Decisions · updated 2026-09-07

Source plans - what Sheila + Michelle actually decided

Compiled 2026-08-19. Everything below is traced to a Fathom call. Nothing here is invented.

Which calls

Plan A - 2026-08-08 (the launch plan)

The Sept 1 target - this is the app scope

• Ship a lead-gen app by Sept 1, not the full course. The MVP drives downloads and builds the user database while 101 gets finished.
• MVP contents, verbatim from the call:
• A sign-in that unlocks free resources and lead-gen tools
• A front-facing agent on login with two paths:
• Career path: "tell us your goals" -> agent writes your plan for your first AI project
• Business path: "write up your goals" -> agent gives you a plan to create your first internal AI tool based on your IP
• A short intro video from Sheila (also starts building the video database/pipeline)
• Sept 1 = app + lead-gen content submitted to the App Store, not live. Apple review is ~10 days, and every future product launch or update carries the same ~10 day tax.
• Apple takes 15% under the Small Business Program. Accepted as fine for a high-ticket course.

Curriculum changes that the app has to accommodate

• 101 is too surface-level, especially modules 5-8. Needs more depth, not more formatting.
• New module: the agentic framework, taught as three layers using Dossier as the worked example:

1. Shared context layer (one Supabase database all agents pull from)
2. Nodal agents (agents strung in sequence, each node emits variables that feed the next)
3. Front-end UI (built in Fable) so a non-technical person can trigger the agents

• New module: case studies / the "why", structured as A-vs-B comparisons of SaaS you rent vs software you own:
• Website hosting: Squarespace -> custom AI backend
• Design: Canva / Figma / Adobe -> custom AI design agent
• Forms: Typeform -> custom AI form builder
• Database: hosted SQL -> Supabase managed with AI help
• Finance software -> replaced
• Framing: after the 101 foundations, here is what they unlocked. Ends on the template for building your first agent. Getting from the template to a working build is what 102 sells.
• Both new modules land in 101, not 102. Michelle pushed back that it was a lot for 101; the resolution was to frame them as case studies plus the "why," and use them as the on-ramp to 102.

Packaging

• 101 stays purchasable alone (Michelle held firm on this - some people will be overwhelmed).
• 101 + 102 sold as a discounted bundle, or incentivized as an upsell at the end of module 7.
• 101 = the why + the agentic framework + case studies. 102 = the how, the environment setup, the hosting.

Community and post-MVP

• Start with a community forum plus office hours, watch where people get stuck, then automate.
• Engagement is the known risk. Nomiki's platform has 300+ AI-forward users and near-dead engagement.
• Future product idea: an AI "female CTO" agent that gives on-demand technical guidance (the thing Brian does for Sheila) - explicitly parked as post-MVP, not launch.
• Positioning guardrail: this attracts people advancing a career or a business, not people chasing founder/VC status.

Project management

• ClickUp. Separate projects for: content and curriculum / content production / marketing / app development and launch.
• Every "ooh what about this" idea during the app build goes straight into a post-MVP folder, later tied to its own product and marketing plan.

Plan C - 2026-07-18 (how we work)

No product decisions in this one, but it sets the working agreement, and it directly governs how this repo gets created.

• Collaboration runs through Claude Code + GitHub, not Google Drive. No more downloading and re-uploading files.
• Branch per person. michelle-updates and sheila-updates. Sheila owns the merge into main. Local machines pull main to pick up the other person's work.
• One scoped folder per project, always. Sheila's rule, stated plainly on the call: open Claude Code inside the narrowest folder that contains the work. Tandem HIPAA code never shares a session with the website. The same discipline applies here - WTFISAI-APP is its own repo, not a subfolder of the curriculum repo.
• Michelle's setup is done. Windows 11, Claude Desktop app (not terminal Claude Code), Git for Windows installed with the "available to all programs" option, paid Claude account. Verified working.
• The curriculum repo does not need Cloudflare. Only repos that publish to the web get wired to a Cloudflare Worker. Worth noting because this app repo will need that connection.
• The blocker from that call was never cleared: Sheila still owes Michelle the shared GitHub repository. That is a prerequisite for Michelle contributing scripts and copy to the two new 101 modules, which are on the Sept 1 critical path.

What this means for WTFISAI-APP

1. This folder becomes its own GitHub repo, separate from the curriculum repo and the website repo.
2. Michelle gets a branch on it if she is writing in-app copy, which she should be - the onboarding and plan-output copy is the highest-leverage writing in the MVP.
3. The repo connects to Cloudflare for app.wtfisai.co.

Plan B - 2026-06-25 (the production and hosting plan)

• Curriculum thesis: teach AI evaluation and shared-context management, not tool tourism. Context capacity dictates tool choice - limited context automates a task, shared context builds a platform, agentic tools go gather their own input.
• Human role: manage the shared context and critically evaluate the output. That is the durable skill.
• Video production: AI voiceover platform at $40-50/mo narrating the slide decks, so a script edit on one slide does not force a re-record of the whole video.
• Hosting: a course. subdomain with PIN-based login - no password management. Free to ~50 users, then ~$5/mo per 100. (The call said course.beintandem.co; since superseded - it is app.wtfisai.co.)
• Lead magnets: simple guides ("Top 5 Ways to Evaluate AI Output") drive sign-ups.

Where the plans disagree with each other, and with the repo

These need a decision before anything gets built. Detail in 03-open-decisions.md.

1. Auth model. 06-25 said PIN login on the web. 08-08 said native iOS app with Apple sign-in and App Store billing. The repo's TODO-week-of-2026-07-06.md recommended Podia and "no login at all for low-ticket." Resolved 08-19: one account, one identity service, used by both the web app and a native SwiftUI iOS app.
2. Where the money is taken. Apple's 15% applies to digital content sold in-app. Stripe Payment Links (the current plan in TODO-week-of-2026-07-06.md) cannot be used inside an iOS app for course access without violating App Store rules.
3. Course numbering. The calls say 101 and 102. The repo files say 101 and 201 (WTF-is-AI-201-Build-It-Yourself.md). Same product, two names.
4. Price. The calls say "high-ticket." The repo says $499 course-alone / $1,350 course + tools, and wtf-products.md still lists a $1,350 4.5-month 101. Three price stories.
5. Michelle's repo access. Outstanding since 07-18 and still blocking her from contributing.
6. Timeline. Today is 2026-08-19. Sept 1 is 13 days out. Apple review is ~10 days. Submitting by Sept 1 is achievable; being live on Sept 1 is not.