Organisations are putting serious money into AI for service and support, and agentic AI is fast becoming the next big investment for the IT service desk. A Gartner survey of 199 service and support leaders, published in August 2026, found that AI spending has grown by 38%, while overall service and support budgets have grown by just 2%.
That money is often being redirected from people and overhead into technology. Gartner’s own analysts caution that the real challenge is making sure those investments deliver measurable business value. It is a pattern we unpacked in why workplace AI adoption isn’t paying off yet.
For IT leaders, the same pressure applies to the service desk. Adding an AI chatbot is easy. Getting it to actually reduce workload and improve the employee experience is harder. This is where agentic AI comes in, and where many organisations are getting it wrong.
Agentic AI on the IT service desk: answering is not resolving
Most service desk chatbots are good at one thing: finding an answer. An employee asks how to reset their password or request a software licence, and the bot returns a knowledge article. The employee still has to do the work, or log a ticket anyway.
Agentic AI changes the goal from answering to resolving. Instead of explaining the process, an AI agent carries it out:
- Verifies who the employee is
- Resets the password or provisions the software
- Updates the ticket and asset records
- Confirms the change actually worked before closing the request
That last step matters. An agent that triggers an action and assumes it succeeded is no better than a script. A well-designed agent checks the result, and when something fails, it tries a safe alternative or hands over to a human with the full history attached.
The employee asks once. The work gets done.
Fix the foundations before the front end
An AI agent is only as capable as the systems it can act on. If your service management, identity, endpoint and asset tools don’t talk to each other, the agent hits the same walls your analysts do.
It is tempting to start with a polished new chat interface. But a smart front end sitting on disconnected back-end systems simply gives employees a faster route to frustration.
The better starting point is integration. Make sure the agent can read from and write to the systems that hold the answers: your ITSM platform, directory services, endpoint management and asset records. In most enterprises, this means getting more out of the tools you already own, not replacing them. We explore this further in the hidden cost of disconnected systems.
Governance: what the agent may do, not just what it can access
This is the question CIOs should spend the most time on. There is a real difference between what an AI agent can technically access and what it should be allowed to decide.
An agent might have system access to user accounts, licences and devices. That does not mean it should be able to grant admin rights, approve a costly software purchase or wipe a laptop without a person signing off.
Before switching on autonomous actions, define clear decision rights:
- Fully autonomous: low-risk, high-volume tasks such as password resets and standard software requests
- Autonomous with notification: routine changes that should be visible, such as adding a user to a standard distribution group
- Human approval required: anything involving elevated access, spend above a set limit, data deletion or security exceptions
Written down and agreed with security, risk and the business, these boundaries let you scale automation with confidence. They also give your audit and compliance teams a clear answer when they ask who approved what.
Avoid building AI silos
Many organisations start with good intentions and end up with a collection of separate bots: one for IT, one for HR, one for facilities. Each works well on its own. None of them share context.
For the employee, the experience is still fragmented. A new starter needs a laptop, system access, a building pass and payroll set up. If each request goes to a different bot with no shared view, they repeat themselves at every step, exactly as they did before AI.
The value of agentic AI grows with how well your agents work together, not with how many you deploy. Enterprise service management, where IT, HR and other departments share one service platform, gives AI agents a common foundation to work from.
Measure resolution, not deflection
For years, service desk AI has been judged on deflection: how many tickets the bot kept away from analysts. But a deflected ticket is not a solved problem. If the employee gives up and phones a colleague instead, the issue has simply moved out of sight.
When AI agents can take real action, the measures need to change too:
- Verified resolution rate: the share of requests closed with a confirmed result, such as a successful password reset or a deployed application
- Time to resolution: how long the employee waits from request to working outcome
- Quality of handovers: whether the agent escalates at the right moment and passes on full context, so the employee never has to repeat themselves
- Employee effort: how many steps, channels and repeated details it took to get help
These measures show whether AI is genuinely improving the employee experience, and they give you the evidence to justify the investment.
Where to start with agentic AI on your IT service desk
You don’t need an enterprise-wide overhaul to see results. A focused, practical approach works best:
- Pick one high-volume, low-risk request. Password resets and standard software requests are ideal first candidates.
- Check your integrations. Confirm the agent can act on the systems involved, not just read from them.
- Agree the decision rights. Decide what the agent can do alone and what needs approval.
- Run a time-boxed pilot. Measure verified resolutions and employee effort, not deflection.
- Expand what works. Apply the same pattern to the next request type, then across departments.
The platforms are ready. Ivanti has introduced agentic AI capabilities in Ivanti Neurons for ITSM, with persona-based agents designed to resolve requests and reduce service desk workload. The difference between a successful rollout and an expensive experiment is the groundwork: integration, governance and the right measures.
How Think Tank can help
As an Ivanti Premier Partner, Think Tank Software Solutions helps South African enterprises get more from their service management investment. We assess your readiness for agentic AI on the IT service desk, connect the systems your AI agents depend on and help you design governance that your security and compliance teams can support.
Ready to move from answering to resolving? Contact Think Tank to discuss where agentic AI fits in your service desk.
