Call Summary: Catholic AI Education Strategy and Alpha Pilot Reflections

Overview

Strategic conversation between Gary and Chris Tyrrell covering Chris's new Citibank role, detailed reflections on his Alpha Schools Catholic pilot, critique of Alpha/Two Hour Learning's pedagogy and product quality, vision for classical Catholic AI-enabled education, technical architecture ideas for open-source tutor infrastructure, and planning for potential fresh build approach.

Key Topics Discussed

Chris's New Role at Citibank

• Position: Running digital asset risk for Citibank's services division
• Timing: Started recently, will update LinkedIn in February
• Fit: "When I read the job description, I was like, someone looked at my resume, wrote a job description, lined it up side by side"
• Context: Banks want crypto expertise but from someone who "looks like a banker" - not the stereotypical crypto person
• Quote: "You have to have dealt with the nonsense in a professional environment that they trust... there aren't a lot of us"

Alpha Schools Catholic Pilot - One Semester Review

• Setup: Found opportunity April, parents May, commitments August, launched September 2025
• Cost: ~$20k per student (half Alpha, half other) vs. $9k typical Catholic school tuition
• Funding: Private donor investors subsidized the difference
• Key learning: "We've learned enough that if we had to do it from scratch, we would do it from scratch"
• Texas implications: Two Hour Learning priced at Texas voucher price - can't do Texas with Alpha, need different approach

Critique of Alpha/Two Hour Learning

Product Quality Issues:

• Answers don't show up
• Answer boxes don't have enough characters for actual answers
• Questions don't match the text
• Misspelled words in questions
• Board member (software PM parent): "I would never let this product get to UAT"

Pedagogical Problems:

• Optimizes for extrinsic motivation (money, competition, beating the system)
• Curriculum is primarily Common Core
• Optimizing for MAP testing speed, not actual learning
• No standardized module lengths - some hours, some minutes (not claimed "25 minutes")
• Never actually built the pedagogical hierarchy they claim ("broken up all of knowledge into 25-minute modules")

"Imago Joe" Concept:

• Alpha built in Joe Liemandt's image: competitive, money-focused
• Joe didn't like school; likes making money and winning
• Sent "ultimate SWAT force school builder" who had "no commitment to what learning was for"
• Pokemon card project example: entire focus was "let's build a Pokemon other people will buy" not creativity/self-expression
• Teachers wanted to teach tithing off first Alpha rewards; Two Hour Learning guy was "mortified"

Vision for Classical Catholic AI Education

Philosophy:

• Build for truth, not adherence
• Intrinsic motivation vs. extrinsic
• Classical view of education: orienting toward knowledge and truth acquisition
• Human formation as the goal

Economics:

• "I don't have any ambition to make my money this way... I just want it to exist"
• Goal: Scale Catholic education cheaper than current $9-12k
• Target: Texas and Florida voucher prices
• "I only want the economics to work so that it can scale fast"

Scaling Vision:

• "You could put a school in a box with very low faculty cost and modest facilities"
• "Scale almost on an unlimited basis in rural areas of Texas"
• Texas bishops see opportunity for new rural religiously-affiliated schools with state money

Content Approach:

• Leverage existing classical pedagogy work (CLT, Colby Academy, Seton)
• Wouldn't build from scratch - "Frankenstein" different pedagogies or commit to one
• Get principles of moral/spiritual development into content not necessarily facially religious

Technical Architecture Discussion

Interpretive Layer Model (from contacts in Rome):

• Three layers: LLM base → Interpretive layer (rules) → Persona layer (tutor)
• Interpretive layer is where controls/rules live
• Persona/tutor layer is specific outputs
• When models upgrade, only redo middle layer - predictable maintenance cost

"Linux of Adaptive AI Tutor" Concept:

• Gary's framing: Need open-source standard for the middle interpretive layer
• Chris: "Better to create a Red Hat on top of an existing LLM"
• Agreement: Open source standard for the middle, let people build tutor models on top
• Gary: "I'm not convinced Joe has figured out that middle thing"

Gary's Situation and Interests

Current:

• Applied AI Society work with Travis Oliphant (NumPy inventor) - building AI community in Austin
• Exploring how to get involved in Texas school choice first year
• Connected with Prenda CEO (2k/student, 200+ microschools, Joe invested)
• Knows Tauren Wells (gospel singer) who wants to build school at his church

Alpha Experience:

• Graduated stack rank first/tied for first
• Paid $600k/year
• Left because unmotivating - no future, internal deception, no product-market fit
• "Everyone's just trying to not be on Joe's kill list. It's like a Hunger Games type thing"

Philosophy:

• Thinking about making money separately from education work
• "The default is to just make money from the world and then with your abundance channel that or your free time"
• Wants role that feels supernatural/God-designed, like Chris's new job fits him

Relationship with Joe Liemandt - Next Steps

Chris's approach:

• Wants to be "very up and up with Joe"
• Will propose fork for classical version (originally proposed in May, Joe said not now)
• Expects Joe will probably say no again
• Then will raise money to build fresh
• "I don't want to do to Joe what Two Hour Learning was doing to IXL" (leveraging assets to build competitor)

Timeline:

• Investor conversations over next week (3-4 key investors)
• Conversation with Joe after that
• Sort out current school and team by February
• Target: Fraction of a product by September that can go alongside existing platform

Prenda Discussion

• Gary introduced Prenda: 200+ microschools, $2k/student/year
• Uses assessment for reading/math level, parent/guide sets goals, curated resources
• Not "adaptive AI tutor" level but good starting point for pastors Gary's talking to
• Almost confirmed on Texas TEFA vendor list (99%)
• Chris unfamiliar but interested to watch walkthrough videos

Key Quotes

Chris on his mission:

• "I don't have any ambition to make my money this way... I just want it to exist"
• "The biggest flaw in all the AI models is that they're building for adherence, not for truth"
• "Forming a billion souls. The image matters."

Chris on Alpha quality:

• "The quality level of the product itself is not hard to compete with"
• "It's in prototype... It's in Alpha Beta. No pun intended."

Chris on pedagogy:

• "It's imago Joe. It's in Persona, Joe."
• "He was there to beat the system with the kids, be their buddy"
• "The ends are so different that you have these collisions that become so obvious after they happen"

Gary on Alpha:

• "Everyone's just trying to not be on Joe's kill list. It's like a Hunger Games type thing"
• "This is not an organization that's going to trick anyone into getting a billion customers"

On collaboration:

• Chris: "Our call was providentially timed for you to receive it"
• Gary: "I see God orchestrating things... your experiment was a public service"

Action Items / Takeaways

• Follow-up call in one week to discuss "what would we build if we got enough money"
• Chris to talk to 3-4 key investors about building fresh
• Chris to have conversation with Joe about forking (expecting no)
• Gary to send Chris Prenda walkthrough videos for comparison
• Both aligned on need for brutally honest assessment of what it takes to build "civilizational infrastructure"
• Potential collaboration on kingdom-focused classical Catholic AI education platform