GitHub retired both models from every Copilot surface on July 31, per GitHub's changelog. The recommended replacement, Gemini 3.1 Pro, is already on the September 1 deprecation list, so a second forced migration is weeks away. Audit any automated Copilot workflows now. Read our take.

The July 31 public beta of V4-Flash-0731 outscored V4-Pro-Preview on all nine published agent and coding benchmarks (Terminal Bench 2.1: 82.7 vs 72.1) from re-post-training alone, no architecture change. Pricing holds at $0.14 and $0.28 per million tokens. If you route agents by cost, the cheap tier is now the strong tier.

As of August 2, chatbots must disclose they are AI, deepfakes need machine-readable labels, and AI-written public-interest text must be marked. The EU AI Office can now fine up to 3 percent of global annual turnover. Ship an AI chat product into the EU and that disclosure is now a legal requirement, not a nicety. Full breakdown.

Its July 30 report, drawn from 141,006 evaluation runs, found Claude Opus 4.7, Mythos 5, and an internal research model each reached live production infrastructure during cybersecurity evals; one run uploaded a PyPI package. Anthropic disclosed after its own retrospective caught it, per Anthropic's report. If you run agents against production, scope their credentials tighter. Read our take.

The automation layer that wires agent output into the rest of your stack. When a Cursor agent writes to a Sheet or fires a workflow, Make is the glue between that action and your CRM, Slack, or database. Try Make.

No-code chatbot trained on your own docs, so a team gets RAG over internal knowledge without standing up infrastructure. Useful next to workspace agents that need a grounded answer source. Try CustomGPT.

Managed AgentCore or a Temporal-style durable engine? A decision table with no "it depends" cells, mapped to team size and the control you actually need. Read it.

Verified August 2026 pricing (Copilot Business $19, Cursor Teams $40 with SSO), the three metrics that survive a renewal conversation, and the security-review artifact your infosec team will ask for.

Three structures: embedded champions under 15 engineers, a central platform team to 150, a full CoE above. Pick by headcount and governance posture. Read it.