Agentic AI is rapidly gaining traction across the data and AI landscape. If you're already using Qlik Cloud, you already have Qlik Automate.
You can start building your own intelligent agents to automate tasks, which will reduce operational overhead.
Introduction: what is agentic AI?
If you joined one of our recent webinars, you'll know something important. Agentic AI is not just another AI trend.
It represents a fundamental transformation in how we apply AI to real-world workflows. Agentic AI combines the reasoning power of large language models (LLMs) with functional orchestration: the ability to make decisions and take action across software platforms. Analytics began with basic reporting, then evolved into interactive dashboards that helped us understand what happened in the past, followed by predictive analytics and machine learning which aimed to forecast what could happen in the future. Agentic AI builds on all of this by giving LLMs the ability to act - these agents can understand context, evaluate problems, choose a strategy, and trigger actions such as modifying a script, sending a notification, or running a process.
In the world of data and analytics, this opens up exciting new capabilities. From automated root cause analysis to self-healing data pipelines and intelligent platform management.
In this blog, we walk through how we built an agentic AI. We used Qlik Cloud using Qlik Automate and OpenAI. This agent monitors reload failures, analyses load scripts and error logs using an LLM. It intelligently decides whether it can apply a fix, if so - it can resolve the issue, and it tests and promotes automatically.
Security and cost considerations
To build this prototype, we used the OpenAI ChatGPT API. This is a pay-as-you-go service.
For reference, we purchased $20 of credits and used one of the most cost-effective models (GPT-4o Mini).
It's priced at $0.15 per million tokens. During development, the total cost incurred was just $0.01.
Importantly, we only send load scripts and reload logs to the API. It's critical to be mindful of the data you transmit to any external LLM.
Avoid sharing sensitive company or customer information. If in doubt, always consult your IT or data security team.
Why I chose reload failure resolution
If you're responsible for data delivery, you'll know that reload failures are disruptive. They're time-consuming, and often trivial to fix once understood.
An agentic AI approach can:
- Reduce manual troubleshooting
- Accelerate time to resolution
- Improve pipeline reliability and uptime
- Demonstrate the practical potential of AI-driven automation
Whether your Qlik Cloud environment supports finance reports, operational dashboards, or embedded analytics, something happens. Proactive remediation of reload issues boosts confidence and reduces fire-fighting.
Project goals
As part of this project, we set out to achieve three objectives. These were key technical and operational objectives:
1. Demonstrate automated orchestration with Qlik Automate
Our aim was to showcase how Qlik Automate can manage an entire workflow. From detecting issues to deploying solutions without manual intervention.
This kind of orchestration allows universities to make repetitive processes more efficient. It saves time and reduces operational overhead.
2. Incorporate AI-driven decision making
We wanted to go beyond traditional automation. We integrated a Large Language Model (LLM).
It could intelligently decide within set safety thresholds whether an identified issue should be resolved. Or flagged for human review.
This approach balances efficiency with control. It makes sure that critical decisions are made transparently and with oversight.
3. Create AI-based code review and self-healing
The project also tested the ability of the LLM to analyse error messages. And relevant code.
It autonomously suggests and implements fixes. Then automatically tests those fixes in a safe, controlled environment.
This "self-healing" capability has the potential to drastically reduce downtime. It frees up technical teams to focus on more strategic work.
The architecture at a glance
We built the solution using:
- Qlik Cloud Automations for orchestration and logic flow
- OpenAI (GPT-4o) for reasoning and decision-making
- A confidence scoring mechanism to decide whether to act autonomously or escalate
Step-by-step: building the agentic AI reload assistant
1. Scheduled monitoring
A scheduled Qlik App Automation checks for failed reloads using the List Reloads block. It uses FAILED as a filter.
(In production you should only look at production reloads, for specific processes)
For each failure, we retrieve:
- App ID
- Load script
- Reload log content, trimmed to only include the error information.
2. LLM evaluation: can it be fixed?
We send the load script and error log to OpenAI. We use the Chat Completion block.
The model is asked to evaluate the likelihood that the script can be fixed. It can be automatically fixed.
It responds with a confidence % as a decimal which is convert to a number.
If the score is below 90%, the issue is flagged for manual review.
3. If fixable: request the revised script
If the model scores 90% or higher, we prompt it again. We ask it to provide only the corrected Qlik load script
(With strict instructions to omit explanations).
The new script is added to the app, and reloaded.
We then check whether the reload was successful.
4. Validation and safe testing
If the reload failed again, we then revert the script to the previous version, and notify the relevant administrator.
If the reload succeeded, we just notify the relevant administrator that the Agent changed. We detail what is was and what it changed to.
Why this is just the beginning
This is a simple demonstration of using Qlik Automate and LLMs. It resolves simple reload failures.
But to make the process more robust and truly intelligent, we can improve the agent. We can do this in several ways:
Schema change detection
Automatically inspect source tables or QVDs. Compare their metadata against the expected schema before reload.
If changes are detected, the agent can flag or even adjust the script. It does this accordingly.
(Can you imagine not needing to worry about a user changing a manually maintained spreadsheet format)
Assessment and notification of root cause
You can have your failed reloads reviewed, categorised, relevant people notified. Suggested resolutions direct to your inbox.
Dynamic retry logic
If the root cause is classified as intermittent (e.g. temporary API timeout), retry the reload. Do this with delay and monitor outcome.
Change tracking and version control
Log each LLM-generated change and associate it with reload outcomes to improves learning and governance.
These improvements not only reduce the risk of reload failure, they also bring us closer to a self-aware, self-correcting data platform.
Extending agentic AI across your data stack
Our team is actively exploring how to extend this framework across other tools such as Snowflake, Databricks, and Power BI.
You could go much further with Qlik Automate for agentic AI, in terms of:
- AI-driven metadata classification and tagging
- Proactive monitoring and resolution or notification of data quality issues
- Intelligent scheduling based on usage patterns and resource availability
- Smart alerts that recommend or trigger remediation workflows
- Cross-platform diagnostic agents for hybrid data environments
Want to see it in action?
If you're curious about agentic AI or Qlik Cloud automations, we'd love to talk. We can demo our solution, evaluate your use cases, and help you take first step.
Comments