Feature
Despite promises that AI will help boost productivity, AI tools are actually slowing down experienced software developers, according to a Cornell University study published last month.
Conducted between February 2025 and June 2025, researchers examined 246 coding tasks performed by 16 experienced open-source contributors, each with an average of five years working on the projects in question. Tasks were randomly assigned to allow or disallow the use of AI tools such as Cursor Pro and Claude 3.5/3.7 Sonnet.
Prior to the study, researchers forecasted a 24% time reduction — a number that dropped to 20% after experiencing a slowdown. Economics and machine learning experts anticipated time savings of 39% and 38% respectively.
Instead, the results found that developers completed tasks 19% slower when using AI compared to working unaided.
Why Does AI Slow Developers Down?
Researchers explored 20 potential factors — including project size, code quality requirements and developers’ prior AI experience — that might explain the slowdown. While they noted experimental artifacts could not be completely ruled out, consistent effects across analyses makes it unlikely that the finding is simply the result of the trial design.
Another recent study by researchers at MIT’s Computer Science and Artificial Intelligence Laboratory (CSAIL) uncovered similar challenges. Issues they uncovered include:
- AI coding systems do not invoke tools. Researchers claim AI needs to be able to select which tool to use, decide how to use it and interpret the outputs.
- Human-AI collaboration is not seamless. Human specifications are often too vague, there is little controllability over coding with LLMs and human-AI collaboration interfaces are still limited.
- Large projects require long-term planning. LLMs struggle with designing good, lasting abstractions and respecting modularity and code quality principles.
- Code writing requires semantic understanding of the codebase. LLMs struggle with seeing how various parts of code go together, knowing what is implemented where, understanding how algorithms work and keeping track of program invariants at certain program points
Security is another major factor. AI-generated code may overlook important security rules, leading to time-consuming fixes. Compatibility issues can also arise when AI-produced code does not work well with a company’s existing libraries and tools, causing further delays.
Related Article: Vibe Coding Explained: Use Cases, Risks and Developer Guidance
Where AI Tools Struggle in Real-World Coding
Vivek Mishra, a senior member of the Institute of Electrical and Electronics Engineers (IEEE), said AI coding tools are helpful and can quickly suggest solutions. But in real-world industry coding, they can’t be relied on completely.
“While they assist in writing code, they might not understand a company’s specific classes, bugs, coding standards or business needs,” he said. “Over-relying on AI tools can sometimes slow things down due to issues with library compatibility or security concerns.”
Protecting personal data is also a big worry, he added, especially in fintech companies, where sending sensitive information to AI tools is risky. “To get code that meets industry standards, security and compatibility, you often need to ask multiple questions, which can make coding slower."
Stevan Le Meur, principal product manager, developer tools, at Red Hat, pointed out that many experienced developers maintain and evolve legacy production systems, which are more intricate and complex than fresh codebases. Integrating AI into these environments often creates more churn than benefit at first, and because these systems are business critical, developers approach AI-generated outputs with heightened caution, reviewing and testing extensively before committing to the changes.
“Furthermore," said Le Meur, "mature code bases have style and culture embedded in them. This can lead to excessive tweaking of AI-generated code.”
Related Article: 10 Top AI Coding Assistants
Where AI Coding Tools Actually Shine
AI coding tools still excel in certain areas, according to Le Meur, particularly when tackling simple or repetitive tasks. “They shine in scripting, code completion, writing straightforward functions, generating test cases, handling regexes and breaking down long and complex error messages." In these scenarios, AI delivers immediate and tangible time savings, creating a strong sense of increased productivity.
“Combined with the hype, marketing and compelling demos circulating in the industry, it’s easy to walk away with the impression that these tools can supercharge all aspects of development,” said Le Meur.
Webinar
Oct
16
Agentic AI Playbook: Real-World Customer Service Use Cases You Can Deploy Now
Boost self-service by 30% and slash call volume by 63% with agentic AI.
Register
Webinar
Oct
22
Beyond Storage: Smarter Content, Bigger Impact with DAM + AI
Discover how the DAM + AI duo makes content smarter, stronger and more accessible.
Register
Webinar
Nov
6
Fix the Content Bottleneck: Build a Better WebOps Strategy
Content stalled? Dev overloaded? You’re not the only one. Learn how streamlined WebOps bridges the publishing gap.
Register
Webinar
On demand
CMS Briefing: A Live Look at What’s Next in AI-Driven Platforms
Learn how leading organizations are using AI‑driven tools to publish faster, personalize smarter and stay secure.
Watch Now
Webinar
On demand
Ready or Not: How Data-First Organizations Are Unlocking Agentforce Potential
Learn how to cut through the noise, activate Agentforce and build a Salesforce AI strategy that actually delivers.
Watch Now
Webinar
On demand
AI in Customer Service: Faster Resolutions, Happier Customers
Don’t let rising demand burn out your team. See how to build a smarter, more resilient support org.
Watch Now
He recommended developers start small and focus on using AI tools where they deliver the most value, such as in test-driven development, generating simple functions, assisting with PR reviews or automating repetitive tasks.
"Avoid the temptation to switch tools too frequently, even in a fast-moving landscape where new models and features appear almost daily." Constantly changing platforms or models forces developers to relearn interaction patterns, which can slow progress. Instead, treat tool evaluation like software testing: create repeatable test scenarios, run them consistently and gather measurable data on performance. “This structured approach bases decisions on evidence, not hype, and [helps] your AI integration remain stable and productive over time,” said Le Meur. “Don’t treat AI as just a tool; it’s a new kind of teammate.”
To avoid common pitfalls when integrating AI into development workflows, teams must start with mastering prompt engineering, he noted. “Crafting the right prompt is a skill: Too little detail, and the AI flounders; too much, and it may hallucinate or overwhelm. Balancing clarity with restraint is key.”
