The AI Productivity Paradox Report 2025

AI coding assistants increase developer output, but not company productivity

Generative AI is rewriting the rules of software development—but not always in the way leaders expect. While over 75% of developers are now using AI coding assistants, many organizations report a disconnect: developers say they’re working faster, but companies are not seeing measurable improvement in delivery velocity or business outcomes.

Drawing on telemetry from over 10,000 developers across 1,255 teams, Faros AI’s recent landmark research report confirms: 

• Developers using AI are writing more code and completing more tasks
• Developers using AI are parallelizing more workstreams
• AI-augmented code is getting bigger and buggier, and shifting the bottleneck to review
• Any correlation between AI adoption and key performance metrics evaporates at the  company level

This phenomenon, which we term the “AI productivity paradox,” raises important questions and concerns about why widespread individual adoption is not translating into significant business outcomes and how AI-transformation leaders should chart the road ahead. 

For engineering leaders looking to unlock AI’s full potential, the data points to both promising leverage and persistent friction. 

Our key findings continue below. 

#1 Individual throughput soars, review queues balloon

Developers on teams with high AI adoption complete 21% more tasks and merge 98% more pull requests, but PR review time increases 91%, revealing a critical bottleneck: human approval. 

AI‑driven coding gains evaporate when review bottlenecks, brittle testing, and slow release pipelines can’t match the new velocity—a reality captured by Amdahl’s Law: a system moves only as fast as its slowest link. Without lifecycle-wide modernization, AI’s benefits are quickly neutralized.

#2 Engineers juggle more workstreams per day

Developers on teams with high AI adoption touch 9% more tasks and 47% more pull requests per day. 

Historically, context switching has been viewed as a negative indicator, correlated with cognitive overload and reduced focus. 

AI is shifting that benchmark, signaling the emergence of a new operating model: in the AI-augmented environment, developers are not just writing code—they are initiating, unblocking, and validating AI-generated contributions across multiple workstreams. 

As the developer’s role evolves to include more orchestration and oversight, higher context switching is expected.

#3 Code structure improves, but quality worsens 

While we observe a modest correlation between AI usage and positive quality indicators (fewer code smells and higher test coverage from limited time series data), AI adoption is consistently associated with a 9% increase in bugs per developer and a 154% increase in average PR size.

AI may support better structure or test coverage in some cases, but it also amplifies volume and complexity, placing greater pressure on review and testing systems downstream. 

#4 No measurable organizational impact from AI

Despite these team-level changes, we observed no significant correlation between AI adoption and improvements at the company level. 

Across overall throughput, DORA metrics, and quality KPIs, the gains observed in team behavior do not scale when aggregated. 

This suggests that downstream bottlenecks are absorbing the value created by AI tools, and that inconsistent AI adoption patterns throughout the organization—where teams often rely on each other—are erasing team-level gains.

Four AI adoption patterns help explain the plateau

Even with rising usage, we identified four adoption patterns that help explain why team-level AI gains often fail to scale, namely: 

1. AI adoption only recently reached critical mass. In most companies, widespread usage (>60% weekly active users) only began in the last two to three quarters, suggesting that adoption maturity and supporting systems are still developing. 
2. Usage remains uneven across teams, even where overall adoption appears strong. And because software delivery is inherently cross-functional, accelerating one team in isolation rarely translates to meaningful gains at the organizational level.
3. Adoption skews toward less tenured engineers. Usage is highest among engineers who are newer to the company (not to be confused with junior engineers who are new to the profession). This likely reflects how newer hires lean on AI tools to navigate unfamiliar codebases and accelerate early contributions. In contrast, lower adoption among senior engineers may signal skepticism about AI’s ability to support more complex tasks that depend on deep system knowledge and organizational context.
4. AI usage remains surface-level. Across the dataset, most developers use only autocomplete features. Advanced capabilities like chat, context-aware review, or agentic task execution remain largely untapped. 

What should engineering leaders do next?

In most organizations, AI usage is still driven by bottom-up experimentation with no structure, training, overarching strategy, instrumentation, or best practice sharing. 

The rare companies that are seeing performance gains employ specific strategies that the whole industry will need to adopt for AI coding co-pilots to provide a measurable return on investment at scale.

Explore the full report to uncover these strategies plus the five enablers—workflow design, governance, infrastructure, training, and cross‑functional alignment—that prime your organization for agentic development.

{{ai-paradox}}

Methodology Note

Background
This study analyzes the impact of AI coding assistants on software engineering teams, based on telemetry from task management systems, IDEs, static code analysis tools, CI/CD pipelines, version control systems, incident management systems, and metadata from HR systems, from 1,255 teams and over 10,000 developers across multiple companies. The analysis focuses on development teams and covers up to two years of history, aggregated by quarter, as teams increased AI adoption.

Definitions
We define AI adoption in this report as the usage of developer-facing AI coding assistants—tools including GitHub Copilot, Cursor, Claude Code, Windsurf, and similar. These are generative AI development assistants that integrate directly into the software development workflow—typically through IDEs or chat interfaces—to help developers write, refactor, and understand code faster. Increasingly, these tools are expanding beyond autocomplete to offer agentic modes, where they can autonomously draft pull requests, run tests, fix bugs, and perform multi-step tasks with minimal human intervention.

Approach
To isolate the relationship between AI adoption and engineering outcomes, we:

• Standardized all metrics per company to remove inter-org variance
• Used Spearman rank correlation (ρ) to assess relationships of metrics to AI usage 
• Reported only those metrics with data from ≥6 companies and statistically significant correlations (p-value < 0.05)
• For each team, we calculated the percent change in metric values between the two quarters with the lowest AI adoption and the two quarters with the highest
• Excluded outlier data and metrics with insufficient historical coverage

This approach enables comparisons within each company over time and avoids misleading aggregate assumptions across different org structures.

Versioning note: This version of the report reflects analysis as of June 2025. Future editions may expand coverage as AI usage matures across more organizations and product features evolve.

About Faros AI

Faros AI improves engineering efficiency and the developer experience. By integrating data across source control, project management, CI/CD, incident tracking, and HR systems, Faros gives engineering leaders the visibility and insight they need to drive velocity, quality, and efficiency at scale. Enterprises use Faros AI to transform how software is delivered—backed by data, not guesswork.

Learn more at www.faros.ai