Latest data reveals a troubling gap between AI adoption and actual productivity gains, plus what enterprise leaders need to know.
The software development landscape is experiencing its most significant transformation since the advent of cloud computing. Our comprehensive analysis of Stack Overflow’s 2025 Developer Survey, GitHub’s Octoverse report, and groundbreaking METR research studies reveals a striking paradox: while AI adoption among developers continues to surge, the actual productivity benefits are far from the promised gains.
For manufacturing and supply chain leaders who increasingly rely on custom software solutions, from IIoT implementations to supply chain optimization platforms, understanding this reality is critical for making informed technology investment decisions.
The following data represents the current state of AI in software development based on responses from over 49,000 developers worldwide and rigorous controlled studies:
The AI Adoption Statistics — 2026
| Key Metric | 2024 | 2025 | Change | Impact |
| Overall Adoption | 76% | 84% | +8% | Near-universal adoption |
| Daily Usage | 45% | 51% | +6% | Professional mainstream |
| Trust in Accuracy | 40% | 29% | -11% | Growing skepticism |
| Actual Productivity | Assumed +24% | -19% | -43% gap | Reality vs expectation |
| Code Acceptance Rate | Unknown | <44% | N/A | Quality concerns |
Source: Stack Overflow Developer Survey 2025, METR Research Study
Three Critical Discoveries:
Despite quality concerns, AI tools have achieved unprecedented adoption rates across the global developer community. The data shows clear momentum that enterprise leaders cannot ignore:
AI Tool Adoption by Developer Experience — 2026
| Experience Level | Daily Usage | Weekly Usage | Monthly Usage | Never Use | Total AI Usage |
| Early Career (0-4 years) | 56% | 18% | 12% | 12% | 88% |
| Mid-Career (5-9 years) | 53% | 17% | 13% | 13% | 87% |
| Experienced (10+ years) | 47% | 17% | 13% | 17% | 83% |
| Overall Professional Average | 51% | 17% | 13% | 14% | 86% |
Source: Stack Overflow Developer Survey 2025
Key Insights:
GitHub’s Octoverse data reveals explosive global growth in AI-capable development talent. Based on data from GitHub’s platform (separate from Stack Overflow’s survey data), we see significant developer population expansion:
Developer Population Growth by Region — 2024
| Region | Developer Growth | # of Developers | Strategic Implication |
| India | 28% YoY | >17M | Largest developer population by 2028 |
| Philippines | 29% YoY | >1.7M | Fastest growing in Asia Pacific |
| Brazil | 27% YoY | >5.4M | Leading Latin American market |
| Nigeria | 28% YoY | >1.1M | African tech hub development |
| Indonesia | 23% YoY | >3.5M | Emerging Southeast Asia leader |
| Japan | 23% YoY | >3.5M | Advanced tech infrastructure |
| Germany | 21% YoY | >3.5M | European manufacturing center |
| Mexico | 21% YoY | >1.9M | Growing North American hub |
| United States | 12% YoY | Largest (>20M) | Mature market stabilization |
| Kenya | 33% YoY | >393K | Highest growth rate globally |
Source: GitHub Octoverse 2024
Note: This data reflects developer activity on GitHub’s platform and represents different methodology than the Stack Overflow survey responses. GitHub tracks actual platform usage while Stack Overflow surveys developer sentiment and practices.
For enterprise leaders, this global expansion means access to a larger pool of AI-capable developers, but also increased competition for top talent in key technology hubs.
The data reveals a clear pattern of where developers embrace AI versus where they resist its implementation:
AI Usage Patterns by Development Task — 2026
| Task Category | Currently Using AI | Willing to Try | Won’t Use AI | Enterprise Risk Level |
| Search for answers | 54% | 23% | 23% | Low – Learning/research |
| Generate content/data | 36% | 28% | 36% | Low – Documentation |
| Learn new concepts | 33% | 31% | 36% | Low – Training support |
| Document code | 31% | 25% | 44% | Low – Maintenance tasks |
| Write code | 17% | 24% | 59% | Medium – Implementation |
| Test code | 12% | 32% | 44% | High – Quality assurance |
| Code review | 9% | 30% | 59% | High – Critical oversight |
| Project planning | 8% | 23% | 69% | High – Strategic decisions |
| Deployment/monitoring | 6% | 19% | 76% | Critical – System reliability |
Source: Stack Overflow Developer Survey 2025
Strategic Implications for Manufacturing:
Despite widespread adoption, developer trust in AI accuracy has hit concerning lows, creating a fundamental tension in the market:
Developer Trust in AI Accuracy — 2026
| Trust Level | Percentage | Year-over-Year Change | Experience Level Most Affected |
| Highly trust | 3% | -2% | Early career (4%) |
| Somewhat trust | 30% | -8% | Mid-career (29%) |
| Somewhat distrust | 26% | +3% | Experienced (31%) |
| Highly distrust | 20% | +5% | Experienced (25%) |
| Net Trust | 32.7% | -12% | All levels |
| Net Distrust | 46% | +8% | All levels increasing |
Source: Stack Overflow Developer Survey 2025
Critical Finding: More developers actively distrust AI accuracy (46%) than trust it (33%), with only 3% reporting high trust in AI-generated output.
