Codex Goes Mobile, Anthropic Kills Compute Arbitrage, and Agents Hit Finance at Scale

Tools

OpenAI Ships Codex Mobile: Your Phone Becomes a Remote Control for Running Coding Agents

Codex is now in the ChatGPT iOS and Android apps as a remote supervision layer for sessions already running on your Mac or devbox. You can approve decisions, steer execution, and review diffs without being tethered to a desk, which matters a lot as agents run multi-hour tasks. Over 4 million weekly Codex users means this hits a real workflow gap immediately.

Dotmatics Luma Agent Ships as an AI Co-Scientist That Executes Multi-Step Lab Workflows on Structured Scientific Data

Luma Agent is notable for solving the governance problem that kills most life sciences AI deployments: every action is logged with full audit trails, requires human approval before data changes, and is built on ontology-backed structured data rather than raw unstructured text. Gartner predicts 80% of agentic AI initiatives in healthcare and life sciences will fail governance checkpoints in 2026, and this architecture is specifically designed to pass that gate. For pharma and biotech builders, this is the production-ready template to benchmark against.

Honeycomb Ships Agent Timeline and Canvas Skills to Give Engineers Full Visibility Into Multi-Agent Workflows

Honeycomb added Agent Timeline for visualizing multi-agent, multi-trace workflows and a rebuilt Canvas workspace that works as both a chat interface and an autonomous debugging agent with reusable Skills. This addresses a real production pain point: when an agent chain fails, you currently have very little tooling to reconstruct which LLM call, tool invocation, or downstream effect caused it. Agent Timeline is in Early Access now.

News

Broadridge Goes Live with Agentic AI Across 40+ Institutional Clients in Post-Trade and Wealth Ops

Broadridge just flipped its agentic capabilities from BPO pilot to production, covering trade fails, account workflows, valuation exceptions, and client inquiry automation at firms processing trillions in daily volume. The key differentiator is a completed financial services data ontology built from 60 years of operational data, which is the actual moat here, not the model layer. Up to 30% day-one cost reduction is a credible number when the data layer is pre-normalized.

Analysis

Metered Agent Billing Is Coming for Everyone: Why Anthropic's Credit Split Is a Preview of the Whole Industry

Analysts quoted in this piece frame the Anthropic credit overhaul not as a one-off but as the leading edge of a broader shift where every AI vendor separates interactive chat limits from programmatic agent usage and bills the latter like cloud compute. GitHub Copilot is already moving the same direction, and OpenAI has always priced this way on the API side. The practical takeaway is that "treat your Claude usage like AWS" is now the correct mental model for anyone building production agent pipelines.