# dora-google-cloud-roi-ai-assisted-software-development-j-curve-2026-04-21

## Veille

Joint **DORA × delta** report (Google Cloud Professional Services), 60 pages, version **v. 2026.1** (citations February 2026, PDF created April 21, 2026), license **CC BY-NC-SA 4.0** — the first official **DORA ROI** framework dedicated to AI in the SDLC, with an **interactive calculator** at dora.dev/ai/roi/calculator. Pivotal thesis: ***"AI is an amplifier"*** — AI **amplifies** simultaneously the strengths of high-performing organizations and the dysfunctions of struggling organizations; it does not create performance, it **multiplies it where it already exists**. New central concept: the ***J-Curve of AI value realization*** — every AI adoption goes through a **temporary productivity dip** (learning curve + verification tax + pipeline adaptation) before **exponential growth**, a metaphor for the *"tuition cost of transformation"* to be **explicitly budgeted**. Reference calculation: organization of 500 FTE / fully loaded salary $176k / 12.5% time saved per developer (≈ 1h/8h day) → **value $11.6M / investment $8.4M / ROI 39% / payback period 8 months (0.7 year)**. Modeled costs: licenses ($250/user/year), additional API ($80/user/year), training ($9,600/user/year), infra ($100k/year), J-Curve cost ($3.3M for a 15% drop over 3 months). Modeled value: **headcount reinvestment capacity** ($11M — freed capacity to reinvest, **NOT headcount reduction**), revenue from extra feature deployments ($990k, based on a 33% idea success rate, Larsen 2023), **negative downtime impact** (−$344k, "instability tax"). **Explicit reinvestment strategy**: ***"we strongly recommend organizations do not adopt a headcount-reduction strategy"*** — reinvest in innovation, retain talent, capitalize on institutional knowledge. Five pillars of value: Productivity / User Experience / Cost Efficiency / Developer Experience / Business Growth (from most direct to most indirect, *cumulated business value*). Five systemic keys of adoption: **Trust + Platform + Data + Users + Guardrails**. Two-phase roadmap: (1) **Build context layer (CapEx)** — quality IDP + healthy data ecosystems; (2) **Empower human in loop (OpEx)** — context engineering + trust in AI. Indicators: leading = experiment frequency + deployment frequency; stability gauge = change failure rate + rework. Three scenarios to model (Conservative 0.8 value × 1.5 cost / Realistic 1.0 / Optimistic 1.2 × 0.8). External data mobilized: 78% of executives report ROI on ≥1 gen AI use case (Google Cloud), 88% of early agentic AI adopters see positive ROI, **35-40% greenfield productivity vs ≤10% brownfield/legacy** (Stanford), inference cost ÷280 between Nov 2022 and Oct 2024 (Stanford AI Index 2025), **727% ROI over 3 years** for Google Cloud AI customers, average market AI payback of **8 months**. Acknowledged weaknesses: *"all models are wrong"* — the model needs contextualizing, the calculator needs adjusting; risk of double-counting value (time saved → both avoided hire AND extra revenue); a "loose" user experience link, hence excluded from the calculator. **Deontological insight**: ***"We don't measure AI by the code it writes but by the bottlenecks it clears"*** — measured by bottlenecks cleared, not code volume. **Major relevance** for CIOs/CTOs who need to build a defensible AI business case for a CFO/board; for France/Europe, to be articulated with Wescale (realistic X3-X4), Tatsyi/Raiffeisen Bank Ukraine (bank case study, −75 people but deliberate reinvestment), Frizzo (3-5× median), Curran/Intercom (3× R&D over 16 months), DORA Report 2025 (on which this ROI builds).

## Titre Article

The ROI of AI-assisted Software Development

## Date

2026-04-21

## URL

https://cloud.google.com/resources/content/dora-roi-of-ai-assisted-software-development