The most significant quality issues driving this trust erosion directly impact enterprise software development:
Top Developer Frustrations with AI Tools — 2026
| Issue | Percentage Affected | Impact on Development Time | Enterprise Impact |
| “Almost-right” solutions | 66% | +15-25% debugging | High – Subtle errors in critical systems |
| Increased debugging time | 45% | +19% overall slowdown | High – Hidden technical debt |
| Reduced developer confidence | 20% | Unmeasured quality impact | Medium – Team capability concerns |
| Code comprehension issues | 16% | +10% review time | High – Maintainability problems |
| No significant problems | 4% | Baseline performance | Low – Rare positive experience |
Source: Stack Overflow Developer Survey 2025
The Bottom Line: Two-thirds of developers report that AI generates solutions that are “almost right, but not quite,” leading to increased debugging time and reduced confidence in AI-generated code.
The most groundbreaking finding comes from METR’s rigorous randomized controlled trial, which studied 16 experienced developers across 246 real-world tasks. This research represents the first scientifically rigorous measurement of AI’s actual impact on developer productivity.
METR Productivity Study Results — 2026
| Metric | Developer Expectation | Actual Measured Result | Perception Gap | Study Conditions |
| Task Completion Time | -24% (faster) | +19% (slower) | 43% gap | Real-world codebases |
| Code Quality | Assumed equivalent | <44% accepted unchanged | Significant quality gap | 22,000+ GitHub stars avg |
| Review Time Required | Minimally increased | +9% of total task time | Major overhead | 1M+ lines of code |
| Developer Confidence | Maintained high | Remained overconfident | Persistent delusion | Post-task surveys |
Source: METR Early-2025 AI Study on Open-Source Developer Productivity
The study revealed precisely where AI productivity claims break down:
Where Development Time Goes with AI Tools — 2026
| Time Category | Without AI | With AI Tools | Change | Manufacturing Impact |
| Active coding | 65% | 52% | -13% | Less hands-on implementation |
| Planning & design | 15% | 12% | -3% | Reduced strategic thinking |
| Reviewing AI output | 0% | 9% | +9% | New overhead category |
| Debugging & fixes | 12% | 18% | +6% | Increased maintenance burden |
| Idle/waiting time | 3% | 6% | +3% | Tool responsiveness delays |
| Documentation | 5% | 3% | -2% | AI assists with docs |
Source: METR Research Analysis
Critical Finding: The 9% of time spent reviewing AI outputs often exceeded the time supposedly saved by AI generation, creating a net productivity loss rather than gain.
