😺 New Codex, Copilot, Hermes, and Microsoft Build 2026 AI updates

2026-06-02 · Source: The Neuron · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Software Development & Engineering, Robotics & Autonomous Systems · Depth: Intermediate, long

Summary

Microsoft Build 2026 unveiled a strategic shift towards an "agent-first computer" paradigm, introducing Windows AI API updates like Aion 1.0 Instruct and Aion 1.0 Plan for local reasoning, alongside the Surface RTX Spark Dev Box, which offers 1 petaflop of AI compute and 128GB unified memory to run 120B parameter models locally. GitHub Copilot also launched a new app integrating OpenAI, Anthropic, and Google models. Simultaneously, OpenAI's Codex has surpassed 5 million weekly users, with non-developer adoption growing three times faster, now featuring role-specific plugins and interactive "Sites." Nous Research released Hermes Desktop, an open-source agent for Mac, Windows, and Linux, enabling cross-platform communication and task automation within sandboxed environments. These developments coincide with President Trump signing a frontier AI security order and Senator Bernie Sanders proposing a bill for 50% public ownership in major AI companies.

Key takeaway

For AI Engineers and Directors of AI/ML evaluating future infrastructure, prioritize solutions that support agent-first computing with robust local execution capabilities. Invest in hardware like the Surface RTX Spark Dev Box and explore open-source desktop agents such as Hermes to enable secure, on-device workflows and reduce reliance on cloud-only solutions. This approach will enhance data privacy, improve performance for sensitive tasks, and provide greater control over agent behavior and integration within existing enterprise environments.

Key insights

The AI industry is rapidly shifting towards agent-centric computing, emphasizing local execution, integrated workflows, and robust security.

Principles

• AI agents require a comprehensive stack beyond chat interfaces.
• Instructions alone do not confer new model capabilities; tools do.
• Effective prompt engineering involves iterative debugging with test cases.

Method

To debug prompts, build a tiny evaluation suite including control, edge, and capability-boundary cases, then diagnose failures as prompt, missing tool, or workflow issues, and test the smallest change.

In practice

• Utilize local AI dev machines like Surface RTX Spark Dev Box for on-device model execution.
• Implement agentic workflows by splitting complex prompts into generate-evaluate-repair loops.
• Explore open-source desktop agents like Hermes for cross-platform task automation.

Topics

• AI Agents
• Microsoft Build 2026
• OpenAI Codex
• Hermes Desktop
• On-Device AI
• Prompt Engineering
• AI Regulation

Code references

• pewdiepie-archdaemon/odysseus
• OpenMOSS/MOSS-TTS

Best for: CTO, VP of Engineering/Data, AI Architect, AI Engineer, Director of AI/ML, Tech Journalist

Related on AIssential

• Hermes Unlocks Self-Improving AI Agents, Powered by NVIDIA RTX PCs and DGX Spark — Hermes Agent and Qwen 3.6 LLMs offer a powerful, reliable, and efficient local AI agent solution on NVIDIA hardware.
• Microsoft Unveils AI-Powered Windows Features for Developers — Windows is evolving into a secure, AI-centric developer platform with enhanced local AI and agent capabilities.
• not much happened today — Open models are rapidly advancing in coding and privacy, while enterprise AI platforms are consolidating agentic capabilities and specialized hardware.
• Hermes Agent v2.0! Huge New Updates: WebUI, Qwen 3.6 Plus FREE, Computer Use, & More! — Hermes Agent's new updates enhance its autonomy, computer interaction, and multi-agent orchestration capabilities.
• Codex launch & OpenClaw/Moltbook chaos: This week in AI agents — Agent orchestration is emerging as a key differentiator and potential premium revenue source in the competitive AI agent market.
• Creating highly efficient agents: 450M tool-calling tokens distilled for post-training from top open-source models — Distilling large model capabilities into smaller, efficient agents enables advanced tool-calling on lightweight compute.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by The Neuron.