us out.

    The verdict is delivered in writing and signed. If the strategy is sound, we say so and the work ends there.

    Weeks 3 to 4 · handover
Who takes part
The CEO and the owner of the current strategy
What we need
The documents you already produced: strategy, architecture, data inventory
What is left
A written verdict, with what holds and what does not
Scoped

Diagnostic

We interview the whole operation and redesign the process.

$40–100k · 6–8 weeks
  1. 01

    We interview the whole operation, not just the committee.

    We talk to the people who execute, not only to those who report. Site-to-site variability shows up in the interview, never in the org chart.

    Weeks 1 to 3
  2. 02

    We map the AS-IS and design the TO-BE.

    The AS-IS is measured in real time and real steps. The TO-BE is designed on that map, stating which step disappears and which stays under human oversight.

    Weeks 4 to 6
  3. 03

    We return three to five costed initiatives.

    Each initiative comes with investment, savings, timeline, owner, and its place in the sequence. The deliverable is the same regardless of how much surface we examine.

    Weeks 7 to 8 · with sequence and owner
Who takes part
The operating areas, their leads, and the people who run the process
What we need
Access to the operation for three weeks and to the source systems
What is left
AS-IS, TO-BE, and three to five initiatives with business cases
Recurring

Advisory

The portfolio stays live and is re-scored when the business moves.

Committed monthly hours
  1. 01

    The portfolio is re-scored when the business moves.

    When model cost, volume, or regulation changes, each case score changes with it. The portfolio is re-scored, not re-commissioned.

    Monthly
  2. 02

    One session a month with the sponsor, fixed agenda.

    The session always has the same agenda, which is why it takes an hour: what moved, what stalled and why, and what enters.

    Committed hours by tier
  3. 03

    The diagnostic never gets re-commissioned from scratch.

    The knowledge stays in the live portfolio, so next quarter does not start with a fresh diagnostic.

    Recurring · the portfolio stays live
Who takes part
The executive sponsor and the owner of each initiative
What we need
One hour a month and the progress data for each case
What is left
The portfolio re-scored, with the sequence up to date
Product

AI Agent Factory

And the consulting output becomes an agent in production.

$150/mo per module
  1. 01

    The approved case enters the factory.

    Only cases already approved at the gate enter. The factory is not a lab: what gets configured there has an owner and a business case.

    Approved cases only
  2. 02

    It is configured, evaluated against baseline, and packaged.

    Tools, guardrails, and the oversight threshold are declared in the agent, not in the head of whoever built it.

    Without writing custom code
  3. 03

    Your team pushes it to your own production environment.

    Publishing is the client's. We do not operate your production environment, which keeps accountability where it belongs.

    Product · activated by contract
Who takes part
Your technical team and the process owner
What we need
An environment to publish into and access to your systems
What is left
Packaged agents, with guardrails and evaluation
The product

AI Agent Factory

An agent builder: tooling schemas and guardrails configured and packaged, ready for the client's production environment. Only approved cases go in.

See platforms
Let's talk

Let's start with a conversation.

Ninety minutes with your leadership team. You leave with a first opportunity map and the recommended next step.

What you're interested in
Response
within one business day
First session
90 minutes with your leadership team
Output
a first opportunity map and the recommended next step
Format
in person or remote
Language
Spanish or English
hola@criterium-ai.com
ARIA

A 12-minute conversation with ARIA, our agent. When you finish, you receive a preliminary proposal for discussion by email.

With confidence.