Students name the problem, then build the agent

This was one of the larger LUBS group, 41 students answered the opening question and 50 took part in polling across the session. They did not arrive neutral. The opening word clouds put curious, excited, useful and powerful directly alongside privacy, replacement, fake news, data security and laziness. Some students were worried about being replaced by the thing they were about to spend two hours learning. A session that only sold the upside would have lost half the room in the first ten minutes.

• A hands-on session for a large LUBS student group
• Live polling on confidence, current AI use and concerns
• Clear framing of generative AI versus agentic AI
• Use case discovery grounded in personal and employability frictions
• Group ideation, naming and shaping agent concepts
• Discussion of business value, risk and guardrails

The session opened with polling rather than presenting, which meant the concerns were on the screen in front of everyone before any claims about AI had been made. Privacy, accuracy, fake news, over-reliance and job replacement all came up in the students' own words. Naming them early made the rest of the session easier, nobody had to sit on an objection for two hours. The framing work was generative AI versus agentic AI: the difference between asking a tool for something and giving it a job. Then the room was pushed to find a friction of their own rather than a business case in the abstract. What came back was unfiltered and better for it. Pascal - a private pet chameleon that helps with job applications. CV Buddy, aimed squarely at the repetitive grind of tailoring applications. EasyTravel for trip planning. PantryBoss for food and pantry planning. GymGPT for fitness. Hungri for meals. And Lovebug, for finding a partner. Two things stand out. First, employability came up twice independently, in Pascal and CV Buddy - the job hunt is the friction this cohort feels most. Second, nobody needed a template to name an agent. Given a pain, a group of students will define a gain and give it a name in minutes. That instinct is the thing worth building on. The concerns raised at the start did not go away, and that turned out to be useful: guardrails, validation and privacy stopped being theory and became part of designing the thing.