Reading up on GitHub Copilot

Paul Bakaus argues AI agents need “skill engineering” to give designers precise control, rejecting full automation in favor of human judgment steering the final 20% of creative work.

Several vendors now offer intelligent model routers that automatically select the cheapest sufficient LLM per task, promising 20-30% cost savings, a trend expected to become table stakes.

Git worktrees let developers work on multiple branches simultaneously without stashing or context-switching, with each branch in its own folder.

Coding agents have become more capable but face fundamental bottlenecks like limited context, assumption-making, technical debt accumulation, and coordination overhead that prevent them from scaling productivity gains proportionally in larger organizations.

Satya Nadella argues Microsoft's AI advantage lies in providing a platform for enterprises to build their own hill-climbing machines, not in owning a frontier model.

GitHub COO Kyle Daigle discusses how AI agents caused platform activity to surge 1400% in 2026, straining infrastructure and forcing GitHub to rethink open-source social contracts and reliability.

The article claims that AI coding assistants like ChatGPT and Copilot have made programming books largely obsolete, citing massive user growth as evidence.

Microsoft is canceling Claude Code licenses for thousands of employees due to unsustainable token costs, signaling that enterprise AI coding's unit economics don't work at current prices.

A staff engineer now uses LLM agents for nearly all code changes, bug hunting, and testing, with light supervision, but avoids AI for communication and UI work.

Flat-rate AI subscriptions broke in April 2026 as agentic workloads caused unsustainable unit economics, forcing providers to shift to per-token billing.

Shopify approaches 100% daily AI tool adoption among employees, with CLI tools like Claude Code and Codex gaining share while IDE-based tools decline, though top percentile token users grow faster than the rest.

Running multiple $10K MRR businesses on a $20/month stack with SQLite, a VPS, and minimal services is possible, but the approach is controversial among developers.

Developer portals should optimize for AI coding agents by structuring documentation with llms.txt, token-efficient formatting, and capability signaling via skill.md.

Microsoft's APM is an open-source dependency manager for AI agents that lets teams declare agent configurations in a manifest file and reproduce them across tools like Copilot, Claude, and Cursor.

Professional AI-assisted coding yields mixed results: high productivity on greenfield tasks but significant technical debt and deskilling risks in complex codebases.

GitHub markets its AI-powered platform, featuring Copilot for code writing and agent mode, collaboration tools, and AI-driven security, as essential for modern software development.

A.I. coding agents shift programmers from writing code to orchestrating agents, debugging, and prompt engineering, fundamentally changing the role.

Big tech engineers report AI tools increase PR volume and development speed but also introduce more bugs, reduce code review quality, and shift culture toward quantity over quality.

Stripe previewed a billing feature that lets AI companies track token usage, pass through costs, and automatically apply a profit markup on underlying model fees.

An article laments that AI coding tools are destroying the craft of programming, but commenters largely disagree, finding the tools empowering.

Claude and Codex coding agents are now in public preview for Copilot Pro+ and Copilot Enterprise customers, enabling agent sessions from GitHub.com, mobile, and VS Code.

Hacker News commenters report that Microsoft engineers are turning to Anthropic's Claude Code over GitHub Copilot due to superior quality and reliability.

AgentRC is an experimental Microsoft tool that measures a repository's AI-readiness and generates contextual instruction files for coding agents.

AI is expected to eliminate entry-level chip design tasks but may allow students trained on AI tools to start in more senior positions, while domain expertise remains critical.

As AI makes coding cheaper, operational excellence and reliability engineering become the key differentiator, shifting software engineering's focus from writing code to keeping it running.

Block Engineering's AI Champions program and repo readiness initiatives boosted AI-authored code by 69% and time savings by 37% in three months.

Docfind uses Rust, WebAssembly, and FSTs to create a compact client-side search engine that runs entirely in the browser, with Copilot accelerating development.

The GitHub Copilot SDK technical preview exposes the same agentic execution loop used by Copilot CLI as a programmable layer for any application.

Skills.sh provides a directory of reusable capabilities that can be installed with a single command to enhance AI coding agents like Claude Code, Cursor, and Copilot.

Writing effective AI agent specs requires a high-level vision, structured PRD-like format, modular task breakdowns, and built-in guardrails to keep agents focused and productive.

The article argues that AI coding assistants are degrading because training data quality drops as inexperienced coders accept flawed code, creating a feedback loop.

AI-generated code shifts the bottleneck from writing to verification; human code review remains essential for security, context, and accountability.

AI Observer is a self-hosted, single-binary OpenTelemetry observability backend that tracks token usage, costs, latency, and errors across multiple AI coding assistants via a local dashboard.

Compound engineering uses AI agents, automated testing, and rapid feedback loops to achieve 300-700% productivity gains by treating each completed task as an investment that accelerates future development. The feedback loop becomes the critical bottleneck when code generation speed approaches near-instant.

Code review catches mostly style issues, not production bugs, costing a typical 80-engineer firm $3.6M annually while preventing few real incidents.

Addy Osmani shares a disciplined AI-assisted coding workflow emphasizing planning, scope management, context provision, testing, and frequent commits to maintain developer accountability.

Microsoft is deprecating its free, local IntelliCode AI code completion extensions in VS Code and pushing developers toward subscription-based GitHub Copilot with usage caps.

Peter Norvig pits several LLMs against Advent of Code 2025 puzzles, comparing their reasoning and coding abilities to gauge current AI programming skill.

Anthropic launches Claude Code in Slack, letting developers delegate coding tasks from chat threads, signaling a shift toward AI-embedded collaboration.

Claude failed to accurately recreate the 1996 Space Jam website layout from a screenshot, despite getting close, highlighting spatial reasoning limits.

A well-structured CLAUDE.md file with focused instructions and a table-of-contents approach improves AI coding assistant performance, supported by data on instruction-following quality.

Analyzing 2,500+ agents.md files reveals that effective agent definitions require specific personas, executable commands, code examples, and clear boundaries.