Devlery

Blog

Notes and analysis on AI development.

Copilot turns issue search into an agent work map

Copilot turns issue search into an agent work map

GitHub added semantic issue search and task-based auto model routing to Copilot, pushing coding agents from code generation toward workflow operations.

Docusign Agent Studio turns contracts into an execution layer

Docusign Agent Studio turns contracts into an execution layer

Docusign unveiled an Iris-powered AI assistant, agents, Agent Studio, and an MCP beta. The bigger shift is contracts moving from signed PDFs into enterprise workflow execution.

34% fewer revisits, the agent cost of clean code

34% fewer revisits, the agent cost of clean code

A 660-run SonarSource study with Claude Code suggests clean code may not raise pass rates, but it can reduce tokens and file revisits.

Ramp moved Codex review from hours to minutes

Ramp moved Codex review from hours to minutes

OpenAI's Ramp case study shows Codex code review moving beyond demos into required review flows, trust loops, and on-call agent tooling.

1,000 Desktop Tasks Turn Computer-Use Agents Into Verifiable Systems

1,000 Desktop Tasks Turn Computer-Use Agents Into Verifiable Systems

OpenComputer shifts computer-use agent evaluation from LLM judges to reproducible desktop tasks and app-state verifiers.

Runtime Launch HN Shows the New Isolation Layer for Team Coding Agents

Runtime Launch HN Shows the New Isolation Layer for Team Coding Agents

Runtime's Launch HN shows coding agents moving from personal IDE workflows into team sandboxes, secrets control, observability, and PR governance.

The SDK Generator Shut Down, and Claude Took the Connection Layer

The SDK Generator Shut Down, and Claude Took the Connection Layer

Anthropic’s Stainless acquisition shows Claude’s moat expanding from model quality into SDKs, MCP servers, and reliable API connectivity.

Chrome gives coding agents modern web sense

Chrome gives coding agents modern web sense

Modern Web Guidance is Chrome's early preview agent skill for giving coding agents Baseline-aware web APIs, fallbacks, and evaluation context.

Exa Puts a $250M Price Tag on Agent Search

Exa Puts a $250M Price Tag on Agent Search

Exa's $250M Series C shows how AI agent competition is moving beyond models into search, indexing, latency, and source-verification infrastructure.

MCP Comes to the Phone, and Local Agents Get a New Runtime Boundary

MCP Comes to the Phone, and Local Agents Get a New Runtime Boundary

Google AI Edge Gallery now combines Gemma 4 on-device agents with MCP, local notifications, and persistent sessions.

The 24-hour agent inside Google Search and the new edge of the link economy

The 24-hour agent inside Google Search and the new edge of the link economy

Google Search agents expand AI Mode from link retrieval into background monitoring, booking, commerce, and generated mini-apps.

AI-Q Skill and the Data Boundary for Research Agents

AI-Q Skill and the Data Boundary for Research Agents

NVIDIA AI-Q agent skill lets Claude Code, Codex, and other harnesses delegate enterprise research to a local AI-Q server.