During the first month of deploying managed company AI, you should expect a structured onboarding process that prioritizes defining written authority, connecting approved data sources, and establishing a baseline for review. The focus is not on immediate automation, but on building a governed record where every action is tied to a specific, ratified permission. This phase ensures that the AI understands your business context while strictly adhering to the boundaries you set, preventing unauthorized actions before full operational use begins.
Defining Written Authority and Scope
The most critical step in the first month is establishing what the AI is authorized to do. Written authority is a separately ratified document that names specific capabilities and their limits, ensuring that the AI cannot act on inference alone. In the context of Sabina, this means that your company decides what she may do, what needs your approval, and what stays with your team. This separation between knowledge and permission is fundamental to trust.
For example, if a caller asks for a discount, the system checks the request against the grants you have issued. If no grant covers that specific action, it comes back to you as a question rather than an action taken on a guess. This mechanism ensures that the record closes the entry with clarity on what was asked, what was cleared, and which grant covered it. The implication is that your team spends the first few weeks reviewing these authority treatments, such as 'watch only' or 'per-act approval,' to ensure they align with your operational risk tolerance.
Connecting Data Sources and Forge
Once authority is defined, the next phase involves connecting the AI to your company's information. Forge is the included workspace for agreed business records, serving as the structured memory for the AI. During the first month, the provider, in this case Dominus Foundry, configures the system to read from the sources you approve. This includes company documents, customer records, and operational logs.
It is important to note that moving the agreed work into Forge is preferred, but if you keep another system, API access and the necessary reads and writes must be verified. For instance, a property management firm might connect their leasing database so the AI can understand tenant histories. The implication here is that data quality directly impacts AI performance. If the sources are incomplete or outdated, the AI will need clarification and correction. Your team should expect to spend time reviewing the initial data connections to ensure accuracy before the AI begins handling live tasks.
Teaching the Business Context
With data connected, the AI begins to learn the specific nuances of your business. Company context is the accumulated knowledge of who your customers are, what you have quoted, and what you committed to. During the first month, your team teaches the AI how work should be handled by providing instructions, corrections, and decisions. This teaching process is distinct from the authority system; what the AI learns does not automatically widen what she is authorized to do.
Consider a law firm where an associate teaches the AI how to format a specific type of legal brief. The AI records this preference in its memory, but it cannot send the brief without a separate grant of authority. This separation ensures that insight and authority are kept apart on purpose. The implication is that the first month is a period of active collaboration, where your team reviews the AI's drafts and outputs, providing feedback that refines its understanding of your firm's standards and tone.
Establishing Review and Feedback Loops
As the AI begins to handle tasks, a robust review process is essential. Per-act approval is a disposition where the AI proposes a specific action and waits for a yes on that act. During the first month, most capabilities will likely start in a 'watch only' or 'compose only' mode, where the AI observes and drafts but does not act. This allows your team to build confidence in the AI's outputs without risking unauthorized actions.
For example, an accounting firm might have the AI draft email responses to clients. The AI composes the draft, but a human must send it. Over the first few weeks, the team reviews these drafts, noting where the AI missed a nuance or used incorrect terminology. These corrections are recorded and used to improve future outputs. The implication is that the first month is not about replacing human judgment, but about augmenting it with a reliable, governed assistant that learns from your feedback.

Assessing Fit and Adjusting Scope
Toward the end of the first month, you should assess whether the initial scope of work is appropriate. The assessment process involves reviewing the AI's performance against your baseline expectations. If certain tasks are consistently handled well, you might consider expanding the authority for those specific capabilities. Conversely, if the AI struggles with a particular type of work, you might limit its scope or provide more detailed instructions.
For instance, a roofing company might find that the AI excels at scheduling follow-ups but struggles with complex quote revisions. In this case, the team might keep the scheduling authority as 'scoped autonomy' while keeping quote revisions in 'per-act approval' mode. The implication is that the first month is a dynamic period of adjustment, where the scope of the AI's responsibilities is refined based on actual performance and team feedback. This ensures that the deployment is tailored to your specific operational needs, rather than a one-size-fits-all solution.
Key Takeaways
- Written authority is the foundation of managed AI, ensuring that actions are tied to ratified permissions rather than inference.
- Forge serves as the structured workspace for business records, requiring verified API connections if external systems are used.
- Teaching and authority are separate systems; learning about the business does not automatically grant the AI permission to act.
- Most capabilities start in 'watch only' or 'compose only' modes to allow for safe review and feedback.
- The first month is a period of active collaboration, where your team provides corrections and refines the AI's understanding.
- Scope should be adjusted based on performance, with authority expanded only for capabilities that demonstrate consistent reliability.
Frequently Asked Questions
What is the primary focus of the first month of deployment?
The primary focus is establishing written authority, connecting approved data sources, and setting up review loops to ensure safe and governed operation.
Does the AI act on its own during the first month?
No, most capabilities start in supervised modes like 'watch only' or 'compose only,' where the AI observes or drafts but requires human approval to act.
How is data privacy maintained during setup?
Data is held on independent primary operational databases for each company deployment, ensuring that your company's information is private and controlled by you.
Can the AI learn new tasks without new authority?
No, learning and authority are separate. The AI can learn about the business, but it cannot take on new actions without a separately ratified written grant of authority.
What happens if the AI makes a mistake?
Mistakes are recorded, and your team provides corrections. These corrections are used to improve the AI's understanding, but they do not change its authority unless explicitly ratified.
How long does the initial setup take?
The initial setup, including authority definition and data connection, typically occurs within the first few weeks, with the first month dedicated to review and adjustment.
Next Steps for Your Deployment
As you prepare to deploy managed company AI, remember that the first month is about building a foundation of trust and control. By defining written authority, connecting verified data sources, and establishing clear review loops, you ensure that the AI operates within the boundaries you set. To begin this process, you can start your assessment to identify the specific tasks and responsibilities that would benefit from AI support. This initial step helps you prepare for a discussion with the provider, ensuring that the deployment is tailored to your company's unique needs and operational goals.