## Keywords

DORA ROI of AI-assisted software development, Google Cloud DORA report 2026.1, J-Curve of AI value realization, AI is an amplifier, code is a liability not an asset, tuition cost of transformation, learning curve verification tax pipeline adaptation, five pillars of value Productivity User Experience Cost Efficiency Developer Experience Business Growth, headcount reinvestment capacity, do not adopt headcount-reduction strategy, instability tax, ROI 39% payback period 8 months, sample calculator 500 FTE 176k salary 12.5 percent time saved, J-Curve productivity drop 15% three months, idea success rate one third, 0.5 percent revenue impact per successful feature, change failure rate 5 to 6 percent, lead time for changes deployment frequency, throughput vs instability, five systemic keys of adoption Trust Platform Data Users Guardrails, build context layer CapEx empower human in loop OpEx, experiment frequency leading indicator, optionality option premium option value, internal developer platform IDP as product, AI-accessible internal data, healthy data ecosystems, documentation quality, context engineering, trust in AI, clear and communicated AI stance, user-centric focus, working in small batches version control test automation continuous integration, scenario planning conservative realistic optimistic, value multiplier cost multiplier, brownfield 10 percent vs greenfield 35-40 percent productivity, inference cost divided by 280 Stanford AI Index, 727 percent ROI Google Cloud AI three years, 78 percent executives ROI gen AI use case, 88 percent early adopters agentic AI positive returns, MIT NANDA shadow AI economy, GenAI Divide State of AI in Business 2025, agentic era, autonomous agents end-to-end transformation, we don't measure AI by code it writes but by bottlenecks it clears, all models are wrong but useful, free headcount reinvest in innovation, software delivery as value engine, retain and train institutional knowledge, governance cost vs inference cost shift, Google Cloud delta team, Eva Dong, Andre Ellis Jr., Nathen Harvey, Vivian Hu, Ursula Lübbert-Passing, Eric Maxwell, Aaron Wanjala, dora.dev/ai/roi/calculator, CC BY-NC-SA 4.0

## Authors

Rapport conjoint **DORA team × delta team** (Google Cloud Professional Services). Auteurs principaux : **Eva Dong** (AI Value Realization Americas, ex-McKinsey 8 ans, Master Financial Engineering Michigan), **Andre Ellis Jr.** (Cloud Financial Operations Lead, Morehouse + Wharton MBA), **Nathen Harvey** (DORA team lead, co-auteur multiples DORA reports + 97 Things Every Cloud Engineer Should Know), **Vivian Hu** (10X Technology Consultant, contributrice DORA 2025 State of AI-assisted Software Development), **Ursula Lübbert-Passing PhD** (AI Value Realization EMEA, 20 ans benchmarking + value advisory, PhD effort estimation software projects), **Eric Maxwell** (lead 10X Technology consulting, ex-Chef Software, contributeur DORA), **Aaron Wanjala** (cloud developer advocate Spring Boot/Angular). Conseillers et contributeurs : **Ben Jose, Eric Lam, Matt Orr, Allison Park, Ryan J. Salva, Jerome Simms, Dave Stanke, Cedric Yao**. Design : Human After All (humanafterall.studio). Document publié sous licence **CC BY-NC-SA 4.0**, version v. 2026.1, citations retrieved February 2026.

## Ton

**Profile**: **DORA × Google Cloud delta team** report-framework, 60 pages, a **reference document for decision-makers** (CTO / CIO / CFO / VP Engineering / boards), published under a Creative Commons CC BY-NC-SA 4.0 license to allow adoption and internal adaptation. Hybrid format between a **vendor whitepaper** (methodology, calculator), a **DORA research report** (continuity with the 2025 State of AI-assisted Software Development) and a **financial toolkit** (formulas, scenarios, interactive calculator). Primary target audience: engineering and finance leaders who need to **build a defensible AI business case for a CFO or a board**. Secondary audience: transformation sponsors, CTO/CIO communities, consulting firms, Value Realization and FinOps teams.

**Style**: Institutional DORA voice, clear English, **pedagogical-prescriptive** without being condescending. High figure density but **always framed by explicit methodological caveats** — ***"all models are wrong"*** is cited as early as the calculator's introduction. **Honest register on uncertainty**: Google Cloud documents **three market perspectives** (positive / neutral / pessimistic) rather than pushing a single narrative; documents **the calculator's limitations**; acknowledges that the link between user experience and revenue is *"loose"* and was **excluded from the calculation**. No hype: this is **consulting-grade rigor** applied to a narratively saturated topic.

**Key aphorisms**:
- ***"AI is an amplifier"*** (the central thesis — the document's pivotal metaphor).
- ***"AI magnifies the strengths of high-performing organizations and the dysfunctions of struggling ones."***
- ***"Code is often seen as a liability, not an asset."*** (citation from Software Engineering at Google, Winters/Manshreck/Wright 2020).
- ***"We don't measure AI by the code it writes but by the bottlenecks it clears."*** (ethical-deontological measurement formula).
- ***"All models are wrong, but we hope this one proves useful."*** (acknowledged methodological lucidity).
- ***"The tuition cost of transformation."*** (the J-Curve metaphor that structures the financial communication).
- ***"We strongly recommend organizations do not adopt a headcount-reduction strategy."*** (explicit normative position — freed productivity should be **reinvested**, not cut).