The most commonly used programming languages reflect the breadth of modern software development, from web applications to enterprise systems:
Top Programming Language by Usage — 2026
| Language | Primary Use Case | Adoption Rate | AI Development Impact | Enterprise Relevance |
| Python | AI/ML, Data Science, Backend | 58% | High – Primary AI development language | High – Analytics, automation, IIoT |
| JavaScript | Web Development, Full-stack | 66% | Medium – Enhanced tooling | High – User interfaces, APIs |
| Java | Enterprise Applications, Android | High adoption | Medium – Legacy system modernization | Critical – Enterprise backends |
| TypeScript | Large-scale Web Applications | Growing rapidly | Medium – Type-safe development | High – Scalable frontend systems |
| C# (.NET) | Enterprise Software, Games | High adoption | Medium – Microsoft ecosystem | Critical – Windows applications, cloud |
Source: Stack Overflow Developer Survey 2025, GitHub Octoverse 2024
Key Trends:
Based on the trust data and productivity research, manufacturing leaders need comprehensive governance frameworks. Here’s what the data suggests:
AI Governance Requirements by Risk Level — 2026
| Risk Category | AI Usage Restriction | Required Safeguards | Measurement KPIs | Manufacturing Examples |
| Critical Systems | Prohibited or heavily restricted | Manual approval + senior review | 100% human verification | PLCs, safety systems, real-time control |
| High-Stakes Code | Mandatory review + testing | Enhanced QA + security scan | <5% defect rate | ERP integrations, financial systems |
| Quality-Sensitive | Guided usage + oversight | Automated testing + lint | Standard quality metrics | Data pipelines, reporting systems |
| Development Support | Encouraged with training | Best practices + style guide | Developer satisfaction | Documentation, prototypes, learning |
Code Review Enhancement Requirements:
| Current Review Process | AI-Enhanced Requirements | Additional Time Investment | Quality Improvement |
| Standard peer review | +Technical lead approval | +25% review time | Moderate improvement |
| Senior developer sign-off | +Security/quality scan | +15% review time | Significant improvement |
| Automated testing | +AI-specific test cases | +10% test development | High confidence gain |
| Documentation standards | +AI decision explanations | +20% documentation time | Long-term maintainability |
Based on the comprehensive data analysis, here are specific recommendations for manufacturing leaders:
ROI-Driven AI Implementation Strategy — 2026
| Implementation Phase | Investment Focus | Expected Timeline | Measured Success Criteria | Risk Mitigation |
| Phase 1: Foundation | Training + governance | 3-6 months | Policy compliance >95% | Enhanced review processes |
| Phase 2: Limited Deployment | Documentation + learning | 6-12 months | Developer satisfaction +20% | Low-risk use cases only |
| Phase 3: Selective Expansion | Guided implementation | 12-18 months | Productivity neutral/positive | Objective measurement |
| Phase 4: Optimization | Advanced tooling | 18+ months | Clear ROI demonstration | Continuous monitoring |
The trust and productivity data suggest a fundamental reallocation of AI budgets away from pure tooling toward the processes needed to manage AI effectively.
Enterprise AI Development Budget Distribution — 2026 Recommendations
| Category | Recommended % of AI Budget | Justification | Expected ROI Timeline |
| Training & Change Management | 35% | Address trust/adoption gap | 6-12 months |
| Enhanced Review Processes | 25% | Mitigate quality risks | 3-6 months |
| Measurement & Analytics | 20% | Track actual vs perceived benefits | 6-18 months |
| Tool Licensing & Infrastructure | 15% | Support expanded usage | 3-6 months |
| Risk Management & Governance | 5% | Prevent costly errors | Ongoing protection |
This allocation reflects the reality that the largest costs and risks in AI adoption are not the tools themselves, but the organizational changes required to use them effectively.
AI Development Tool Evolution — 2025-2026 Projections
| Technology Category | Current State | 2026 Prediction | Manufacturing Impact |
| Local/Private AI Models | 15% adoption | 45% adoption | High – Data security compliance |
| Specialized Industry Models | Rare | 25% availability | High – Manufacturing-specific knowledge |
| Enhanced Code Review AI | Basic | Advanced quality detection | Medium – Improved catching of errors |
| Infrastructure Automation | Limited | Widespread deployment | High – IIoT system management |
The 2025 data reveals a development landscape where AI adoption is widespread but benefits remain unevenly distributed. For manufacturing and supply chain leaders, the key strategic insights are:
Immediate Actions (Next 90 Days):
Medium-Term Strategy (6-18 Months):
Long-Term Vision (18+ Months):
The opportunity lies not in wholesale AI adoption, but in strategic implementation that leverages AI’s strengths while mitigating its documented weaknesses through proper governance, measurement, and human oversight.
Ready to navigate AI integration in your software development process?
USM Business Systems specializes in helping manufacturing and supply chain leaders implement AI governance frameworks that drive real business value. Our Agentic AI for SDLC services provide expert guidance on balancing innovation with operational excellence.
