This is it. Eight weeks of foundations, models, reasoning, agents, and governance — all leading to this. Today you show us what you built. Have fun with it.
Tight schedule today. Each team gets 15 minutes total: 12 min presentation + 3 min Q&A. We'll go straight through with short transitions. Break after all presentations, then we wrap up the course.
Fill this in before class. Either randomize or let teams volunteer for order. First slot is hardest, last slot benefits from seeing others. Consider randomizing to keep it fair.
Hard stop at 15 minutes — it's not fair to other teams if you go over. Practice your timing beforehand. The most common mistake is spending too long on background and rushing through the demo and evaluation. The demo is the highlight — give it space.
This is the same rubric from the project description. I'm showing it again so it's fresh in everyone's mind while watching presentations. When I'm evaluating, I'm looking at all five dimensions. The biggest differentiator is usually evaluation rigor — teams that systematically tested their system and honestly reported results (including failures) score significantly higher than teams with flashy demos but no evaluation.
Encourage students to ask substantive questions. Good Q&A makes the whole session more valuable. The reflection questions often lead to the most interesting answers. Remind students: engagement during presentations is part of their participation grade.
Transition to presentations. Call up the first team. Keep time strictly.
Let's take a step back and see the full arc. We started with how these models work (Week 1), then which ones to use (Week 2), then how to use them well (Week 3), then how to make them act in the world (Weeks 4-5), then how to do it responsibly (Week 6), and finally you built something real (Weeks 7-8). Each week built on the previous one. The projects you just presented integrate concepts from across the entire course.
Five things I want you to remember a year from now. First: context engineering is the highest-leverage skill — the same model with better context dramatically outperforms a bigger model with lazy prompting. Second: always evaluate systematically — gut feel is not enough for production. Third: agents are the future but they need guardrails — fully autonomous AI is not here yet. Fourth: governance isn't bureaucracy, it's engineering discipline. Fifth: the specific models and tools you used this semester will be obsolete within a year. The concepts — attention, context engineering, agentic patterns, evaluation, governance — will last much longer.
What to watch. Multimodal is expanding beyond text and images into video, 3D, and eventually robotics. Agent reliability is the key bottleneck — expect rapid progress here. MCP or something like it will become the standard way agents connect to tools, just like HTTP standardized web communication. Evaluation will mature from ad-hoc testing to standardized frameworks. And regulation will move from frameworks to enforceable requirements. If you stay in this space, these are the trends that will define the next few years.
This course gave you a foundation. To stay sharp: follow the blogs and newsletters for what's new, keep building with Claude Code on real problems, and go deeper if you want to understand the internals. The best way to learn GenAI is to use it every day. Find a workflow in your job or personal life where AI can help, and iterate on it. That's how you build real fluency.
Final logistics. Make sure all deliverables are submitted. Peer review is individual and confidential — be honest about contributions. And please fill out course evaluations. This is a new course and your feedback is critical for improving it next year.
Close it out. Thank the students for their work, the TAs if applicable, the guest speaker. Remind them that the skills they built here are immediately applicable. Encourage them to keep building.