**Elaborated metaphors**:
- ***J-Curve of AI value realization*** — a visual metaphor that **makes financially legible** a non-linear phenomenon (temporary dip before exponential growth). A metaphor especially useful for CFOs/boards who need to **validate the strategy during the dip** without panicking.
- ***Tuition cost of transformation*** — the initial investment is **tuition**, not a loss. A semantic frame that turns debt into educational investment.
- ***AI as amplifier*** — AI is not a **tool** in the classic sense, it is a **multiplier** that acts on the system's pre-existing state. Breaks the "AI = solution" narrative.
- ***Verification tax*** — the hidden cost of reviewing/verifying generated code, taxed against the time saved on generation. A key concept explaining that apparent gains are not net.
- ***Free headcount*** — the capacity freed by productivity = equivalent to additional headcount to allocate elsewhere. An HR/finance metaphor for framing **reallocation** instead of **reduction**.
- ***Optionality / option premium / option value*** — borrowed from derivatives finance: each experiment is a low-cost option that will only be "exercised" if it proves its value. **Frames experimentation as a portfolio of low-cost options**.
- ***We don't measure AI by the code it writes but by the bottlenecks it clears*** — redirects the metric away from **code volume** (a vanity metric, even toxic since *code is a liability*) toward **bottleneck removal**.

**Epistemic stance**: exemplary **methodological rigor** + **lucidity about uncertainty**. Three markers:
1. **Three perspectives** (positive / neutral / pessimistic) on market AI ROI — Google Cloud documents the pessimistic one (MIT NANDA shadow AI economy) **without downplaying it**.
2. **Repeated methodological note**: *"high-uncertainty estimate meant to spark a conversation, rather than a rigid mathematical formula"*.
3. **Recommendations excluded from the calculator** when the link is weak (e.g., developer retention, user experience to revenue) — prefers a conservative estimate over a fragile causal chain.

**Authority**: built from (a) the **DORA lineage** (continuity from 2020 ROI of DevOps Transformation → 2025 State of AI-assisted Software Development → DORA AI Capabilities Model → 2026 ROI of AI), (b) the **rigor of the framework** (simple formula ROI = (Value − Investment) / Investment, an open and adaptable calculator), (c) **convergence with broader Google Cloud data** (78% executives, 88% agentic adopters, 727% over 3 years, 8-month payback), (d) the **co-signing delta team**, Google Cloud Professional Services (operational transformation authority).

