AI and automation
Agentic AI: what is it and how do you use it?
Agentic AI is AI that doesn’t just provide answers, but is given a goal, plans the steps itself, executes them in your existing systems, and checks the result. You set the boundaries and decide on everything that matters.
Three types of AI assistance
ChatbotAnswers a questionResponds
WorkflowExecutes fixed stepsFollows rules
Agentic AIPlans, acts and checksExecutes
In this article
The concept
What is agentic AI, and what is it not?
Almost every company has experimented with AIAI Computer programs that perform tasks normally requiring human intelligence, such as writing text or answering questions. by now: having a text written, asking a question, or creating a summary. That is useful, but it remains manual work. You type, the AI answers, and you turn the answer into action yourself. Agentic AIAgentic AI AI that independently performs multiple steps to achieve a goal, instead of merely answering a question. reverses that. You give a command and the system does the work, step by step, in the programmes you already use.
The word comes from “agentAI Agent An AI program that independently performs a series of tasks, such as checking pages or conducting measurements.”: someone who acts on behalf of another. An agentic AI system acts on your behalf, within agreements you establish beforehand. The difference from what you might already know is mainly how much it is allowed and able to do itself.
A calculator does exactly what you type in. A good employee understands what you want to achieve and handles the rest. Agentic AI is closer to the employee, with one important difference: an employee has final responsibility, while an agent needs your rules and your control.
The market
Why agentic AI is emerging now
Research firm Gartner expects that by 2028, a third of enterprise software will contain agentic AI, compared to less than 1% in 2024. At least 15% of daily business decisions will then be taken independently by AI agents. The direction is clear, even if the pace varies by sector.
At the same time, there is noise. Gartner warns of “agent washing”: products sold as agentic when they are just a regular chatbotChatbot A program that automatically responds to visitor questions within a chat window. or a fixed workflow. Gartner also expects that over 40% of agentic AI projects will be discontinued by the end of 2027, mainly due to high costs, unclear returns, or insufficient risk management. A good question for any supplier is therefore: what task does this perform independently, in which systems, and who checks the result?
Gartner, press release of 25 June 2025 on agentic AI in enterprise software and agent washing
Our approach
How EQwise uses agentic AI
We don’t build on a single AI. We combine multiple models, local and online, and connect them to the systems where the work happens. We deliberately keep the technology behind the scenes, because for you, it’s the result that counts. The principle, however, is easy to explain.
The best model for the task, not the biggest model for everything
No single model is the best at everything. For complex reasoning, writing, and up-to-date knowledge, we use powerful online models. For simple, repetitive work and for anything that must remain private, we use local models that run on your own equipment. This saves costs and time, and sensitive data never leaves your environment.
- Online: maximum power, current knowledge, strong in difficult tasks and long texts.
- Local: privacy, predictable costs, fast for routine work. Data remains within the own environment.
Integrations give the agent hands and eyes
A model that can only talk has limited reach. Through integrations (the open standard for this is called MCP, Model Context Protocol), we connect agents to the systems where the work is: your website, search and visitor data, documents, calendar, tasks, and reports. Every integration increases the agent’s reach, and every integration has strictly defined permissions. An agent allowed to read your website is not necessarily allowed to change everything.
How models, agent and integrations work together
Multiple agents, each with its own role
Just like in a team, it works better if one person doesn’t do everything. One agent researches, a second builds, a third checks the result. That checking role is important: the creator never assesses their own work. This is also how our AI swarm for discoverability works.
Three roles, a human at the helm
How it’s structured
- Goal and rules. You determine what the result should be and what must never happen.
- Agents with roles. Researching, executing, checking.
- Models, local and online. The right model for each task.
- Integrations. Access to website, data, documents and other systems, with limited permissions.
- Human at the helm. Control, approval and final responsibility remain with you.
What this delivers in practice
We are currently using this for our own work and that of our clients:
- Building and managing websites. An AI agent works directly in the website, based on our own building blocksGutenberg The default WordPress editor where you build a page using individual blocks.. This primarily delivers quality and oversight: a consistent structure, less manual work and fewer errors. The work also goes faster and smoother, after which a specialist handles the finishing touches and final check.
- Discoverability in Google and AI. A combined check of SEOSEO Everything you do to improve your visibility in Google without paying for advertisements. and GEO (visibility in AI answers), where agents do the research and we assess what really matters.
