Two weeks ago, we published the AI Productivity Paradox Report 2025, a landmark study that exposes the disconnect between the adoption of AI coding assistants and their organizational impact. Developer output increases, but engineering outcomes are flat. 

We also identified common AI adoption missteps that explain this paradox, including slow uptake, uneven usage, adoption that skews to less tenured engineers, and surface‑level tool usage. 

{{ai-paradox}}

Today, we examine another angle of the report: The systemic barriers that sap productivity momentum even after AI coding assistants reach critical mass, and what top‑performing companies are doing to beat the odds.

Why AI gains stall: Three systemic barriers

Developers using AI complete 98% more code changes and 21% more tasks. But these gains evaporate at the company level, where neither a positive nor a negative impact can be observed. 

Why is this happening? Three systemic barriers keep coming up in operational fieldwork: 

1. Downstream bottlenecks cancel out upstream gains

AI accelerates code creation, but review queues, brittle test suites, and sluggish release pipelines remain stuck in yesterday’s gear. By Amdahl’s Law, your delivery engine only moves as fast as its slowest stage—so faster coding simply piles more work onto the choke points.

2. Grassroots adoption lacks structure and scale

AI adoption is still driven by bottom-up experimentation, with developer enthusiasm undermined by a lack of centralized enablement. Developers spend time navigating tools without guidance, users receive little to no formal training, and there's rarely a strategy tailored to role or experience—resulting in inconsistent outcomes and uneven utilization. Without shared best practices and strong internal communities to socialize tips and recommendations, the organization struggles to convert adoption into lasting impact.

3. Directionless deployment drains ROI

Simply handing out licenses to Copilot, Claude Code, or Cursor isn’t a strategy. Without clear goals, usage policies, and change‑management plans aligned to business priorities, AI becomes “just another tool” instead of a catalyst for transformation.

What high-performing companies do differently

Some companies are seeing greater success and higher ROI from their AI investments. Their edge stems from  three mutually reinforcing practices:

Blueprint for operationalizing AI engineering

As software teams transition from AI-assisted coding to agentic development, the complexity and autonomy of AI participation will increase. This creates new coordination demands, where code may be written, reviewed, or executed by agents working in parallel with humans.

Read the comprehensive research to discover practical steps that scale AI through the entire lifecycle, set the stage for agentic development, and ready your organization for the next phase of AI‑driven innovation.