## Pense-betes

- **Date / source**: **April 21, 2026** (PDF metadata CreationDate), version **v. 2026.1**, citations retrieved February 2026. Hosting: https://cloud.google.com/resources/content/dora-roi-of-ai-assisted-software-development. Direct PDF: https://services.google.com/fh/files/misc/dora-roi-of-ai-assisted-software-development-2026.pdf
- **Format**: DORA × delta team report-framework, **60 pages**, under **CC BY-NC-SA 4.0** license, with an **interactive calculator** at https://dora.dev/ai/roi/calculator
- **Main authors**: Eva Dong, Andre Ellis Jr., Nathen Harvey (DORA team lead), Vivian Hu, Ursula Lübbert-Passing PhD, Eric Maxwell, Aaron Wanjala
- **7-chapter outline**: Executive summary → Build the business case → Understand the market divide → Calculate the ROI → Build the organizational foundation → Map your AI investment roadmap → Secure long-term ROI → (Acknowledgments / Next steps / Appendix calculator) ### The pivotal thesis — *AI is an amplifier* > *"Artificial intelligence (AI) serves as a powerful amplifier in software development. It magnifies the strengths of high-performing organizations and the dysfunctions of struggling ones."*
- **Implication 1**: buying AI licenses alone **is not enough** — *"purchasing licenses alone will not guarantee a financial return"*.
- **Implication 2**: if the organization has bottlenecks (manual testing, bureaucracy, fragmented data), AI **accelerates** technical debt instead of reducing it.
- **Implication 3**: *"Code is often seen as a liability, not an asset"* (Winters/Manshreck/Wright, *Software Engineering at Google*, 2020) — generating more code **without oversight** increases verification overhead and long-term debt.
- **Implication 4**: the relevant metric is **not code volume** but ***"the bottlenecks it clears"***. ### The J-Curve of AI value realization (central concept) > *"The introduction of a new process almost guarantees an initial negative impact on performance, with the depth of the decline directly correlating to the magnitude of the change."* **Three drivers** of the temporary productivity dip: 1. **Learning curve**: teams learn new interfaces, adapt workflows, master the shift from *prompting → systems built on context, intent, specification*. 2. **Verification tax**: time spent reviewing generated code (distrust of hallucinations + increased volume). 3. **Pipeline adaptation**: downstream processes (testing, change approval) must absorb the new velocity, **exposing legacy constraints**. **Managerial risk**: *"Initiatives often fail not because the technology is flawed but because leadership misinterprets this learning phase as a failure and pulls funding during the inevitable dip."* Hence the need to **explicitly budget** the dip to protect the investment during the learning phase. ### The sample calculator — reference figures | Variable | Sample value | |----------|---------------| | Technical staff size (FTE) | **500** | | Average fully loaded salary | **$176,000** (US blended; +30% US, +100% EU on base) | | Net time saved per developer | **12.5%** (~1h / 8h day; literature range 40-150 min/day) | | Annual AI license / user | **$250** | | Additional annual AI costs / user (API/tokens) | **$80** | | Annual training cost / user | **$9,600** | | Additional infra cost | **$100,000** | | J-Curve productivity drop | **15%** | | J-Curve duration | **3 months** | | Product portfolio revenue | **$100M** | | Cost of downtime / hour | **$100,000** | | Current deployments / year | 50 | | Idea success rate | **33%** (Larsen et al. 2023) | | Revenue impact per successful feature | **0.5%** (range 0.01-1%) | | Current CFR | 5% | | Target CFR | 6% (+20% — *instability tax*) | | Target deployments / year | 56 (+12%) | | Target features / year | 56 | | FDRT (failed deployment recovery time) | 4 hours | **Sample results**: | | Amount | |--|--| | Total hard costs (tooling + training) | **$5,065,000** | | J-Curve cost | **$3,300,000** | | **Total first-year investment** | **$8,365,000** | | Headcount reinvestment capacity | $11,000,000 | | Revenue from extra feature deployments | $990,000 | | Downtime impact (instability tax) | **−$344,000** | | **Total annual value** | **$11,646,000** | | **First-year benefit** | $3,281,000 | | **First-year ROI** | **39%** | | **Payback period** | **0.7 year (8 months)** | ### Five pillars of value (cumulated business value) ``` Productivity → User Experience → Cost Efficiency → Developer Experience → Business Growth (most direct)                                                  (most indirect) ```
- **Productivity**: the most direct effect, best confirmed by DORA 2025 (>80% of respondents perceive a gain).
- **Developer Experience**: retention, less turnover (replacement cost = 1.5-2× annual salary). **Excluded from the base calculator** (variable link).
- **Cost efficiency**: avoided hire (not headcount reduction!) + IT infra savings.
- **User Experience**: app performance → engagement. **Excluded from the calculator** ("loose" link).
