How to Measure AI Coding Productivity Beyond Raw Token Counts

·6 views

A research summary published on DEV Community examines how to accurately measure whether an AI coding session — specifically with Claude Code — is genuinely productive, rather than merely active. Raw token metrics such as input and output counts reflect activity, not outcomes, and ranking developers by these figures risks triggering Goodhart's Law, where behavior optimizes for the metric rather than actual delivery. A 2026 ICSE-SEIP paper by Chen et al., drawing on surveys of 2,989 developers at CMU and BNY Mellon, identified six productivity factors and found that commit frequency and suggestion acceptance rates cover only one of those factors. Rework rate — the percentage of AI-generated code later reverted or rewritten — has been recognized as the fifth official DORA metric since 2025 and is considered essential for measuring quality rather than just output. The article also warns that competitive activity-based rankings can push junior developers to over-rely on AI tools, potentially eroding learning and code ownership over time.

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)

Framework Helps Product Teams Decide Whether and How to Build MCP Apps by 2026

A strategic framework published on DEV Community advises most product teams to deploy at least one Model Context Protocol (MCP) app to an AI client by end of 2026. The framework centers on three diagnostic questions: where buyers are actually working today, what role the product plays in their workflow, and what the cost of absence would be. Based on those answers, teams are directed toward one of four postures — aggressive multi-client shipping, focused single-client deployment, a defensive read-only app, or no MCP build at all — each carrying distinct cost implications ranging from zero to $3 million annually. The framework stresses that choosing a posture before selecting a specific AI client is critical, as skipping that step often leads to multiple half-built integrations that never reach production. The core argument is that an MCP decision is not merely a build decision but a strategic commitment that shapes staffing, distribution, and long-term product positioning.

0 comments Read more at DEV Community

ProgrammingDEV Community ·

How One Educator Uses Generative AI for Lesson Prep Without Outsourcing Judgment

A developer-educator has shared how they use generative AI as an iterative collaborator rather than a shortcut when preparing lessons, most recently for a full Git and GitHub module. Instead of accepting the AI's first output, they describe their syllabus, class profile, and time constraints, then refine the plan through repeated pushback and revision. The approach concretely changed two habits: testing every command and deliberately triggering expected errors before class, and reusing a well-structured lesson template across subsequent sessions with AI assistance. Every piece of AI-generated material is reviewed line by line for both technical accuracy and practical fit with the actual student group. The educator emphasizes that while AI speeds up production, full responsibility for classroom content must remain with the teacher.

0 comments Read more at DEV Community

ProgrammingDEV Community ·

Next.js 15 Migration: Turbopack Gains Come With Caching and Async API Pitfalls

Next.js 15 introduces stable Turbopack support in development, React 19 compatibility, and Partial Prerendering, delivering build speeds reported to be 4–7x faster and hot module replacement latencies under 100ms. However, the upgrade brings two key breaking changes that developers must address during migration. Request-scoped APIs such as cookies(), headers(), params, and searchParams now return Promises, meaning any code that reads them synchronously will compile but fail silently at runtime. Additionally, fetch calls and GET Route Handlers are no longer cached by default, which can cause unexpected spikes in backend traffic for apps that relied on implicit caching in earlier versions. Developers are advised to audit async usage across middleware and server components, and to declare caching behavior explicitly per fetch call or route handler.

0 comments Read more at DEV Community

ProgrammingDEV Community ·

How to Import Excel Data into SQL Server Using the SSMS Import Wizard

Data professionals frequently need to transfer Excel spreadsheet data into SQL Server databases, and the SSMS Import and Export Wizard offers a straightforward way to do this without writing code. The built-in SSMS tool guides users through each step, handling tasks like data type mapping and automatic table creation. Before importing, proper data preparation is essential — columns should have consistent data types, headers must be clean and unique, and blank rows or merged cells should be removed. Date formats should also be standardized to avoid misinterpretation by the wizard during the transfer process. The guide covers the full workflow from data preparation to troubleshooting, making it accessible to both beginners and experienced database users.

0 comments Read more at DEV Community