- Measure, fix, re-measure. Speed and quality issues are flagged, resolved and then tested again, as a closed loop.
- Reporting and administration. Data from different sources is brought together in overviews that you refresh without manual work.
- Target audience test. A panel of AI personas reacts to a design or text, so we know before publication whether your target audience understands and appreciates it.
Our rule: not the list of 200 improvement points, but the 5% that delivers results. Agents provide the breadth, our craftsmanship determines the choice.
For your business
Agentic AI in SMEs
For a smaller company without its own marketing or IT department, the benefit is greatest. Agentic AI doesn’t replace a team you don’t have. It does give you the capacity of a team you couldn’t afford. Typical starting points:
- Automatically compiling recurring reports and overviews.
- Keeping your website up-to-date, fast and easy to find.
- Handling quotes, invoices and reminders faster.
- Preparing or sorting customer queries, with a human as the final check.
Start small. One process that recurs often and has clear rules provides quick proof and shows where the risk lies.
For larger organisations
Agentic AI at scale: staying manageable
For larger organisations, it’s less about whether it’s possible and more about how you keep it manageable. It revolves around four things.
Four things that keep it manageable
Data protection
Sensitive data remains in its own environment. This is where the local layer of a mixed setup comes into its own. Think of GDPRGDPR The European privacy law that regulates how organisations must handle personal data. and the European AI Act.
Permissions management
For each agent and each integration, you determine exactly what may be read and what may be changed.
Traceability
Every action is recorded, so you can always check what happened and why.
No vendor lock-in
Those who can swap multiple models remain flexible if prices, quality or rules change.
Rules of play
Risks and how to mitigate them
AI can make mistakes and sometimes hallucinate. That is no reason to forgo agentic AI, but rather to set it up smartly:
- Give an agent only the permissions necessary for the task.
- Always have actions with consequences, such as publishing, paying or sending, approved by a human.
- Have a second agent or a human check the result, never the creator themselves.
- Record what happens so you can make adjustments.
Getting started
Starting with agentic AI in five steps
- Choose one process that recurs often and has clear rules.
- Define the boundaries: what is the agent allowed to do and what is strictly forbidden?
- Connect only what is necessary and give those integrations limited permissions.
- Measure the result against the current situation: time, quality, costs.
- Scale up only once it demonstrably works.
Questions
Frequently asked questions
Hoe begin ik met agentic AI in mijn bedrijf?
Kies één proces dat vaak terugkomt en duidelijke regels heeft, bepaal wat de agent wel en niet mag, koppel alleen de systemen die nodig zijn en meet het resultaat. Pas daarna schaal je op.
Is agentic AI hetzelfde als een chatbot?
Nee. Een chatbot geeft antwoord op een vraag. Een agentic AI-systeem voert een opdracht uit over meerdere stappen heen, gebruikt daarvoor koppelingen met andere software en past zijn aanpak aan als iets niet lukt.
Is agentic AI ook geschikt voor het mkb?
Ja, juist voor het mkb. Je begint klein met één terugkerend proces, zoals rapportages, websitebeheer of administratie, en breidt uit zodra het aantoonbaar werkt.
Waarom combineer je lokale en online AI-modellen?
Geen enkel model is overal het beste in. Online modellen leveren de meeste kracht en actuele kennis. Lokale modellen houden gevoelige gegevens binnen je eigen omgeving en zijn voor eenvoudig werk sneller en goedkoper. De combinatie geeft kwaliteit, privacy en beheersbare kosten.
Wat is agentic AI in gewone woorden?
Agentic AI is AI die een doel krijgt, zelf de stappen plant, die stappen uitvoert in je bestaande systemen en het resultaat controleert. Een mens bepaalt de grenzen en beslist bij belangrijke keuzes.
Wat zijn de risico’s van agentic AI?
AI kan fouten maken of dingen verzinnen. Daarom krijgt een agent alleen de rechten die nodig zijn, wordt elke actie vastgelegd en blijft een mens eindverantwoordelijk voor beslissingen met gevolgen.
Free website scan
Wondering what agentic AI could mean for your business?
We’ll tell you honestly where it works, where it’s not worth it, and where you can start. You’ll get a concrete plan, including for your website and your discoverability.
A concrete report within two working days. No obligations.
Written by