- **Business Growth**: revenue, conversion. The most downstream, the hardest to attribute. ### Five systemic keys of adoption (organizational foundation) 1. **Trust**: *"clear and communicated AI stance"* — reduces verification tax via psychological safety. 2. **Platform**: Internal Developer Platform (IDP) treated **as a product** — guardrails for devs AND for agents. 3. **Data**: AI-accessible internal data + healthy data ecosystems + machine-readable documentation quality. 4. **Users**: user-centric focus — velocity directed toward user value, not commit volume. 5. **Guardrails**: non-optional security/quality gates, automated checks, pre-commit hooks. ### Two-phase investment roadmap | Phase | Budget type | Capabilities | Goal | |-------|-------------|--------------|------| | (1) **Build the context layer** | **CapEx** | Quality IDP + healthy data ecosystem + machine-readable docs | Minimize agent friction — *garbage in, garbage out* | | (2) **Empower the human in the loop** | **OpEx** | Trust in AI + context engineering | Devs become *high-level orchestrators* — reduce verification tax | | (3) **Validate progress** | (gauge) | Leading: experiment frequency + deployment frequency / Stability: change failure rate + rework | Confirm J-Curve trajectory | ### Three scenarios to model | Scenario | Value multiplier | Cost multiplier | Assumption | |----------|------------------|-----------------|-----------| | **Conservative** | **0.8** | **1.5** | Slow adoption + hidden integration overhead | | **Realistic base** | 1.0 | 1.0 | Standard trajectory | | **Optimistic** | **1.2** | **0.8** | Elite team + mature IDP absorbing the tools | ### External data mobilized (evidence) | Data point | Value | Source | |--------|--------|--------| | Executives reporting ROI on ≥1 gen AI use case | **78%** | Google Cloud, *The ROI of AI 2025* | | Early adopters seeing positive returns from agentic AI | **88%** | Google Cloud, *The ROI of AI 2025* | | Productivity gain greenfield (simple) | **35-40%** | Stanford Software Engineering Productivity Research | | Productivity gain brownfield (legacy) | **≤10%** | Stanford | | Inference cost reduction (Nov 2022 → Oct 2024) | **÷280** | Stanford 2025 AI Index | | Avg payback period for AI tools (Google Cloud data) | **8 months** | Google Cloud, *How Businesses Achieve Strong ROI* | | Avg ROI for Google Cloud AI customers (3 years) | **727%** | Google Cloud | | Idea success rate (features that increase revenue) | **~33%** | Larsen et al. 2023 (A/B testing methodology) | | Developer replacement cost | **1.5-2× annual salary** | Standard HR | | Fully loaded salary cost overhead | **+30% US / +100% EU** | On base salary | ### Explicit normative position — *do not reduce headcount* > *"We strongly recommend organizations do not adopt a headcount-reduction strategy, which has a negative impact on morale and organizational culture, can reduce efficiencies, and can even incentivize workers to not improve their work processes. Instead, this effort should be reinvested into new, innovative, or more productive work."* Position **explicitly opposed** to cases like **Tatsyi/Raiffeisen Bank Ukraine** (−75 people over 12 months). Productive tension: Tatsyi reports a **deliberate reallocation** of the freed capacity while still reducing headcount; DORA recommends **retaining headcount** and reinvesting the freed capacity into innovation. The two positions are not irreconcilable — Tatsyi is retrospective on a decision already made, DORA is prescriptive **a priori** to preserve morale, institutional knowledge, and incentive structure. To be used as a **discussion pivot** in executive committees. ### Connection to the corpus #### Convergence on "AI is an amplifier" / "organizational system >> tool"
- **Tatsyi/Raiffeisen** (2026-05-05): *"AI expanded our production possibility frontier, and we deliberately allocated the freed capacity"* — an exact analogue of the DORA position.
- **Wescale Usine Logicielle Augmentée** (2026-05-03): *governance injected as a "near-military layer"*, realistic X3-X4.
- **Habert PROJ-AI** (2026-05-05): *"technology 20% / team discipline 80%"* — a direct reformulation.
- **Osmani *Cognitive Surrender*** (2026-05-05): 6 structural organizational guardrails.
- **MIT NANDA *GenAI Divide*** (2025-08-23): 95% of AI pilots fail to deliver ROI — explicitly cited by DORA as a *"pessimistic perspective"*.
- → **Strong convergence**: organizational maturity is the **moat**, not the tool. #### Convergence on "J-Curve / tuition cost"
- **Frizzo** (2026-05-05): *"writing muscle atrophy"*, *"the new bottleneck is supervision"* — lives the verification tax day to day.
- **BCG Brain Fry** (Bedard et al., 2026-03-05): 14% AI brain fry — the learning cost on the human side.
- **Beck *Starving Genies*** (2026-04-03): voluntary scarcity to preserve manual practice vs FOMO over 24/7 agents.
- → DORA provides the **financial framework** for the dip these authors document qualitatively. #### Convergence on productivity ratios (committed median 3-5×)
- **DORA sample 12.5% time saved** = equivalent to a 1.14× on an 8h base, **far more conservative** than the 3-5× median ratios in the 2026 corpus.
- **Why**: DORA is **conservative by design** to defend the case before a CFO. Practitioner ratios (Frizzo 3-5×, Wescale X3-X4, Curran 3×, Tatsyi multi-tool ×1.5-3 / Claude stack ×4.5) capture **the entire transformation of the job** (scope change, new products, task reallocation) that the DORA calculator does not capture — it measures *avoided hire*, not *new product space*.
