From information access to intelligent task execution — what every library professional needs to know.
I still remember the days when I would walk into a library with a question in mind, move from catalogue to shelf, and slowly piece together meaning on my own, the system gave me access, but the responsibility to interpret always stayed with me. No wonder, we built our entire professional identity around this model.
But somewhere along the way, quietly and steadily, this arrangement has begun to change.
Today, we are not just talking about better search or faster discovery. We are witnessing a shift from information access to intelligent task execution. What we once navigated ourselves, machines are now beginning to navigate on our behalf.
Let me explain this journey as I have come to understand it.
In the early phase, our systems, OPACs, databases, catalogues, they were efficient, but limited. They responded to queries, but they never understood intent. That gap between what the user meant and what the system retrieved, that was our space as librarians. That was our value.
Then came the first wave of AI. Chatbots, recommendation tools, smarter interfaces. They made things easier. They reduced effort. But they still depended on the user to drive the process. They could assist, but not act.
Now, we are entering a new phase. Agentic AI.
This is where things become interesting. And slightly unsettling.
These systems do not stop at answering questions. They take a goal, break it into steps, use multiple tools, check results, and refine their own process. In simple terms, they do the work. The model has shifted from asking and receiving to planning and executing.
By the way, this does not mean humans are out of the picture. It means the role of navigation is shifting, from the user to the system, but always under human supervision.
When I first explored how these systems actually function, I realised they are not magic. They are structured, layered systems.
First, they ingest data. PDFs, scanned texts, MARC records, web content. They clean it, structure it, and make it searchable.
Second, they think, or at least simulate thinking. They take a goal and create a plan. They break tasks into steps and track progress.
Third, they connect. To Koha, DSpace, CrossRef, ORCID, and many more systems. Their power grows with every connection.
And finally, they act. Not by giving you text alone, but by producing usable outputs, catalogue records, bibliographies, reports, corrections. Outputs that enter real workflows.
Just imagine what this means for our daily work.
Cataloguing, which once demanded hours of manual effort, is now being partially handled by AI systems that extract metadata, suggest classifications, and prepare draft records. Still, human review remains essential. Accuracy without oversight is a risk we cannot afford.
Subject classification is evolving too. Tools like Annif learn from existing catalogues and apply that knowledge to new materials. But their effectiveness depends heavily on the data we provide, especially for regional and multilingual collections.
Reference services are also changing. A single query can now trigger a chain of actions, retrieving records, analysing sources, and generating structured reading lists. Yet, when it comes to deep research guidance or sensitive queries, the human touch remains irreplaceable.
Collection development has become more data-driven. AI can identify gaps, track demand, and suggest acquisitions with measurable impact. This was difficult to achieve at scale earlier.
And then there is workflow automation, the silent transformation. Deduplication, metadata enrichment, batch processing. These are not glamorous tasks, but automating them saves significant time. In some institutions, up to 50 percent efficiency gains have been reported.
Of course, with every opportunity comes risk.
I have seen how AI can confidently produce incorrect information. Fabricated references, wrong classifications. Without proper checks, this can damage trust.
Bias is another concern. Systems learn from past data, and past data carries historical bias. If we are not careful, we risk reinforcing outdated or unbalanced perspectives.
Privacy also demands attention. User queries, reading habits, borrowing patterns, these are sensitive. Libraries have always protected this data. We must continue to do so, even in AI-driven environments.
And then there is vendor lock-in. If we lose control over our systems and data, we compromise our independence. Open standards and flexible systems are not optional, they are necessary.
All this brings me to what I feel is the most important part, the human side.
AI will not replace librarians. But it will change what we do.
Routine, repetitive tasks will reduce. But roles will evolve.
Cataloguers will supervise AI outputs. Reference librarians will engage in deeper research support. Collection specialists will rely more on data and analytics.
This means we must learn new skills. Prompting, evaluating AI outputs, understanding data, working with APIs. Alongside this, we must also engage with policy, ethics, and privacy.
These are not technical add-ons. These are becoming core professional competencies.
If you ask me how to begin, I would say start small.
Pick one use case. Test it. Learn from it.
Then expand gradually. Introduce human oversight. Train your team.
Integrate systems carefully. Build connections through APIs.
And only then think about optimisation and scale.
This is not a race. It is a structured transition.
Looking ahead, I see a phased future.
In the next few years, assisted cataloguing and automated workflows will become common. Reference queries will increasingly be handled by AI systems.
A little later, we will see multiple AI agents working together across different library functions.
And in the long run, we may see deeply integrated systems with strong institutional memory, almost like AI co-librarians.
But let us be clear about one thing.
Fully autonomous libraries are not realistic. Nor are they desirable.
What will succeed is a balanced approach. AI systems that handle defined tasks. Humans who guide, supervise, and take responsibility.
In the end, I do not see this as a story of replacement. I see it as a story of augmentation.
The library has always stood for access, equity, and intellectual freedom. These values do not change. What changes is how we deliver on them.
And perhaps, for the first time, we are not just helping users navigate knowledge. We are building systems that can navigate it with us.
That is a powerful shift. And a responsible one.
