No more open-ended man-day invoices.Buy outcomes, not man-days.We offer Implement at a fixed price based on the validated scope.

Agentic AI · Solutions

What is agentic AI? A guide to AI agents for businesses

Agentic AI is an AI approach that plans the steps towards a given goal, reads data and prepares transactions in the systems it is authorised to use, observes the result and chooses the next step. A unit working this way is called an AI agent. An agent does not just answer the question, it tries to complete the work; at critical steps the decision stays with a person.

AI agents are built on top of a mature operation.

Last updated: · MindDX

Coming soon

We do not offer this service yet; we are preparing it. Write to us if you want to hear first when it opens, or to discuss an early assessment for your own processes.

01What an agent is

Not an assistant that answers a question — an agent that completes the work.

Works towards a goalInstead of answering one question, it breaks the goal into steps: it gathers information, checks it and chooses the next step.
Acts in your systemsIt reads and prepares records across ERP, CRM and production data, within the permissions it is given — it does not just chat.
Stays under human approvalCritical steps wait for a person’s approval. Every action leaves a trail of on whose behalf, and with which data, it was taken.

02Use areas

Where an agent will help in a mid-sized manufacturer.

These are planned use areas; this is not a service we offer today.

Orders and quotesMatches an incoming request to product, stock and price data in the ERP, and leaves a draft for the sales team to check.
Supply follow-upWatches late purchase orders, prepares the reminder to the supplier and alerts the planner.
Production and maintenanceTurns a deviation in machine data into a maintenance request and hands it, with its context, to the right team.
Customer and IT requestsClassifies an incoming request, matches it to past records and routes it to the right person with a proposed fix.
Period reportingCompiles the draft report from scattered sources and flags inconsistent items before sign-off.

03Foundation first

An agent is only as reliable as the system it works in.

That is why Agentic AI sits at the top of the maturity ladder. In a business whose foundation is not ready, an agent does not speed up the work — it speeds up the errors.

See the maturity levels
Level 3 · IntegratedConnected systemsThe systems an agent reads and writes must share one record. An agent working on disconnected sheets multiplies the inconsistency.ERP
Level 4 · OptimizedMeasured processesHow a process really flows must be known. Handing an unmeasured process to an agent hands over its flaws as well.Process intelligence
At every levelPermissions and trailWhat an agent may do is defined per role, and every action leaves an auditable record.
At every levelMeasurable valueEvery agent is tied to a business measure defined before it starts: time, errors or cost. An agent whose value cannot be measured is not built.

04How we will start

X-ray first, again.

  1. 01MeasureMindDX Checkup shows the maturity level and the processes ready for an agent.
  2. 02ChooseAssess validates one measurable use area and its scope.
  3. 03BuildThe agent goes live in a bounded area, together with its human-approval steps.
  4. 04TrackIts effect is tracked against the measure set at the start; the decision to extend rests on the data.

Agentic AI guide

Understanding AI agents, managing their risk and starting the right way

This guide explains what agentic AI is, how it differs from chat assistants, RPA and workflow automation, how an agent works, which risks are managed with which controls and which regulations are taken into account in enterprise use.

What are agentic AI and an AI agent?

Agentic AI is the name of an approach, and an AI agent the name of the unit that works this way. What sets an agent apart from a chat assistant is that it decides for itself which step to take, and when, towards the goal it has been given, and that it can carry out these steps in the systems it is authorised to use.

For example, a chat assistant is asked «which purchase orders are late?» and gives an answer. An agent is given the goal «follow up the late orders»: it reads the orders from ERP, finds the late ones, prepares reminders to the suppliers, alerts the planner and leaves the steps awaiting approval to a person.

The difference between an AI agent and a chat assistant, RPA and workflow automation

All of these tools automate something, but who decides the next step and what happens in an unexpected situation differ. The table below sets the differences side by side.