[Schedule your AI readiness assessment →]
Stack Overflow. (2025). 2025 Stack Overflow Developer Survey. Retrieved from https://survey.stackoverflow.co/2025/
[2] GitHub. (2024). The State of the Octoverse 2024: AI leads Python to top language as the number of global developers surges. Retrieved from https://github.blog/news-insights/octoverse/octoverse-2024/
[3] Becker, J., Rush, N., Barnes, E., & Rein, D. (2025). Measuring the Impact of Early-2025 AI on Experienced Open-Source Developer Productivity. METR. Retrieved from https://metr.org/blog/2025-07-10-early-2025-ai-experienced-os-dev-study/
Chemical products such as medicines, plastics, soap, and paint are still often based on fossil raw materials. This is not sustainable, so there is an urgent need for ways to make a ‘materials transition’ to products made from bio-based raw materials. To achieve results more quickly and efficiently, researchers at Radboud University in the Big Chemistry programme are using robots and AI.
The material transition from fossil-based to bio-based (where raw materials are based on materials of biological origin) is a major challenge. Raw materials for products must be replaced without changing the quality of those products. This requires knowledge of the properties and behaviour of those raw materials at the molecular level. Wilhelm Huck, professor of physical-organic chemistry at Radboud University: “Moreover, you don’t want to optimize the properties of a single molecule, but of a mixture. And we can greatly accelerate that search with our robots and models.”
The difficulty, Huck explains, is that most chemistry is ‘non-additive’. ‘Whether you dissolve one sugar cube in water or ten, essentially the same thing happens with ten cubes. That is predictable. But if you know how one molecule behaves and you know how another molecule behaves, you might think: if I put them together, I’ll get the combined or the average of the two. And that’s almost never the case in chemistry. In many cases, the interaction between molecules leads to an interaction that you couldn’t have predicted.”
Because raw materials can interact with other raw materials in all kinds of ways, the number of possible interactions increases rapidly. Huck: ‘And when you consider that suppliers of ingredients for cleaning products, cosmetics, paints and coatings, ink, perfume, medicines, you name it – that they can supply tens of thousands of components. And that you can combine them in different ways. That quickly adds up to hundreds of millions of interactions that you can’t possibly study all. So you need a model that can predict the properties of mixtures. And to train that model, you need a lot of data, which you collect in experiments.”
This fall, three grants were awarded to projects by Radboud researchers within the larger Big Chemistry program of the National Growth Fund. Led by chemists Wilhelm Huck, Mathijs Mabesoone, and Peter Korevaar, the programme involves collaboration with companies to conduct research into the properties of bio-based raw materials for paints and soaps, among other things.
Peter Korevaar will be conducting research into paints together with Van Wijhe Verf. These are often still (partly) based on oil, because they have to be waterproof. And that is just one of the requirements that paint has to meet: ‘Paint has to mix well. That mixture has to remain stable. It must not be too watery or too viscous. It has to be washable, but it shouldn’t wash off your house when it rains. It simply has to be good stuff. If you try to design that based on new, bio-based ingredients, you need a lot of experimental data.”
Mathijs Mabesoone will be conducting research into soaps together with the company Croda International. ‘If you have a pure soap solution, it has a certain cleaning capacity, for example. But in mixtures of soaps, that same property can suddenly occur at a hundred times lower concentration. That is also very difficult to predict, so we are going to take a lot of measurements. We will create a large database of informative measurement points, which we can then use to train a model to better predict the interactions.”
The third project that received funding this fall deals with polymers on a more fundamental level: large molecules that often occur in mixtures. Huck: “For most polymers, there is insufficient data for theoretical calculations. For the development of new, bio-based polymers, we will collect more data in collaboration with TNO and Van Loon Chemical Innovations (VLCI), so that we can train AI models to make better predictions.”
Generating unique data, and lots of it, is the goal of all three projects. And the scientists are doing this with the help of robots. A large robot lab at Noviotech Campus in Nijmegen will follow in the fall of 2026. But the researchers are already working with robots the size of a small refrigerator that continuously take measurements. Mabesoone: “You supply such a robot with a few samples of basic solutions, and then you put it to work testing, mixing, and measuring. The robot decides which are the best samples to make, and you only need to supply a small amount to obtain a lot of data.”
Will consumers notice anything from this research, and if so, when? Huck: ‘If we don’t do this, you may find that at some point you can no longer get certain products because they contain substances that are no longer permitted or available. But if we do it right, you won’t notice much. You had good stuff and you want to keep good stuff. Only, in the long run, those good products will be more often biodegradable. And we can probably make the good products even better—with robotics and AI, we can try out so many more combinations than we ever thought possible that we are sure to discover completely new properties.”