For parents

Start at $20/month. Earn the upgrade.

Students can start free, but serious project work quickly benefits from a paid subscription. Hardware and high-tier subscriptions come later, after the student has proven they will use them well.

Screenshot of AI subscription usage limits — Usage limits are part of the learning curve. Students should learn to spend compute intentionally.

Investment roadmap

Upgrade only when the work creates the need.

Start with the smallest useful tool stack, then match the subscription or hardware to the student's current level of output.

Try the workflow

Free

Good for installing Codex, testing the starter harness, and seeing whether the student actually returns to the work.

Upgrade signal: limits interrupt the same project more than once.

Start building

$20/month

The practical first tier for Level 1 and early Level 2: Codex sessions, harness edits, sequencing practice, and first project builds.

Upgrade signal: the student has active repos, datasets, or a site they are improving weekly.

Scale serious work

$100/month

Useful once longer coding sessions, website builds, research planning, and project review are happening often enough that the smaller plan slows momentum.

Upgrade signal: the subscription limit is slowing a student who already works with discipline.

Control compute cost

$200/month, lab access, or hardware

For advanced students doing high-volume work: orchestration, local LLMs, lab machines, or a high-RAM computer can move bulk tasks off expensive frontier compute.

Upgrade signal: the student can explain exactly what work will run locally and why.

Screenshot estimating counterfactual model API costs — At the advanced stage, students learn why frontier models should direct the work while bulk analysis can move to local or lab compute.

Advanced compute

Why local and lab machines matter later.

Frontier models are best used for judgment. Let them plan the approach and review the student's choices. Once a student starts running large research sweeps or repeated document analysis, the expensive part is often volume.

That is when local LLMs or Counselor Jay's lab machines become useful. They let the student route bulk work to available compute while keeping the strongest subscription models focused on the decisions that matter most.

Use frontier models for strategy and verification.
Use local or lab compute for high-volume drafts, scans, and analysis.
Buy hardware only after the student has proven the workflow.

Investment rule

Do not buy the machine before the habit.

A high-RAM MacBook or AI workstation can be powerful. It is also premature for most students at the beginning. Start with Codex, the starter harness, sequencing, GitHub, and portfolio proof. Upgrade only when the student's work creates the need.