Today I’m going to give a quick view on few short term realistic business cases for Agentic AI as this is going to be a major hype next year where I see many opportunities of real application especially in those businesses that struggle with hiring while having costs optimization targets and seeking for ways to release internal organizations from repetitive tasks to increase value add.
Few cases not complex to deliver that I foresee:
- Automated ServiceDesk. This is a case quite obvious, applying to IT (but not only going forward). Here a nice video from ServiceNow speaking about how that would apply. Value definitely there and possibility to automate first level activities today not yet automated.
- Claim process in Insurance, automating the first steps evaluation and assessment and leaving the latest step to human to accept/deny (at least for the first phase)
- Automating in Pharma the first phase of seeking and evaluating scientific articles that could be plague and bring only a subset relevant to deeper analyze from experts
- Automating answers for enterprises policies and regulations in the area of HR Administrator, Operations, compliance and repetitive tasks like Holidays approval
- Preventive maintenance in manufacturing with autonomous corrective actions, monitoring trends and adjusting dynamically
- Adjusting automatically inventory based on production plans, market availability, avoiding risk of out of stock. Automating procurement repetitive activities and overall supply chain automation
- Automating several steps in the sales customer experience. Here a great example from Salesforce
- Automatic software testing agentic, like for example with the recent release of products like Devin and other examples that proven that AI can help even to identify long existed bugs not detected.
Some threats:
- Losing innovative answers. Did you notice that most of the pictures we are now using look having same style? Same applies to answers in data when the source of data is too much repetitive or even worst when is generated thru AI data.
- Boundary limits and cybersecurity. Without conditions on what to do, what not, to whom answer and how, Agentic would execute out of expectations. Imagine what would be for the company reputation if one agentic would answer automatically to customers and could bring data not expected or if an HR administrative agent could reply bringing sensible information to employee or be used to reach other agents from other entities on behalf of companies
- Code quality could degrade if no proper analysis and review is happening. Repetitive paths could introduce bugs in software not easy to detect.
Some tips:
- Focus only on repetitive, non creative activities
- Set the limits in which your Agentic will be allowed to interact with the customers and between them to avoid the risk to lose control of the way they operate. The time spent here is super valid and relevant because some agentic could be turned by the requestors and deliver answers and contents unexpected
- Take the time to test with proper quality data, train on what are the limits and monitor what you automate as not critical process in first stage because could bring unpredictable results
- Decide the governance and scope of agentic and monitor the behavior.
- Remember an Agentic is going to be a sort of employee of the company making repetitive tasks so the enterprise will have to consider a certain accountability around their execution and behavior.
- Limit code software analysis to tests and consider agentic here like junior programmers that help senior programmers. Don’t accelerate too much the process to automate everything.
Agentic AI will bring from my perspective a competitive advantage and the way to compose them will bring an enormous potential to scale outside the enterprise boundaries.
GG