- **Stanford 35-40% greenfield vs ≤10% brownfield** cited by DORA: confirms the **uneven distribution** by technical context.
- → **Correct reading**: DORA = **financially defensible floor**; practitioner ratios = **organizationally observed ceiling**. Both are true depending on measurement scope. #### Convergence on "free headcount / reinvest / do not reduce headcount"
- **DORA**: *"do not adopt a headcount-reduction strategy"*.
- **Curran/Intercom** (2026-04-16): 3× R&D productivity **without massive reduction** — internal reallocation.
- **Tatsyi/Raiffeisen** (2026-05-05): −75 people but with **deliberate reallocation** of freed capacity toward features / stability / technical debt.
- → **Productive tension**: DORA prescriptive (do not reduce) vs Tatsyi descriptive (reduced while reallocating). To be used for **balanced** presentations. #### Convergence on "code is a liability"
- **DORA** cites Software Engineering at Google (Winters/Manshreck/Wright, 2020).
- **Cherny** (2026-05): *"100% of generated code"* — but with oversight and compaction.
- **Frizzo** (2026-05-05): *"writing muscle atrophy"* — the cognitive cost of volume.
- **Osmani Cognitive Surrender** (2026-05-05): *"PRs ~100 lines max"* — an explicit anti-volume limit.
- → ***"more code is bad code"*** is a **2026 stylized fact** confirmed by heterogeneous sources. #### France/Europe vs Anglo-Saxon position
- DORA is **American but documents a European position** (Ursula Lübbert-Passing PhD, EMEA Value Realization).
- **Calculator includes** a +100% salary overhead for Europe (vs +30% US) — sensitive to local contexts.
- To be used in FR presentations as a **reference standard** for CFO/boards, complementary to Wescale (FR firm), Habert (FR), Tatsyi (Central Europe). #### Convergence on "experiment frequency = leading financial indicator"
- **DORA**: optionality framework, experiment frequency as a leading indicator.
- **Cherny** (2026-05): *"few dozen PRs/day"*, parallel exploration.
- **Karpathy** (2026-04-29): MenuGen vs Nanobanana, *jagged intelligence*, experimentation as a new mode.
- **Habert PROJ-AI**: *7-dimension Decision Records* legitimizing documented exploration.
- → **Convergence**: AI makes it possible to **turn every feature into a low-cost option**. ### Limitations to flag
- **Calculator deliberately simplistic** (acknowledged by the authors) — **excludes**: retention/turnover savings, user experience to revenue, agentic AI compounding effects year 2+, downstream business process savings (HR, etc.).
- **Sample figures very US-centric** ($176k salary, $250 license/year — likely underestimates enterprise costs with large-account negotiation + agents).
- **12.5% time saved is very conservative** compared to practitioner feedback — but this is **by design** to defend the case before a skeptical CFO. Inverse risk: **undervaluing** the potential.
- **Time saving capped by the instability tax** in the model — the temporary dip (15% drop over 3 months) **weighs $3.3M** out of $8.4M total investment. To verify: is this dip really always this large? For mature organizations, it may be much smaller.
- **Idea success rate of 33% (Larsen 2023)** comes from standard A/B testing — may be too **low** for well-researched features and too **high** for experimental ones.
- **No explicit discussion** of regulatory risks (GDPR, EU AI Act, sector-specific oversight) — **surprising** for a Google Cloud document with EMEA enterprise reach.
- **The Google Cloud link** (727% ROI over 3 years, 8-month payback): statistics **internal to Google Cloud customers**, likely selection bias.
- **The calculator does not capture** what Tatsyi calls the *"production possibility frontier"* (the **new products** that didn't exist before) — it measures *avoided hire*, not *new product space*. A **structural** limitation of the model.
- **"All models are wrong"** is repeated 3 times in the document — a self-disarming move, but it **does not exempt** the model from methodological critique on blind application. ### To be used for
- **Executive committee / board / CFO presentations**: official Google Cloud × DORA framework — institutional authority to defend an AI budget.
- **Building an AI business case**: reuse the calculator as a structure, **adjust the assumptions to the client context** (explicit recommendation from the authors).
- **Raising awareness of the "tuition cost"**: protect the investment during the J-Curve dip, do not cut funding during the learning phase.
- **Team / HR debate**: the normative position *"do not reduce headcount"* is a usable argument against purely budget-driven headcount reduction impulses.
- **Sourced figures**: 39% ROI / 8-month payback / 35-40% greenfield vs 10% brownfield / inference cost ÷280 / 727% ROI over 3 years — **solid figures** to integrate into training, tech-watch notes, and strategic presentations.
- **Connecting FR content with Wescale / Habert / Tatsyi / Frizzo**: DORA as the **institutional financial foundation**, other notes as complementary **operational testimonials**.
- **Strategic discussion on IDP / context engineering**: the DORA capabilities (Quality IDP + AI-accessible internal data + Documentation quality) become identifiable and budgetable **priority investments**.

