AI Round-up: June 4, 2026

OpenAI's freshest thread was agent workflow, not a single article repeat: its GPT-Rosalind post stressed agentic coding and tool use for life-science workflows, the Wasmer case study put Codex into an edge-runtime build, and Reddit kept testing Codex limits in practice. Anthropic added services-track and cyber-threat work, xAI previewed Grok Imagine 1.5, and Microsoft supplied a Foundry manufacturing case study.

Security and cost pressure shaped the rest of the snapshot. Anthropic mapped a year of AI-enabled cyber threats, while Copilot discussion centred on usage-based billing, fast credit burn, Codex comparisons and whether Claude Code is now the better practical coding-agent spend.