Hey Devs! This week brought new models, better coding agents, and more tools for building reliable AI workflows. Here are the main updates from July 19–26, 2026.
OpenAI improves Codex multi-agent workflows
Codex CLI 0.145.0 improves Multi-Agent V2 with configurable subagent models, reasoning levels, roles, and concurrency.
Codex also adds better imports from Claude Code and Cursor, experimental thread history and memory, audio input, inline visualizations, and multi-folder project support.
This makes it easier to balance speed, cost, and reasoning quality across different agents.
Anthropic releases Claude Opus 5
Claude Opus 5 is designed for complex coding, debugging, large repositories, and long-running agent tasks.
It includes a 1-million-token context window and configurable effort levels. It is also rolling out in GitHub Copilot.
Read the Anthropic announcement
Google launches Gemini 3.6 Flash
Gemini 3.6 Flash focuses on coding, multimodal applications, computer use, and production AI agents.
Google also introduced Gemini 3.5 Flash-Lite for faster and cheaper high-volume tasks such as classification, extraction, and data processing.
GitHub adds controls for AI agents
GitHub Issues can now show an agent’s confidence and reasoning before it applies a change.
Teams can automatically accept high-confidence actions while sending uncertain changes for human review. GitHub also made its Copilot cloud agent integration for Linear generally available.
Read about GitHub’s agent controls
Microsoft releases its Agent Framework Harness
Microsoft’s new framework for Python and .NET includes agent loops, planning, memory, crash recovery, approvals, context management, and observability.
Declarative Workflows 1.0 also allows developers to define agent workflows and approval steps using YAML.
Explore Microsoft Agent Framework
Cursor introduces automatic model routing
Cursor Router automatically selects a suitable model for each request.
Developers can optimize routing for intelligence, balance, or cost. This could reduce the need to manually select a model for every coding task.
JetBrains improves context for coding agents
JetBrains Context creates a semantic repository index for Claude, Codex, and Junie.
It helps agents find relevant code across one or multiple repositories while potentially reducing agent turns, latency, and cost.
Read the JetBrains announcement
I Tested GPT-5.6 Sol Reasoning Levels So You Don’t Have To
Does higher GPT-5.6 Sol reasoning actually improve your code? See how four different settings affect coding task results and speed.
Watch on YouTube:
Thanks for reading! Subscribe to AI News for Developers for more practical AI and developer updates.
Cheers, proflead! ;)