State of AI Software Delivery 2026
Headlines, social variants, boilerplate, high-resolution figures, and brand assets — for journalists, analysts, and anyone covering this report.
Embargo: Public from publish date 2026-05-17. No embargo. · Press contact: research@newlightai.com
Canonical URL
Always cite the canonical URL. The page evolves from Volume 0 to Volume 1 at the same URL; readers who cite Volume 0 today will automatically cite the most current version when crawlers re-fetch.
https://www.stride.page/research/state-of-ai-software-delivery-2026
Alternative headlines
Pick whichever angle best fits the publication. Every headline is factually accurate and avoids overclaiming.
Stride 2026 — Volume 0: the AI coding-productivity numbers range from +55% to −19%, depending on what you measure
Volume 0: pre-registering 1,500 surveys and 14,000 telemetry stories before looking at the data
AI software-delivery research, calibrated: where Peng 2023 ends and METR 2025 begins
The 39-point self-vs-measured productivity gap, and the study that will publish whether it closes
Stride publishes its pre-registration before its findings — Volume 0 of the State of AI Software Delivery 2026
Twitter — single post
The two most-cited RCTs on AI coding productivity report +55% (Peng 2023) and −19% (METR 2025). Volume 0 of State of AI Software Delivery 2026 is the honest landscape synthesis + pre-registered design for Stride’s 2026 primary study.
Twitter — thread starter
Pairs naturally with a screenshot of the effect-range chart (Figure 2 on the report page).
AI productivity research is broken. The most-cited RCTs disagree by ~75 percentage points. We pre-registered a study to figure out why. Volume 0 published today —
Today we’re publishing Volume 0 of the State of AI Software Delivery 2026. The most-cited RCT-style productivity studies on AI coding assistants disagree by a factor of four — Peng (Microsoft/GitHub) measured +55.8%, METR measured −19% on the same population question. Volume 0 doesn’t resolve that contradiction. It synthesises what the four landmark studies (Peng 2023, METR 2025, DORA 2024, Stack Overflow 2024) actually claim, and pre-registers the design for our 2026 primary study (1,500-person Prolific survey + 14,000-story telemetry analysis at our customer base) before any data is touched. The hypotheses we register today are the hypotheses we’ll test in July. The dataset will publish under CC-BY-4.0. Methodology, pre-registration, and FAQ are all in the body.
Hacker News — title
State of AI Software Delivery 2026 — Volume 0: pre-registered design
Hacker News — author seed comment
Posted right after submission to set context, surface caveats up front, and invite questions.
Author here. Two notes: (1) Volume 0 is intentionally a landscape + pre-registration; the primary findings ship as Volume 1 at the same URL when the survey closes in July. (2) Every number in the body is sourced to a public RCT or survey — no proprietary Stride numbers yet. The interesting tension is the +55% / −19% gap between Peng 2023 and METR 2025, which is largely task-type (greenfield vs context-rich legacy) not study quality. Happy to answer questions about the pre-registration, dataset license (CC-BY-4.0), or methodology.
Mastodon
Volume 0 of State of AI Software Delivery 2026 is up. Landscape synthesis of the four landmark public studies (Peng 2023, METR 2025, DORA 2024, Stack Overflow 2024) + pre-registered design for the 2026 primary study. Dataset under CC-BY-4.0 when Volume 1 ships in July. https://www.stride.page/research/state-of-ai-software-delivery-2026
Boilerplate — short (≈50 words)
Inline at the bottom of a release or as the publisher footer.
Stride Research is the research arm of Newlight Solutions, publishing pre-registered studies on AI-native software delivery. State of AI Software Delivery 2026 is its annual flagship.
Boilerplate — medium (≈100 words)
For an 'about the publisher' card or press-bio section.
Stride Research is the research arm of Newlight Solutions, publishing pre-registered studies on AI-native software delivery. The annual State of AI Software Delivery report combines a public landscape synthesis (Volume 0, May 2026) with primary survey + telemetry findings (Volume 1, July 2026) at a single canonical URL, so citations to the work age forward as the dataset deepens. Every dataset publishes under CC-BY-4.0 with a Zenodo DOI for permanent citability.
High-resolution figures
Each figure regenerates the inline-SVG chart at 2400px wide as a transparent PNG, suitable for print or large-format syndication. Right-click → Save link as.
- Download PNG (2400px) →Figure 1 — Adoption rates across surveysBar chart of Stack Overflow 76%, McKinsey 65%, and DORA elite-quartile adoption rates.
- Download PNG (2400px) →Figure 2 — Effect-size range (Peng vs METR)Floating-bar chart with Peng +55.8% in green and METR −19% in red, non-overlapping confidence intervals.
- Download PNG (2400px) →Figure 3 — Perception–measurement gapFour points on a horizontal axis showing the 39-percentage-point gap between developer self-perception and measured result.
- Download PNG (2400px) →Figure 4 — Literature timeline 2023–2026Timeline showing Peng, McKinsey, Stack Overflow, DORA, Octoverse, Anthropic Economic Index, METR, and Stride V0 + V1 markers.
Brand assets
Stride + Newlight logos for use alongside coverage. All assets are SVG; PNG raster fallbacks live at the matching paths under /brand/logos/.
- Stride lockup — horizontal (black, for light backgrounds)Open SVG →
- Stride lockup — horizontal (white, for dark backgrounds)Open SVG →
- Stride logomark (parallelograms only — square)Open SVG →
- Stride lockup — stacked (for narrow placements)Open SVG →
Executive summary PDF
6-page server-rendered PDF: cover, TL;DR, key findings, effect-range chart, methodology one-pager, citation + references.
Download executive summary PDFReach out for an embargoed pre-brief, additional figures, or a researcher Q&A: research@newlightai.com.