# AI in Teaching, Entrepreneurship, and Personal Workflow with Kerp
- Session: b7f1c22f-b6a4-4c0e-8179-6cbe5111212a
- Channel: Discord #🐣│cohort-voice
- Started: 2026-06-11T17:32:36.123Z
- Ended: 2026-06-11T18:10:15.563Z
- Participants: takekek, duckanbro, ECWireless, Kerp, Aphilos • Pharo
- Tags: ai-in-education, applied-ai, workflow-automation, context-management, model-orchestration, student-assessment, claude-code, canvas-lms, startup-judging, future-of-work
## TL;DR
Kerp discussed how AI is reshaping university teaching, student assessment, startup support, personal tooling, and creative execution. The conversation focused on proxy collapse in education, AI as a patient tutor, context and memory management, model orchestration, AI-assisted grading and judging workflows, and the broader shift from execution scarcity toward taste, judgment, and distribution.
## Summary
The meeting was a fireside-style conversation with Kerp, a University of Virginia McIntire School of Commerce professor, assistant director of student entrepreneurship, and longtime RaidGuild member. He described his current work teaching information systems, entrepreneurship, and applied AI, as well as his background in marketing, startups, blockchain, and legal technology.
A major theme was how AI is forcing educators to rethink assessment. Kerp described 'proxy collapse': traditional assignments such as reflection papers or written submissions used to function as proxies for understanding, but AI makes those proxies unreliable. In response, he is exploring AI-run oral exams and interview-style assessments, including Claude-based skills that can evaluate whether students actually understand course material.
Kerp emphasized AI as an infinitely patient tutor, especially in tools like Claude Code. He said working with AI has helped him learn more by doing, because he can continually ask why something was done and receive explanations without exhausting a human mentor. He also discussed how AI helps him plan syllabi, design interactive class experiences, and quickly generate course structures that would otherwise require much more teaching experience.
He shared several active projects and tools. These include Course Tools, a Canvas-integrated suite for managing courses, polls, peer feedback, prediction markets, surveys, AI study tools, syllabus management, grading dashboards, and other classroom utilities. He also described Hypercontext, a tool for preserving and moving personal/project context across machines and AI tools, and Juris, a revived local AI negotiation client built from older legal-tech and dispute-resolution work.
The group discussed current pain points, especially context management, memory, and model orchestration. Kerp noted that he often uses the strongest available model for tasks that could likely be handled by cheaper or smaller models, but lacks confidence in choosing the right model automatically. OpenRouter was mentioned as a possible direction.
Kerp also reflected on student anxiety about AI and jobs. He argued that while students are understandably nervous because they are investing in education to get better jobs, the narrative that AI will simply replace all work is overstated and often tied to company incentives. His view is that roles will shift, with more emphasis on reskilling, QA, product judgment, and people who understand systems deeply enough to absorb and validate much higher output.
He expanded this into a broader theory that as AI drives down the cost of execution, the remaining differentiators become taste, judgment, sequencing, and distribution. He suggested this expands the design space for many small, useful, previously uneconomical public-good or niche tools, similar to how blockchain expanded the design space for DAOs and digital coordination.
Near the end, Kerp described practical AI workflows: LLM-based first-round judging for a startup competition, a Claude skill for grading handwritten exams using rubrics and instructor feedback loops, skills for filling out recurring university forms, research workflows for podcast guest preparation, AI-assisted pitch decks, an 'idea writer' interview skill for fleshing out concepts, and heavy use of AI in drafting a book from his own outlines, frameworks, and conversations.
## Action Items
- Share AI tools and project links: Kerp shared links to Juris, Hypercontext, and Course Tools in the voice-channel chat for others to review. (owner: Kerp)
- Explore OpenRouter or model-routing options: The group raised model orchestration as a pain point, including choosing cheaper or smaller models for simpler tasks instead of always using the strongest model.
- Follow up on Kerp's AI workflows: Potentially collect more detail on Kerp's Claude grading skill, startup judging workflow, Canvas-integrated Course Tools, and idea-writer skill for future cohort learning or documentation.
## Notable Quotes
- Kerp: "The AI breaks the thing where it used to be too costly to fake things."
- AI has undermined traditional assignment formats that were used as proxies for student understanding.
- Kerp: "It is an infinitely patient tutor."
- AI tools can teach and explain technical choices repeatedly in a way human mentors often cannot sustain.
- Kerp: "I feel like I know so much more since I started working with Claude for every second."
- Kerp feels that constant interaction with Claude has accelerated his practical learning.
- Kerp: "The students are understandably anxious."
- Students are worried about whether AI will undermine the job-market value of their education.
- Kerp: "Code generation is a task that engineers do, but it is not actually their primary function."
- Engineering value lies more in systems thinking, process understanding, and judgment than simply producing code.
- Kerp: "If everybody can do that part, just the execution piece, then what's left?"
- As AI commoditizes execution, differentiation shifts toward taste, judgment, sequencing, and distribution.