Sprint Estimation Reality 2026

https://www.stride.page/research/sprint-estimation-reality-2026

Software estimates have been ~30% optimistic on average for 43 years. AI may be hiding that, not fixing it. Volume 0 of Sprint Estimation Reality 2026 synthesises the literature and pre-registers a 500-person study of calibration in the AI era.

43 years after Boehm published the Cone of Uncertainty, software estimation has not gotten meaningfully better. The 2026 question isn't whether AI fixes that — it's whether AI hides it. Volume 0 just dropped —

Today we're publishing Volume 0 of Sprint Estimation Reality 2026.

The academic literature on software estimation is more mature than most engineering thought leadership recognises. Boehm's Cone of Uncertainty (1981) still empirically holds. Halkjelsvik & Jørgensen's 2012 meta-analysis of 200+ time-prediction studies put the average overrun at ~30% with no clear improvement over decades. Jørgensen's 30-year program of research is the most comprehensive empirical body of work in the field. Yet almost none of this gets cited in modern Agile or DevOps writing.

Volume 0 reads what 43 years of research already established. Volume 1 (Q4 2026) lands the primary findings from a 500-person study of AI-era estimation calibration. The hypotheses are pre-registered now. The Stride sprint-capacity-calculator (free, no-signup) is the measurement instrument; we're reading the literature and the pre-registration before we look at any data.

If you're a senior software-delivery practitioner and want to participate in Volume 1 — Prolific arm or organic arm — research@newlightai.com.

Sprint Estimation Reality 2026 — Volume 0: pre-registered landscape synthesis

Author here. Two notes: (1) Volume 0 is intentionally a landscape synthesis + pre-registration, not a primary-findings report. The 500-person calibration study fields Q3 2026; Volume 1 ships at the same URL when it closes. (2) The four landmark studies in the comparison table (Boehm 1981, Halkjelsvik & Jørgensen 2012, Jørgensen 2014, Eveleens & Verhoef 2010) are all real, peer-reviewed, publicly accessible. The interesting tension Volume 1 tests: does AI-aided estimation improve calibration (accuracy), or only confidence? Happy to answer questions about the pre-registration, the calibration instrument, or the dataset (CC-BY-4.0).

Volume 0 of Sprint Estimation Reality 2026 is up. 43 years of estimation research synthesised (Boehm, Jørgensen, McConnell, Lichtenstein) + pre-registered design for a 500-person AI-era calibration study. Dataset under CC-BY-4.0 when Volume 1 ships Q4 2026. https://www.stride.page/research/sprint-estimation-reality-2026

Stride Research is the research arm of Newlight Solutions, publishing pre-registered studies on AI-native software delivery. Sprint Estimation Reality 2026 is the second Volume 0 in the 2026 research series after State of AI Software Delivery 2026.

Stride Research is the research arm of Newlight Solutions, publishing pre-registered studies on AI-native software delivery. The 2026 research series combines public landscape syntheses (Volume 0) with primary survey + telemetry findings (Volume 1) at a single canonical URL per study, 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.