AI agent builds 208-card CKA Anki deck with every card verified against Kubernetes docs

According to a post on dev.to, a developer ran an experiment in which an autonomous AI agent had to design, build and sell a small digital product while operating under strict honesty rules. The finished product is a 208-card cloze-deletion Anki deck for the Certified Kubernetes Administrator (CKA) exam — and, unlike most AI-generated study material, every card in it was machine-checked against the official Kubernetes documentation before release.

What the deck contains

Each card is a single sentence taken verbatim from the official Kubernetes documentation, pinned to version 1.35, with one term blanked out for the learner to recall. The reverse side of every card links to the exact documentation page and section the sentence came from, so learners can jump straight to the original context.

The product is listed on Gumroad for $7 and ships in three formats: the Anki deck itself, a PDF and a CSV export.

How the cards were verified

The most notable part of the workflow is the verification step. According to the dev.to post, a script re-fetched every source page after the deck was assembled and confirmed that each quoted sentence actually appears in the documentation. All 208 cards passed that check.

On top of the automated check, the agent audited its own output by reading random samples of cards and progressively tightening its inclusion filter. In the final sample, one card out of 32 was judged to lack enough surrounding context, and it was removed. The author is explicit that this was sampling-based quality control, and that some unclear cards may still remain in the deck.

The stated limitations

The post is unusually candid about where the product falls short. The cards cover facts and definitions only, while the CKA is a practical, hands-on exam — so the deck cannot substitute for lab practice. Coverage is also skewed relative to the exam blueprint: troubleshooting accounts for 30 percent of the CKA exam but only 22 percent of the cards.

The author also discloses a direct conflict of interest: they operate the agent and are selling the product themselves, so they are not a neutral reviewer of its quality.

Licensing

Card text is drawn from the Kubernetes documentation, which is copyrighted by The Kubernetes Authors and licensed under CC BY 4.0. The product is not affiliated with or endorsed by The Linux Foundation or CNCF, and both CKA and Kubernetes are acknowledged as trademarks of The Linux Foundation. Reported errors or unclear cards will be fixed in a free update.

Why it matters

This experiment is a small but instructive example of how to make AI-generated content trustworthy. Rather than asking users to trust a model's output, the workflow closes the loop: an independent script re-checks every card against the authoritative source, and the product page lists what the deck does not cover. That generate-then-verify pattern is applicable far beyond flashcards, from documentation summaries to compliance reporting.

It also models honest marketing for AI products. Publishing known coverage gaps, describing the limits of the quality-control process and stating the seller's bias up front is a transparency standard that most AI-generated products still fail to meet.