Skip to content
Day0
Get your free report

The research behind Day0

Figures are as reported by each source. Vendor data is labeled. Re-verified before launch.

FindingFigureSource
Output grew far more than deliveryLines of code 8.4×, commits up to 3.4×, releases about 1.3×, no detectable change in usage so far[1][2]
Review became the queueMedian PR review time +441.5%; PRs merged with no review +31.3%

Vendor data, correlational

[3]
Problems grew faster than outputIncidents per PR +242.7%; deployments per week −11.7%

Vendor data

[3]
Testing is overloadedTime in QA +300.6%

Vendor data

[5]
Even AI labs hit the wallCI job volume up 25× in six months; writing code “no longer the constraint”[6]
Coding is a minority of the workWriting and testing code is 25–35% of idea-to-launch time[7]
Redesign beats toolsAssistants alone 10–15% gains; with end-to-end redesign 25–30%[7]
Feeling fast is not being fastExperts believed AI made them 20% faster; measured 19% slower[8]
Trust depends on checking48% trust AI output when they can easily verify it; 78.7% say source attribution matters[9][10]
Review work is growing67% of developers spend more time reviewing AI-generated code[11]
A low-risk lane works at scale331,000+ changes landed; reverted at a third of the rate, incidents at a fiftieth, median time in review −35%

The lane selected low-risk changes, so not causal

[12]
Risk can be predictedFlagging 10% of changes caught 60% of incident-causing ones[13]
Throughput without review capacityPRs per developer 2.09×; total PRs 3.1× while reviewers grew 1.5×; thorough human review fell from about 39% to 21%[14]
Agent projects fail on value and controlOver 40% of agentic AI projects predicted to be cancelled by end of 2027[15]
Shipping is not adoption80% of features are rarely or never used; customers with 25+ in-app guides used 25% more features daily[16]

Sources

  1. 1

    Demirer, Musolff and Yang, Writing Code vs. Shipping Code, NBER working paper 35275 (2026)

    https://ideas.repec.org/p/nbr/nberwo/35275.html
  2. 2
  3. 3

    Faros AI, AI Engineering Report 2026 (vendor data)

    https://www.faros.ai/research/ai-acceleration-whiplash
  4. 4

    Faros AI, The AI Productivity Paradox (2025, vendor data)

    https://www.faros.ai/blog/ai-software-engineering
  5. 5

    Faros AI, The Speed Trap (Q3 2026, vendor data)

    https://www.faros.ai/blog/ai-speed-trap-takeaways
  6. 6
  7. 7
  8. 8

    METR, developer productivity study and 2026 design update

    https://metr.org/blog/2026-02-24-uplift-update/
  9. 9
  10. 10
  11. 11
  12. 12

    Meta, Automating Low-Risk Code Review at Meta: RADAR (2026)

    https://arxiv.org/pdf/2605.30208
  13. 13
  14. 14

    He et al., AI Writes Faster Than Humans Can Review (2026)

    https://arxiv.org/pdf/2607.01904
  15. 15

    Gartner prediction on agentic AI project cancellations, via W.Media

    https://w.media/over-40-percent-of-agentic-ai-projects-could-face-the-axe-by-end-of-2027-gartner/
  16. 16
  17. 17
  18. 18
  19. 19
  20. 20