Book a callFree trial

Creatio AI agents: AI that runs inside your processes, not alongside them

You have probably seen it already. Someone demonstrates an AI assistant that answers a customer question or writes a summary, everyone is impressed, and three months later nobody uses it any more.

Book a consultationMore about Creatio
Lachende klantenservicemedewerker met headset in gesprek met een klant
Veelgestelde vragen AI agents

Why stand-alone AI assistants get stuck at demo level

The reason is almost always the same: the assistant sits next to the work instead of inside it. It can talk, but it can't act. It doesn't know the customer, it doesn't know which stage a file has reached, it can't request an approval and it can't record anything. So the employee still ends up doing everything themselves, and then an extra screen is more of a burden than a help.

Creatio AI Studio takes a different approach. Agents are not a layer on top of the platform but participants in your processes: with the same data, the same permissions and the same logging as your employees.

What AI Studio is

AI Studio is the environment in which you build, deploy, orchestrate and govern AI agents across their full lifecycle. Not a feature inside an application, but a dedicated workspace with a designer, a test environment, a deployment mechanism, analytics and a governance layer.

Agents work inside Creatio, reason over your CRM and process data, and can run searches and carry out actions across channels, without any custom development.

Two kinds of agents, for two kinds of work

Prompt agents

The easy-to-start type. You describe in plain language what the agent should do, and at its core that is the whole design. Then you enrich it: add knowledge sources so it draws on your own documentation, connect tools so it can carry out actions, set triggers, and define guidelines that determine what it may and may not do.

A typical example is an agent that answers frequently asked questions from your knowledge library, deployed in the web chat. Something like that is up and running in an afternoon.

Agents come with version control and can be deployed to different environments, so you can test before anything reaches customers. Once published, you follow their behaviour with an analytics suite and adjust through policies.

Workflow agents

The heavier type. These agents are linked to your business processes and are triggered by events: a change to a record, a step in a process, an incoming message, a point in time. They carry out a series of actions instead of giving a single answer.

This is where the difference with a chatbot lies. A workflow agent can assess an incoming request, fetch the data it needs from several systems, do a calculation, start an approval step and move the file to the next stage, all within the process you designed.

What sets Creatio apart: agents and processes work both ways

This is where the platform's BPM background pays off, and it is the argument you can't make for most AI platforms.

Why it matters: AI is probabilistic and processes shouldn't be. By using AI for exactly the steps that require judgement, and leaving the rest to the process model, you get the best of both. You don't get a system that "usually does the right thing", but a process with predictable outcomes in which AI handles the difficult assessments.

More about Creatio BPM

A process can call an agent

When the process reaches a step that requires interpretation (assessing a document, classifying free text, estimating a risk), it calls an agent, gets an answer back, and continues down one route or the other based on that answer. The rest of the process stays deterministic.

An agent can start a process

The other way round, an agent handling a customer query can decide that a formal procedure is needed, and kick off that process with the right data attached.

A different architecture

That is a fundamentally different architecture from an AI assistant that sits on top of an application and hopes the employee handles the outcome correctly.

Team in overleg rond een bureau met laptops op kantoor

Several agents working together

Agents don't have to do everything alone. In a multi-agent set-up you divide tasks between specialised agents that are directed by a coordinating agent.

For composite processes that works better than one agent that has to be able to do everything. Think of a quote request where one agent gathers the customer data and history, a second runs a credit check, a third calculates the pricing and a fourth summarises it all for the account manager. Each agent has a well-defined task, its own knowledge sources and its own permissions, which makes it testable and easy to adjust.

Where your agents show up

Creatio deploys agents across multiple channels: web chat, email, SMS, messaging channels such as WhatsApp, Telegram, Facebook Messenger and Instagram Direct, collaboration tools such as Microsoft Teams and Slack, and voice channels.

  • Behaviour per channel

    For each channel you set how the agent behaves: which personality and tone it uses, which knowledge sources it draws on, and when it hands over to a person. That handover is no side issue: it is the setting that determines whether customers experience the agent as helpful or as an obstacle.

  • Replaying sessions

    You can also debug sessions, which is worth a lot in practice. When an agent gives a wrong answer, you want to be able to trace back where it came from instead of guessing.

No-code and pro-code in the same environment

  • The visual designer

    For most agents the visual designer is enough: describe, connect knowledge sources, add tools, publish. Someone from the business can do that.

  • The developer side

    With the Code Agent Designer and the AI development toolkit you can build agents using coding agents such as Claude Code, Codex or GitHub Copilot, which have access to the structures of your Creatio environment: Freedom UI components, data models, process definitions and application metadata.

  • One governance model

    The beauty of it is that agents built this way are managed in exactly the same way as agents from the visual designer. One platform, one governance model, one place where you can see what is running. That prevents a situation that arises in many organisations: a few tidy agents in the official environment and a shadow collection of home-built tools that nobody keeps track of.

Ondernemer werkt aan een laptop in een kantoor met bakstenen muur

Your own language model, and why that matters

Creatio is explicitly LLM-agnostic. You connect the language model of your choice and so decide for yourself which party processes your data.

For Dutch organisations, that is often the question that decides whether an AI project goes ahead. In healthcare, government, education and financial services, "where is this processed and by whom" is not a detail but a precondition. A platform in which that choice is fixed by the vendor regularly drops out in those sectors.

