Customer support
Tickets arrive sorted and de-duplicated, with a first-line reply ready. Your teams approve the replies and settle the cases that call for their judgement.
Definition and examples
Agentic AI explained, and how it applies to your ERP or CRM
Agentic AI turns a conversation partner into a digital team member that carries out whole processes, working within set limits and under visible supervision.
In brief
What is agentic AI?
Agentic AI is artificial intelligence that completes whole tasks on its own. It chooses the steps, uses several tools and acts inside your software. An AI agent pursues an objective over time, under human oversight, to a written brief. Djtal, the operational AI specialist for Swiss companies, connects such agents to the ERP or CRM a company already runs.
A chatbot gives you an answer and a copilot offers a suggestion for you to approve. An agent carries the task through from start to finish. It sorts your tickets, enriches your CRM and prepares your journal entries while your teams get on with other work. Djtal builds custom AI agents for Swiss companies, connected to the ERP (enterprise resource planning) or CRM (customer relationship management) system they already run and working under supervision.
For context
How a chatbot, a copilot and an agent differ.
| Criterion | Chatbot | Copilot | AI agent |
|---|---|---|---|
| Trigger | On request, question by question | When you invoke it inside a tool | Continuously, within its scope |
| What it produces | An answer | A suggestion to approve | Finished work, with actions taken inside your systems |
| Autonomy | None | Assisted, step by step | Bounded, logged and auditable |
| Example | A website FAQ | A suggested email draft | Tickets sorted and answered, journal entries prepared |
And conventional automation? It follows fixed rules (if this, then that) and breaks as soon as a case falls outside the planned scenario. An agent reasons and adapts. It can handle the unexpected and knows what to hand over to a human.
Our difference
Agentic AI on the tools you already use.
We put the tool you use today to work. Your business data lives in your ERP, your CRM and your email, so we connect agents to these systems to read and enrich your data and trigger actions, with no migration.
Agentic AI connects to Zoho, Salesforce, HubSpot or Microsoft Dynamics, whichever you already run. It is quicker to install and less risky. It also pays for itself sooner.
The reason is structural. Your history, your records and your documents are already structured there, so the agent builds on them from the first day instead of waiting for a migration project to finish before it starts producing.
Use cases
Where agentic AI pays off within the first few weeks.
The best first cases are repetitive tasks at a steady volume, connected to your data. Here are four examples from our deployments. Our AI agents page covers them in detail, and two functions have their own pages: finance and communications.
Sales & CRM
Contact and company records are completed, incoming enquiries qualified, briefing packs prepared for each meeting and dormant opportunities flagged. Your CRM stays clean and current.
Finance & admin
Invoices and expense claims are read, the data extracted and the journal entries prepared. The books are kept up to date every day, and your team checks and approves them.
Market intelligence & knowledge
The market is tracked, the essentials summarised and the company's knowledge captured. Everyone finds up-to-date information when they need it.
What makes projects fail, and our answer
Method is what makes agentic AI succeed.
Gartner (June 2025) expects over 40% of agentic AI projects to be cancelled by the end of 2027, typically after pilots that go nowhere or agents delivered without supervision. Thirty years in professional IT and four waves of technology have taught us where projects break. These are the three common failures, and how we prevent each one by design.
An agent that is delivered, then abandoned.
This is the most common failure: an agent goes live with nobody to watch over it. At Djtal, each agent runs under a supervision subscription, so its actions stay logged and it is maintained and improved for as long as it runs.
An agent that acts on bad data.
An agent connected to inaccurate or incomplete data amplifies the error. We connect each agent to your real data, within a written scope. Every sensitive output goes through human approval.
A black-box agent beyond any control.
You see everything the agent does. Every action is traced, and the level of autonomy is set step by step, from observation to audited autonomous action. You can suspend an agent within minutes.
We run agentic AI in our own company.
Djtal's own work runs on nine AI agents: Sofia, Iris, Cassandra, Victor, Adil, Émile, Athena, Mercure and Basilio. Between them they cover strategy, communications, sales intelligence, projects, administration, finance and knowledge management, and every one works to a written scope under a human supervisor.
On 3 July 2026, Iris produced and published her first LinkedIn post from start to finish. She did the research, wrote the post, created the visual and put it online. A human approved the content before it went out, and that approval is the only step in the chain reserved for a person.
Further reading. Find out how we design and run agents, meet our nine agents, or map your opportunities with an AI strategy audit.
Frequently asked questions about agentic AI
What is agentic AI, in one sentence?
Agentic AI describes AI systems that carry out complete tasks autonomously. They choose the steps, use several tools and act in your software. An AI agent pursues an objective over time, under human oversight, to a written brief.
What is the difference between agentic AI and conventional automation?
Conventional automation follows fixed rules (if this, then that) and breaks as soon as a case falls outside the planned scenario. An AI agent reasons: it adapts to a badly worded email, an invoice in an unfamiliar format or an ambiguous request, and it knows what to hand over to a human. Automation executes a rule. An agent pursues an objective.
Can you add an AI agent to software you already use, such as Zoho or another CRM?
Yes, and it is usually the quickest and most cost-effective route, because your business data already sits in your ERP, your CRM and your email. Djtal connects agents to these systems to read and enrich your data and trigger actions, with no migration and without replacing what already works. We start from what you already run, be it Zoho, Salesforce, HubSpot or Microsoft Dynamics.
What are the risks of agentic AI, and how do you avoid them?
There are three common failures: an agent delivered and then left without supervision, an agent connected to inaccurate or incomplete data, and an agent whose actions nobody checks. Djtal's answer is oversight by design: a written scope, your real data, human approval of sensitive outputs, an audit log and an agent you can switch off at any time.
How do you tell a real AI agent from a product relabelled ‘agentic’?
The phenomenon has a name, ‘agent washing’. Gartner (June 2025) estimates that about 130 vendors offer real agentic capabilities, out of the thousands that claim them. Here are three simple checks. Does the agent act in your systems, beyond conversation? Are its scope and limits of action written down? Are its actions logged and verifiable? A product that fails these three checks is a relabelled chatbot.
Is agentic AI already relevant for an SME?
Yes, agentic AI is already relevant for an SME, and the impact is often clearest there. An SME's processes rest on a few people, so taking repetitive tasks off them (sorting requests, enriching data, reading documents) frees up time that shows straight away. A precise scope and clear supervision are enough to get started, with or without an IT department.
Where should we start with agentic AI?
Start agentic AI with a repetitive process at a steady volume, where the value can be measured quickly. Djtal identifies this first case with you in a 30-minute call. For an overall view of your options before anything is built, the AI strategy audit maps the opportunities and prioritises them in a 90-day plan.
Get started
Your first agent starts with the right process.
Agentic AI suits only some processes: repetitive ones, at a steady volume, connected to your data. Thirty minutes are enough to find out whether one of yours is among them, and if none is, we will tell you that too. There is no obligation.
Further reading: Why so many agentic AI projects fail · Agentic AI in practice
Last updated: