March 11, 2026
10% or 10/10 drama?
Preliminary data from a longitudinal AI impact study
Hype said 3x; study shows ~10% — comments split: win, flop, or human bottleneck
TLDR: A study of 40 companies says AI use jumped 65% but output rose ~10%. Commenters split: some cheer 10% as huge, others claim bigger personal gains with newer models, while process skeptics say meetings and reviews—not coding—are the real bottleneck, urging leaders to reset expectations.
AI was supposed to turbocharge coders, but a new study across 40 companies says reality is… a modest ~10% bump in shipped code changes, even though AI usage jumped 65%. Cue the comment section: half pep rally, half roast, all drama.
The loudest camp is yelling, “Reset expectations!” One pragmatist called a 10% industry-wide lift a once-in-a-generation win, not a letdown. Another group fired back with personal war stories: one dev swore the jump from older models to “Opus/Sonnet 4.5” was “night and day,” insisting the averages miss the real action. Meanwhile, the process skeptics rolled in like hall monitors: coding isn’t the bottleneck—meetings, planning, and approvals are. As one commenter put it, humans (and their calendars) may be the next bottleneck.
A recurring theme? We’re still learning the tool. One user shrugged that every new release gets better, hinting the curve isn’t flat yet. And yes, there were memes: folks joked that AI makes the boring bits less annoying but doesn’t magically triple your output—“the easy tasks got easier; the hard parts are still human.”
Bottom line: the study says “real gains, not a miracle.” The crowd says “celebrate the 10%,” “wait for the next model,” or “fix your process first.” Pick your fighter.
Key Points
- •DX analyzed 40 companies from Nov 2024–Feb 2026 to assess AI’s impact on engineering productivity using PR throughput.
- •AI adoption rose by an average of 65% during the period, while PR throughput increased by 9.97%.
- •Teams with individual PR throughput targets were excluded to mitigate metric gaming effects.
- •Findings align with broader reports of 8–12% productivity gains, below widely touted 2–3x expectations.
- •Developers report coding isn’t the main bottleneck; planning, alignment, scoping, code review, and handoffs remain largely unaffected, and DX will continue studying long-term effects.