02 · What we found

The direction was right. The delivery system made it difficult to realise.

Learning Yogi had strong product knowledge, robust engineering practices and extensive tests. The path from an idea to a tested, releasable improvement was too slow and fragmented.

What already worked

Strong product and engineering assets

  • Well-structured implementation and broad automated checks
  • Rich timetable and curriculum knowledge
  • Repeatable test data and interface examples
  • Clear backend product areas and live-environment checks
  • Active specifications grounded in real product complexity
What made delivery difficult

Coordination and navigation overhead

  • Mandatory reading and planning disproportionate to change size
  • Very large files containing several different responsibilities
  • Several parts of the product able to change the same information
  • Checks that were slow, incomplete or not consistently required before release
  • Extensive documentation without a dependable current map
Three evidence tracks

The friction appeared at every handoff.

None of these findings says the product is simple. They show that the same complexity was being paid for repeatedly, in the wrong places.

Instructions

Process crowded out the product

Before reaching the relevant feature, a small frontend change could involve roughly 9,748 words of guidance. A routine backend change could involve about 11,332.

analysis/investigation/2026-08-04_18-48-25…
Code structure

Finding the right place was hard

The frontend contained 54 live files over 1,000 lines; the backend contained 30. Several central areas mixed many decisions inside files thousands of lines long.

analysis/investigation/2026-08-04_19-10-49…
Release confidence

Test volume did not create a clear release path

The frontend carried broad, expensive checks. The backend had thousands of tests that were not consistently required before publishing a release.

analysis/investigation/2026-08-04_21-29-15…
Context pressure

Instructions and product code competed for the same attention.

The goal is not “no instructions.” It is to help the agent reach the right part of the product quickly, with the guidance needed for that change.

Frontend fixed reading path
9,748w
Backend routine path
11,332w
Backend Python path
14,130w
Guru root router
75 lines
→

The product foundations were stronger than the delivery experience.

The evidence consistently shows valuable product knowledge and quality assets connected by a route that was difficult to navigate, verify and release through.