## RésuméDe400mots

The **Google Cloud DORA × delta team** publishes in April 2026 (v. 2026.1, citations February 2026, **CC BY-NC-SA 4.0**) a 60-page report-framework dedicated to the ROI of AI in software development, with an **interactive calculator** at dora.dev/ai/roi/calculator. The document sits within the DORA lineage (2020 ROI of DevOps Transformation → 2025 State of AI-assisted Software Development → DORA AI Capabilities Model → 2026 ROI of AI).

**Pivotal thesis**: ***"AI is an amplifier"*** — AI simultaneously magnifies the strengths of high-performing organizations and the dysfunctions of struggling organizations. Buying AI licenses **is not enough**: AI injected into a system with manual testing, bureaucracy, or fragmented data **accelerates** technical debt. Citation from Software Engineering at Google: ***"code is often seen as a liability, not an asset"***. Ethical metric: ***"we don't measure AI by the code it writes but by the bottlenecks it clears"***.

**New central concept**: the ***J-Curve of AI value realization*** — every AI adoption goes through a **temporary dip** (learning curve + verification tax + pipeline adaptation) before **exponential growth**, a metaphor for the *"tuition cost of transformation"* to be **explicitly budgeted** so as not to panic during the dip.

**Sample calculator** (500 FTE / $176k salary / 12.5% time saved per developer): **value $11.6M / investment $8.4M / ROI 39% / payback 8 months (0.7 year)**. Detail: hard costs $5.065M + J-Curve cost $3.3M; value = headcount reinvestment $11M + extra features $990k − instability tax $344k.

**Explicit normative position**: ***"we strongly recommend organizations do not adopt a headcount-reduction strategy"*** — reinvest, retain talent, capitalize on institutional knowledge.

**Five pillars of value**: Productivity / User Experience / Cost Efficiency / Developer Experience / Business Growth (from most direct to most indirect). **Five systemic keys**: Trust + Platform + Data + Users + Guardrails. **Two-phase roadmap**: (1) Build context layer (CapEx) — quality IDP + healthy data ecosystems; (2) Empower human in loop (OpEx) — context engineering + trust in AI. Leading indicators: experiment frequency + deployment frequency.

**External data**: 78% of executives report ROI on ≥1 gen AI use case, 88% of early agentic AI adopters see positive ROI, **35-40% greenfield productivity vs ≤10% brownfield** (Stanford), inference cost **÷280** (Nov 2022 → Oct 2024), **727% ROI over 3 years** for Google Cloud AI customers, average payback of **8 months**.

**Connection to the corpus**: strong convergence with Tatsyi/Raiffeisen (production possibility frontier), Wescale (governance + X3-X4), Habert PROJ-AI (technology 20% / discipline 80%), MIT NANDA (95% of pilots fail, explicitly cited). Productive tension with practitioner ratios (Frizzo 3-5×, Curran 3×, Tatsyi ×1.5-4.5): DORA = **financially defensible floor** (12.5% time saved), practitioners = **organizationally observed ceiling**. To be used for executive committees, CFO business cases, transformation sponsors.

