Seventy-eight participants, 26 per arm, building a React course scheduler in 90 minutes, then 30 minutes fixing a bug in their own code with the AI switched off. Group A worked manually.
In the News: August 6, 2026 (Evening)
A controlled study prices the comprehension gate: 14.2 minutes of friction halved the failure rate when the AI was taken away.
Evening edition
Machine-readable
Download Markdown
Story
A comprehension gate cut the failure rate in half, and the study says what it cost
Read story →
Story
Claude Code 2.1.223 closes four ways the permission prompt could be lied to
Four of today's entries are the same class of defect. A Bash permission bypass "where a crafted command could hide parts of itself from permission checks" is fixed.
Story
The term Canon 80 credits to Osmani was coined in September 2025, and the paper citing it quotes a sentence that is not there
The post is about 450 words and it claims the term outright: "When teams produce code faster than they can understand it, it creates what I've been calling 'comprehension debt'." The archive's Canon File 80 records the term arriving through…
Also this cycle
Permalink
Rohit Mehra, Samdyuti Suri, Prithviraj K Tagadinamani, Kapil Singi, Vikrant Kaulgud and Adam P. Burden, Accenture Labs
Agents That Teach: Towards Designing Incidental Learning Back into AI-Assisted Software Development
arXiv, 7 July 2026 · A position paper and prototype, ASE 2026. SHIELD instruments the coding agent to emit telemetry for every change it makes, capturing "what the agent changed, the rationale behind those changes, the alternatives it considered, and its confidence in the selected approach," then routes that to triage, probe and assessment agents. Its central claim is that "The agent's reasoning carries information that cannot be reconstructed from artifacts alone." Two of its six principles are general harness laws rather than teaching ones. Selective: "Surfacing too frequently trains developers to ignore the system." Closed-Loop: verify internalisation "rather than treating delivery as success." Evidence is zero by the authors' own statement, "No datasets accompany this submission." All seven authors are Accenture. Read it for the principles, not for a finding.