aiagentrank.io
Subscribe

How we review courses

The editorial framework that powers /learn. Same tone, different rigor, than our agents and tools methodology.

The three-tier system

Courses are sorted into three editorial tiers. The tier is shown as a chip on every course card. The methodology behind each tier is below.

Tier 1 — Hands-on reviewed

A member of our editorial team took the course end-to-end (every module, every lab, every assignment) and signed off personally. The course's pros, cons, bestFor, and notIdeal lists are based on the editor's actual completion — not on instructor marketing or aggregated student reviews. We will not move a course into Tier 1 without this completion.

Tier 2 — Curated from public signals

The editor reviewed the syllabus, instructor credentials, student outcomes where published (employment data, completion rates), and recency of updates. The editor did not complete every module — but did spot-check at least 30% of content. Pros and cons are derived from public student review aggregates, cross-referenced with the editor's spot-checks.

Tier 3 — Listed for reference

A course we list because it appears in many comparable curricula, but we have not personally vetted. Listed for completeness; not recommended. We disclose this clearly so a reader knows the boundary of our review.

Conflict-of-interest policy

We use affiliate links on course pages — when a reader clicks through and enrolls, we may receive a commission. This is the same model as our agent and tool reviews. The disclosure appears on every course detail page, and we follow these rules to keep editorial decisions independent of affiliate economics:

  • · Tier is assigned before we check affiliate eligibility.
  • · We list courses that have no affiliate program (e.g., Anthropic's GitHub-hosted tutorials) with the same prominence as courses that do.
  • · When two courses on a topic are roughly equivalent, we list both — we do not pick the higher-paying one.
  • · Sponsored placements (if we ever accept them) are labeled clearly and editorially separate from organic reviews — same policy as our main methodology.

What we don't review

Deliberately out of scope (as of mid-2026):

  • · University degree programs and full Master's degrees. The decision factors there are too different from short-form courses (tuition, accreditation, location) to fit our framework.
  • · Bootcamps over 80 hours. Same reason — the framework treats <20 hour courses; longer-form is a different review job.
  • · Courses without a clearly identified instructor. If we can't name the person teaching, we can't evaluate authority.
  • · Courses on languages other than English or German. Coming as we expand i18n coverage.
  • · Courses that haven't been updated in over 24 months. The AI field moves too fast for that to be acceptable.

Refresh cadence

Every course on this site carries a Last verified date. Our refresh policy:

  • · Tier 1 courses are re-checked every quarter — the editor confirms the course is still live, the syllabus hasn't materially changed, and the instructor is still listed.
  • · Tier 2 courses are re-checked semi-annually.
  • · Tier 3 courses are checked annually.
  • · Any course that hasn't been verified within its window gets a visible "Verification overdue" flag on its page until re-reviewed.

What makes a Tier 1 pick

A course earns Tier 1 status when it scores well on five axes — all of which are checked by the editor who took the course:

  1. Instructor authority. The instructor is a recognized practitioner in the field (e.g., the framework's author, a senior engineer at a frontier lab, a published researcher).
  2. Syllabus depth at the stated level. Beginner courses cover beginner topics thoroughly; advanced courses go deep on production concerns (evaluation, observability, failure modes).
  3. Hands-on labs that work. Code labs run without environment setup hell; assignments produce something a learner can show.
  4. Honest about model limits. Courses that pretend AI can do anything get docked. We favor instructors who explicitly cover where models fail.
  5. Recency. Material is updated to current models (GPT-4o / Claude 3.5 / Gemini 2 in mid-2026), not stuck on 2023-era examples.

Disagree with a review?

Feedback is welcome — especially from instructors whose courses we've reviewed and from learners who've completed them. Reach out at editors@aiagentrank.io or via our contact form. We update reviews based on feedback we can verify.

How we review courses — Learn methodology | AI Agent Rank · AI Agent Rank