## GrapheDeConnaissance

- DORA —publie→ The ROI of AI-assisted Software Development (DOCUMENT, 0.98)
- Google Cloud delta team —publie→ The ROI of AI-assisted Software Development (DOCUMENT, 0.97)
- Eva Dong —travaille_chez→ Google Cloud (ORGANISATION, 0.95)
- Nathen Harvey —dirige→ DORA (ORGANISATION, 0.96)
- Ursula Lübbert-Passing —travaille_chez→ Google Cloud (ORGANISATION, 0.95)
- AI —est_instance_de→ amplificateur du système organisationnel (CONCEPT, 0.97)
- DORA —affirme_que→ l'IA magnifie les forces des organisations performantes et les dysfonctionnements des organisations en difficulté (AFFIRMATION, 0.97)
- J-Curve of AI value realization —est_instance_de→ trajectoire à creux de productivité initial puis croissance exponentielle (CONCEPT, 0.96)
- Learning curve —fait_partie_de→ J-Curve of AI value realization (CONCEPT, 0.95)
- Verification tax —fait_partie_de→ J-Curve of AI value realization (CONCEPT, 0.95)
- Pipeline adaptation —fait_partie_de→ J-Curve of AI value realization (CONCEPT, 0.95)
- Tuition cost of transformation —est_variante_de→ J-Curve productivity drop (CONCEPT, 0.94)
- Code —est_instance_de→ liability not asset (CONCEPT, 0.94)
- DORA —recommande→ mesurer l'IA par les bottlenecks levés et non par le volume de code généré (AFFIRMATION, 0.96)
- Sample ROI calculator —mesure→ 39% ROI / 8 mois payback / 11.6M$ valeur / 8.4M$ investissement (MESURE, 0.97)
- Sample ROI calculator —est_basé_sur→ hypothèses 500 FTE / 176k$ salary / 12.5% time saved / 15% J-Curve drop 3 mois (CONCEPT, 0.96)
- DORA —recommande→ ne pas adopter de stratégie de réduction d'effectifs (AFFIRMATION, 0.97)
- DORA —recommande→ réinvestir la productivité libérée par l'IA dans l'innovation et la création de valeur (AFFIRMATION, 0.96)
- Productivity, User Experience, Cost Efficiency, Developer Experience, Business Growth —fait_partie_de→ Cinq piliers de valeur (DORA) (CONCEPT, 0.96)
- Trust, Platform, Data, Users, Guardrails —fait_partie_de→ Cinq clés systémiques d'adoption (CONCEPT, 0.96)
- DORA —recommande→ traiter l'Internal Developer Platform comme un produit (AFFIRMATION, 0.95)
- Build context layer —est_instance_de→ phase CapEx (Quality IDP + Healthy data ecosystem) (CONCEPT, 0.94)
- Empower human in loop —est_instance_de→ phase OpEx (Trust in AI + context engineering) (CONCEPT, 0.94)
- Experiment frequency —est_instance_de→ leading financial indicator (CONCEPT, 0.95)
- Stanford AI Index 2025 —mesure→ inference cost réduit ×280 entre nov 2022 et oct 2024 (MESURE, 0.96)
- Stanford research —mesure→ 35-40% productivité greenfield vs ≤10% brownfield (MESURE, 0.94)
- Google Cloud —mesure→ 727% ROI moyen sur 3 ans des clients Google Cloud AI (MESURE, 0.92)
- MIT NANDA —affirme_que→ une shadow AI economy existe et 95% des pilotes IA échouent à délivrer un ROI (AFFIRMATION, 0.94)
- Larsen et al. 2023 —mesure→ idea success rate ~33% (MESURE, 0.92)
- Verification tax —est_basé_sur→ Code (CONCEPT, 0.95)
- bilan ROI de l'étude —converge_avec→ Tatsyi production possibility frontier, Wescale gouvernance, Habert technology 20 / discipline 80 (CONCEPT, 0.93)
- Ratios praticiens 3-5× (Frizzo, Wescale, Curran, Tatsyi) —surpasse→ Sample 12.5% time saved (CONCEPT, 0.93)
- Position no-headcount-reduction (DORA) —s_oppose_à→ Tatsyi −75 personnes mais réallocation délibérée (CONCEPT, 0.91)
- Optionality framework —permet→ de traiter chaque expérience comme option à faible coût (CONCEPT, 0.94)
- DORA —affirme_que→ l'IA accélère l'accumulation de dette technique si l'organisation est en bottleneck (AFFIRMATION, 0.95)
- Scénarios Conservative 0.8/1.5, Realistic 1.0, Optimistic 1.2/0.8 —fait_partie_de→ Trois scénarios (DORA) (METHODOLOGIE, 0.94)
- Google Cloud —mesure→ 78% des executives rapportent un ROI sur au moins 1 use case gen AI (MESURE, 0.93)
- Google Cloud —mesure→ 88% des early adopters agentic AI voient des retours positifs (MESURE, 0.92)

---
Canonical: https://www.thekb.eu/en/fiches/dora-google-cloud-roi-ai-assisted-software-development-j-curve-2026-04-21/