Comparison of the AI agent and other automation tools
ToolWho decides the next stepDoes it do work in systemsIn an unexpected situationSuitable work
Chat assistantThe userUsually not; it answersProduces an answer or says it does not knowAnswering questions, finding information
RPAA rule written in advanceYes, by imitating screens and stepsStops or returns an errorUnchanging, repetitive transactions
Workflow automationA designed flowYes, through defined connectionsGoes to the exception defined in the flowApproval and notification chains
AI agentThe agent, according to the goal and its observationYes, with the permissions it has been givenTries a new route or hands over to a personMulti-step work with variable inputs

How does an AI agent work?

An agent works in a loop: it takes the goal, plans the steps, reads data or prepares transactions with the tools it can use, observes the result and decides on the next step. When it reaches a critical step, the loop stops and waits for a person’s approval.

  1. GoalA measurable business goal and the limits of authority are defined.
  2. PlanThe agent breaks the goal into steps.
  3. ToolsIt reads from ERP, CRM or operational data and prepares draft records.
  4. ObservationIt checks the result and changes the plan if needed.
  5. Human approvalA critical step is made subject to a person’s approval.
  6. Audit trailEvery action is recorded with on whose behalf and with which data it was taken.
The working loop of an AI agent: from goal to audit trail.

The components of an AI agent

An enterprise agent is more than a single model. For it to work safely and usefully, each of the components below has to be designed.

The components of an enterprise AI agent
ComponentWhat it does
Language modelUnderstands the goal, plans the steps and produces text
Tools and connectionsDefined interfaces through which the agent can read and write in ERP, CRM and other systems
Context and memoryHolds the company knowledge needed for the task and the previous steps
OrchestrationManages the order of steps, retries and the division of work between several agents
Permission and approval layerDetermines which transactions the agent may carry out itself and which only with human approval
Audit trailRecords every action, its input and its result so that they can be reviewed later

The risks of AI agents and their controls

The fact that an agent can do work in systems increases both its value and its risk. The risks are well known and each has a control; these controls are put in place at the start of the design, not afterwards.

Risks and controls for AI agents
RiskExampleControl
Wrong actionA draft order to the wrong supplierHuman approval and starting with read-only access
Excessive permissionsThe agent being able to write to more systems than its task requiresRole-based permissions limited to the task
Data leakagePersonal or confidential data being passed to the wrong placeData classification and access limits
Malicious instructionThe agent being steered by an instruction hidden in an emailInput checks and rules for tool use
Untraceable decisionNot knowing why a transaction was carried outAn audit trail for every action

Regulations and frameworks for enterprise use

The enterprise use of AI agents is not tied to a single regulation, but several sources provide guidance. The EU Artificial Intelligence Act treats AI systems according to their level of risk and may concern systems placed on the EU market or used in the EU. The Turkish Personal Data Protection Authority (KVKK) has published a guide on generative AI and the protection of personal data.

For risk management, the NIST AI risk management framework and, for threats specific to agents, the OWASP publications are frequently consulted sources. They do not replace legal advice; assess your scope together with your legal and information security teams.

Is your company ready for an AI agent?

An agent is only as reliable as the system it works on. If the answer to these questions is not «yes», the foundation is prepared first:

  • do the systems the agent will read from and write to share the same record?
  • has the process to be handed over been measured, and is it known how it really flows?
  • is who may do what defined by role?
  • can the agent’s value be tied to a business measure defined before it starts?
  • has it been made clear under which conditions personal and confidential data will be used?

Measure your readiness: Assess

Where do you meet AI agents today?

The agent approach is entering the business software already in use, without a separate project: AI layers in customer service and IT service management tools that classify requests and suggest replies, and in CRMs that suggest the next step to the sales team, are examples of this. These capabilities are assessed by the same principles: permissions, human approval and an audit trail.

AI in IT service management: the ITSM guide

Frequently asked

Frequently asked questions about agentic AI and AI agents

Sources

Assess · Implement · Transform entry points

Let us first make clear what needs to be done.

The road into this category does not start with the category. Which category and which product fit is proven in Assess.

Assess entry points

Checkup — Readiness and assessmentSoftware-independent
Discover Scope — New implementation · clarifying the scopeRoute A
Audit — Live system · analysisRoute B