TechPrufer
Sign inGet started

Blogs›GitHub›AI-Assisted Code Development and Review

AI-Assisted Code Development and Review

AI-Assisted Code Development and Review

3
posts
2026

This feature thread tracks the evolution of how AI is integrated into the software development lifecycle. Initially, the focus was on AI's potential to accelerate code writing. This post highlights a significant shift: AI's ability to rapidly produce initial code 'probes' or 'price checks' is changing the economics of decision-making around feature requests, moving the expensive part from code implementation to the debate about whether to implement. The emphasis is now on using AI to make human intellectual toil automated, enabling rapid development loops and collaborative agent creation. The post details strategies for effective prompting, architectural design for agent-first repositories, and iteration strategies that mirror blameless culture, leading to accelerated development of new agents and features.

2026

The cost of saying yes has changed

7/17/2026

This post details how AI agents can generate initial code patches as 'price checks' for feature requests, shifting the expensive part of development from implementation to the decision-making process. It emphasizes that the generated code is a probe, not the product, and that human validation and ownership remain critical. The post introduces the concept of 'pricing uncertainty' as a new engineering skill in an AI-assisted world, where the ability to quickly assess the cost of owning a change, rather than just producing it, is paramount. It suggests moving scope discipline closer to evidence by requesting constrained AI attempts to better understand the true cost and complexity of a change.

Agent-driven development in Copilot Applied Science

3/31/2026

This post details the development of an agent-driven development workflow using GitHub Copilot CLI and Claude Opus. It introduces the concept of 'eval-agents' for automating the analysis of coding agent performance trajectories. Key contributions include strategies for effective prompting (conversational, verbose, planning-first), architectural principles for agent-first repositories (refactoring, documentation, testing), and iteration strategies based on 'blame process, not agents' to enable rapid development. The post highlights the successful creation of 11 new agents, 4 new skills, and the concept of eval-agent workflows in under three days by a team of four scientists, demonstrating the efficiency gains of this approach.

Continuous AI for accessibility: How GitHub transforms feedback into inclusion

3/12/2026

This post details the implementation of an internal workflow using GitHub Actions, GitHub Copilot, and GitHub Models to automate the processing of accessibility feedback. It describes how AI is used for triage analysis (classifying issues by WCAG violation, severity, and affected user group) and accessibility coaching, leveraging custom instructions and stored prompts. The workflow is event-driven, with GitHub Actions orchestrating steps like issue creation, Copilot analysis, and metadata population. The post also outlines the design considerations for different user personas (issue submitters, accessibility teams, program managers) and the detailed steps of the feedback flow, including intake, Copilot analysis, and submitter review.