The second argument is strategic. The model market is moving fast; what is the best model today may not be in a year's time. By building your processes on a platform that isn't tied to one model vendor, you keep that choice open.

Governance: the part that decides the purchase

In AI projects at larger organisations, this is where the decision is made, not on functionality.

Creatio offers a central governance dashboard with real-time insight into what agents are doing: which activity, which actions were taken, which policies were triggered and what the outcomes were. On top of that you define behavioural rules, set execution limits, control access at data level and configure compliance policies.

In practice, that means you can answer three questions that management or an auditor always asks. What exactly did this agent do? Based on which data? And what stopped it when it was about to go beyond its limits?

Without those answers, an AI agent in a regulated environment won't pass the internal review. With them, the conversation is usually over within an hour.

The AI agent now handles a large share of standard requests itself. My team has time left for the cases that really need a person.

Sample quote Real customer quote to follow

Where AI agent projects run aground

Five causes we see again and again in practice, and none of them is technical.

Book a consultation

  • You start with the technology

    "What can we do with AI" is the wrong question to start with. "Which work costs us the most time and is the most predictable" gives you a usable candidate within a week.

  • Your knowledge source isn't up to scratch

    An agent that draws on outdated documentation gives outdated answers, with authority. That is worse than no answer. Cleaning up comes before building.

  • Authorisations aren't defined

    What may the agent settle on its own and where does it hand over? A credit note of fifteen euros is something else than a contract cancellation, and that distinction belongs in the policy, not in someone's head.

  • Nobody owns it

    An agent without an owner who is allowed to change it quietly gets worse as reality changes.

  • You let it work on its own straight away

    Let an agent observe and make suggestions first. Only when you have seen for a few weeks that its judgement is sound do you give it the controls.

How we approach a project

  1. We start with the process

    Which step takes time, is repetitive and has a clear rule or a recognisable judgement? This conversation often reveals that half of the question doesn't need AI at all, just a better process design. You can also book our process mapping workshop separately for this.

  2. We decide the technology per step

    Deterministic where possible, an agent where judgement is needed.

  3. We define authorisations and handover points

    Before anything is built.

  4. We clean up the knowledge source

    The least exciting and most decisive part.

  5. We build one agent and test it in a protected environment

    And we let it observe first.

  6. We measure and scale

    Only then the second agent.

  7. We hand over

    Your team must be able to adjust agents themselves, otherwise you'll be back to square one in a year.

An agent is only as good as the process it sits in

The question is not whether AI agents work. They do. The question is whether they can do something with what they come up with, and whether you can explain afterwards what happened.

That is exactly where Creatio is strong: agents that run inside your business processes instead of alongside them, with your own language model, your own permissions and a governance layer that keeps the conversation with your management and your auditor short.

FTE betekenis

Frequently asked questions about Creatio AI agents

  • An assistant responds to what you ask and delivers a proposal; you carry it out. An agent is given a goal, decides for itself which steps are needed, carries them out in your systems and reports back, within the permissions and boundaries you have set. The difference lies in being allowed to act, and that is also where most of the value comes from.

  • Prompt agents are the easy-to-start type: you describe in plain language what the agent should do, add knowledge sources and tools, and publish. Suitable for question-and-answer tasks and well-defined actions. Workflow agents are linked to your business processes and are triggered by events; they carry out a series of actions within a process instead of giving a single answer.

  • Yes, and that is Creatio's strongest point in this area. The relationship works both ways: a process can call an agent for a step that requires judgement and continue based on the answer, and an agent can start a process itself. That keeps your processes predictable while AI makes the assessments where rules fall short.

  • Creatio is LLM-agnostic: you connect the model of your choice and so decide for yourself which party processes your data. For organisations with requirements around EU data storage or an established AI policy, that is often decisive. The second advantage is strategic: you are not tied to one model vendor in a market that is moving fast.

  • Through the governance dashboard, which gives real-time insight into agent activity, actions taken, policies triggered and outcomes. On top of that you define behavioural rules, execution limits, access control at data level and compliance policies. In practice, that means you can show what an agent did, based on what, and what stopped it: exactly the questions an auditor or your data protection officer will ask.

  • For most agents, no. The visual designer is intended for people from the business: describe, connect knowledge sources, add tools, test, publish. For more complex agents there is a developer side where you build with coding agents such as Claude Code or GitHub Copilot, with access to your data models, process definitions and UI components. Both types are then managed in the same environment.

  • Yes. Language models are probabilistic, so an agent can get it wrong. That is why you work with guardrails, well-defined authorisations and a clear moment of handover to a person. Responsibility always stays with your own organisation. Start with agents that observe and make suggestions, and only give them independence once you have seen their judgement over a period of time.

  • A simple prompt agent that answers questions from your knowledge library is up and running in a day. A workflow agent that runs inside a business process and carries out actions in several systems takes days to weeks, depending on the integrations. In practice, the lead time is not determined by the build but by three other things: getting clear on what the agent should do, cleaning up the knowledge source, and defining authorisations.

Curious which step in your process lends itself best to a first agent?

Mobile XL is a Creatio partner and always starts with your process, not with the technology.

Book a no-obligation consultationContact us