Your daily AI digest for developers — Tuesday, March 17 2026
OpenAI Codex has announced the general availability of subagents, which are similar to Claude Code interpreters. This feature allows developers to create custom agents for specific tasks, enhancing the flexibility of AI coding tools.
The article discusses the challenges of testing AI agents and introduces a new framework designed to make this process more practical. Traditional unit tests are insufficient for capturing the complex behaviors of AI agents.
Picsart has launched an AI agent marketplace where creators can hire AI assistants to enhance their creative workflows. The marketplace will initially feature four agents, with plans to expand weekly.
Australia’s Commonwealth Bank developed its own agentic AI threat hunting tools to cope with emerging AI threats, as existing vendor solutions were too slow. The AI reduced threat response time from two days to 30 minutes.
Nvidia announced NemoClaw, an open enterprise AI agent platform built on OpenClaw, aimed at addressing security concerns. This platform is designed to enhance AI security measures in enterprise environments.
The article provides a detailed guide on building and distributing a Claude Code Skill from scratch, sharing insights and lessons learned from the process.
Mistral AI has released Mistral Small 4, a new model designed to unify instruction, reasoning, and multimodal workloads. The model features 119B parameters and is Apache 2 licensed.
This guide helps beginners set up their first GitHub Actions workflow, providing step-by-step instructions to automate software development tasks.
Gartner analyst Dennis Xu humorously suggests banning Microsoft’s Copilot AI on Friday afternoons to prevent users from overlooking potential errors due to end-of-week fatigue.
Jabez Eliezer Manuel from Booking.com shares insights into the company's AI journey, discussing challenges and lessons learned in integrating AI into their systems.