About Us

Blinqx offers AI workflow platforms that empower service professionals to grow and succeed.

Sectors

Blinqx develops AI workflow platforms for financial and business service providers in selected sectors.

Insights

Stay up to date on what's going on at Blinqx: awards, acquisitions, knowledge, cases. You can find it here!

Search

Onboard your agent as if they were a new coworker

Modified on
Written by Jeroen van Eijk

Onboarding an AI agent is more like training a new colleague than installing software. The agent needs context, clearly defined authority, guidance during the first few weeks, and a review period. Platform builders who build that process into their product help customers actually put agents to use.

Why does an agent perform better when you train them?

An agent who is immediately given full access and a vague assignment makes mistakes that erode trust. An agent who is given more responsibility step by step, on the other hand, builds that trust. We’ve known this for a long time when it comes to new employees.

We’ve all been there: a good first week on the job determines how quickly someone gets up to speed. The new colleague is briefed on the team, the clients, and the work process. He’s given access to the systems he needs—and nothing more. A mentor checks in with him. After a few weeks, there’s a discussion about what’s going well and what could be improved.

When rolling out agents, we often skip those steps. An office hires an agent, assigns them a broad range of tasks, and expects immediate results. If things go wrong, the conclusion is quickly drawn that the technology isn’t ready yet. Often, the problem was simply a lack of training.

What context does a new agent need?

An agent needs three types of context: knowledge of the organization, knowledge of the task, and knowledge of the client or the case. Without that context, an agent’s work will be generic. With that context, the results align with how the firm operates.

This translates into specific product choices:

  • Organizational knowledge. Corporate identity, standard phrasing, internal procedures, and quality requirements. An accounting firm uses different standard texts than a law firm, and each firm has its own way of working.
  • Task Knowledge. What exactly is the assignment, what constitutes a good result, and when should the agent stop and involve a human? A clear job description works just as well for an agent as a job profile does for an employee.
  • Case knowledge. The agent must know which client, case, or employee they are working on. In business and financial services, that knowledge is embedded in the software that we, as providers, deliver.

That last point is where platform builders make a difference. We manage the files, the relationships, and the history. The better we make that context available, the faster an agent can get up to speed. We previously wrote about the importance of industry knowledge for agents in “Why Industry Knowledge Becomes More Important as AI Becomes More Autonomous.”

How do you grant an agent the right permissions?

Give an agent the same rights as the colleague on whose behalf they’re working, and start with fewer. Let the agent read and prepare first, and only expand to execution once the results are good. That way, autonomy grows alongside trust.

When a new employee joins, IT sets up their access based on their role. The same principle applies to agents. In WorQX, organizations centrally manage users, permissions, and access for all applications in the workspace. That’s also where agent permissions belong: a single overview of who has access to which files and what they can do with them.

Three design principles:

  1. Permissions follow the user. An agent never sees more than the professional on whose behalf they are working.
  2. Autonomy is configurable. The user determines the pace and degree of autonomy: making proposals only, executing after confirmation, or acting independently within agreed-upon limits.
  3. Authorizations are transparent. A manager or compliance officer can see at a glance which agent is authorized to do what. This aligns with the principle of compliance-by-design.

Who mentors the officer during the first few weeks?

Just like any new colleague, an officer needs a designated mentor: a professional who monitors their work, provides feedback, and offers guidance. This role ensures that mistakes are caught early and that the officer adapts more effectively to the realities of the job. Plan this mentoring in advance, specifying both the timeframe and the mentor’s name.

In practice, a simple approach works best. Assign a mentor to each agent within the office. During the first few weeks, that mentor reviews a set portion of the agent’s work and identifies areas for improvement. That feedback is then incorporated into the instructions, examples, and context.

As a provider, you can support this within your product:

  • Feedback in the flow. Let the user indicate immediately after seeing a result whether it was correct—and, if not, why not.
  • Collect examples. Approved results are the best guidance for an agent. Make it easy to save them as examples.
  • Overview for the owner. Show how many tasks the agent completed, how many were modified, and where the agent most often needs guidance.

How do you evaluate whether an agent is performing well?

Evaluate an agent the same way you evaluate an employee after the probationary period: based on quality, speed, and reliability, measured against predefined goals. The outcome determines whether the agent is given more responsibility, needs guidance, or is let go. Make that decision explicit.

Here are a few metrics we find useful in practice:

  • Acceptance rate. What portion of the results does the professional adopt as is?
  • Correction time. How much time does it take to improve a result, compared to doing it yourself?
  • Errors with consequences. How often does a result lead to an error that affects a client or case?
  • Usage. How often do employees use the agent? Low usage is a sign that the agent isn’t meeting their needs.

Insight into usage also helps with the business case. At Blinqx, agents are billed based on usage. In WorQX, users and managers can therefore see how AI credits are being used, so that value and costs are presented side by side.

On Tech TalQX, the podcast about AI in regulated sectors, we regularly discuss how adoption is unfolding in practice.

What does it mean to onboard an AI agent?

Onboarding an AI agent means bringing the agent online step by step, just as you would train a new employee. You provide the agent with context, clearly defined authority, guidance, and an evaluation opportunity. This helps build trust before the agent begins working more independently.

What permissions do you grant a new AI agent?

Grant an agent the same rights as the user on whose behalf they are working, and start by reading and preparing. Only expand their permissions to include execution if the results are good. Document those rights centrally and clearly.

Who is responsible for supervising an agent?

Assign an owner within the organization for each agent. This professional will monitor the agent’s work during the first few weeks, provide feedback, and adjust instructions. Plan this role in advance, assigning both a timeframe and a specific person.

How do you measure whether an AI agent is performing well?

Measure the acceptance rate of results, the time required for corrections, the number of errors with impact, and how often employees use the agent. Compare these figures to the goals you set in advance. The outcome will determine whether the agent is given more responsibility.

What is WorQX’s role in deploying agents?

WorQX is Blinqx’s digital workspace where tools, data, and processes come together. Organizations can centrally manage users, permissions, and access there, and see how AI credits are being used. This makes it easy to ensure that agents have the right permissions to do their jobs.

Related articles