Study Finds Experienced Developers 19% Slower When Using AI Tools

·1 views

A study titled 'Measuring the Impact of Early-2025 AI on Experienced Open-Source Developer Productivity' found that senior developers took 19% longer to resolve issues when using AI tools, despite perceiving a 20% boost in their own productivity. The nearly 40-point gap between perception and reality is attributed to the overhead of using AI — crafting prompts, waiting for file indexing, and processing context — steps that can outweigh the time savings for highly experienced engineers. The research focused exclusively on senior-level open-source developers, suggesting AI may deliver less productivity benefit at higher experience levels than commonly assumed. However, the study's conclusions are limited by a small sample size of just 16 developers, making broader generalization difficult. The findings have reignited debate in the software industry about whether AI tools genuinely improve output for staff and principal engineers, or are better suited to mid-level developers.

Read the full story at DEV Community

This is an AI-generated summary. ShortSingh links to the original source for the complete article.

Discussion (0)

How Agent-to-Agent Marketplaces Work in 2026: A Technical Overview

Autonomous AI agents are evolving into programmable micro-services capable of discovering, negotiating with, and paying other agents for compute, data, or specialised tasks. A developer guide published on DEV Community outlines the core components needed to build or use an agent-to-agent (A2A) marketplace, including a registry service, structured descriptor schema, messaging bus, and a settlement layer. The settlement layer uses ERC-20 contracts, specifically USDC on the Base blockchain, to lock and release funds based on proof of task completion. Standardised JSON-LD descriptors define each service's inputs, outputs, pricing, and SLA terms, and are made immutable via IPFS or on-chain storage so consumers can verify them before use. The guide also provides illustrative Python code demonstrating how an agent can query a registry, sign a request, escrow payment, and release funds upon successful service delivery.

0 comments Read more at DEV Community

ProgrammingDEV Community ·

Japanese Engineer Turns 11 Years of Legacy .NET Experience Into AI-Assisted Modernization Plan

A Japan-based software engineer with over eleven years of C# and .NET experience set out to find an easy, AI-driven income stream after noticing viral overseas money-making schemes that were largely unavailable in Japan. After consulting both GPT and Claude, both AI tools independently suggested leveraging his existing engineering background rather than chasing passive-income shortcuts. The recommendation pointed toward legacy system modernization, a sector where Japan's Digital Agency reported in 2025 that 61% of companies — and 74% of large enterprises — still operate legacy systems. With .NET 8 reaching end-of-support in November 2026 and .NET 10 already released, demand for framework upgrade projects is actively rising. However, the engineer questions whether AI tools like GitHub Copilot can truly handle the messy, undocumented real-world legacy systems he has spent his career working on.

0 comments Read more at DEV Community

ProgrammingDEV Community ·

Docgrity Tool Aims to Flag Outdated Documentation Used by AI Agents

A developer named Ujjavala has launched an open-source tool called Docgrity, designed to verify the integrity of software documentation. The tool targets a growing problem where AI agents act on outdated or inaccurate docs, potentially producing unreliable outputs. Docgrity is intended for individual developers, teams, and AI agent workflows alike. The project was announced on DEV Community on September 5 and is also listed on Product Hunt.

0 comments Read more at DEV Community

ProgrammingDEV Community ·

How One HR Team Used AI to Tame a 500-Application Recruitment Surge

Viktoriia Kopeikina, HR Manager at 111Techs, describes how a flood of over 500 applications for a single QA engineer role in early 2024 exposed the limits of manual CV screening, taking nearly two months to fill the position. The challenge is not unique: Djinni analytics data from June 2026 shows nearly 90,000 candidates competing for fewer than 8,000 vacancies in Ukraine's tech sector alone. To manage the volume, the team first introduced a structured Google Form to channel applications into a spreadsheet, marking an initial step toward automation. Key pain points included mass-applying candidates who ignored job requirements, mismatched seniority levels, and CVs arriving in inconsistent formats that were difficult to parse at scale. The article argues that AI-assisted screening is becoming necessary not to replace human judgment but to ensure strong candidates are not overlooked simply due to application volume.

0 comments Read more at DEV Community