AI product notes: week of 13 July 2026

Smaller product updates do not always need a full article, but they can still matter if you own the platform, run the admin console, or build against the API.

Updates

• OpenAI: OpenAI sunsets ChatGPT Pulse as proactive updates move into scheduled tasks (19 July 2026). Pulse users who relied on automated daily briefings need to migrate to scheduled tasks before the 14-day grace period ends, and the deprecation signals OpenAI’s consolidation of proactive-AI features into the ChatGPT Work/scheduled-tasks paradigm.

• OpenAI: ChatGPT gets a new speech-to-text dictation model with at least 10% lower word error rate across top languages (18 July 2026). A behind-the-scenes upgrade that improves accuracy for multilingual users and noisy environments across every ChatGPT plan tier, with measurable word error rate gains across 10+ languages.

• OpenAI: OpenAI publishes safety scorecard on teen AI use — ChatGPT Under-18 Principles now in the model spec (18 July 2026). Formalising teen-specific model behaviour as part of the spec — rather than just policy — creates an auditable safety floor that regulators and school districts can point to when evaluating ChatGPT for educational use.

• OpenAI: ChatGPT Enterprise adds workspace-scoped Admin keys and 120-day credit history in the Global Admin Console (17 July 2026). Workspace-scoped Admin keys let large organisations delegate API-level admin access by role without sharing org-wide credentials, a governance gap that blocked some enterprises from safely automating user and spend management.

• Anthropic: Claude Code ships /fork-to-background, WebSearch session caps, and a 9-fix stability sweep (17 July 2026). The /fork workflow change and WebSearch cap directly address two of the most common production complaints — context pollution across parallel tasks and runaway token spend in search-heavy agentic loops.

• Anthropic: Claude Code gets screen reader mode, /doctor checkup command, and vim insert remaps in latest release (17 July 2026). The /doctor checkup is the first built-in self-diagnostic tool for Claude Code installations, directly addressing enterprise ops complaints about configuration drift and context-bloat in long-running agentic setups.

• OpenAI: Codex 5.6-Sol hits server-overload errors on July 17 — mitigation applied (17 July 2026). Back-to-back incidents on GPT-5.6-Sol’s first full week in production signal infrastructure capacity constraints at a time when OpenAI is positioning the model as its enterprise flagship.

• Anthropic: Anthropic publishes how it runs large-scale code migrations with Claude Code (16 July 2026). First-party documentation of how Anthropic itself automates large-scale migrations with Claude Code gives developers a credible reference architecture and validates real-world multi-agent coding workflows at scale.

• Anthropic: Claude Code stability update: —forward-subagent-text flag, permission-preview security fixes, and hook decision hardening (16 July 2026). The permission-preview Unicode sanitisation fix closes a prompt-injection-adjacent security gap in human-approval workflows, making it a noteworthy security patch for teams using Claude Code’s chat-relay feature in production.

• Anthropic: Claude Code adds /optimize command for installation health, unused skills, and CLAUDE.md de-duplication (13 July 2026). As Claude Code setups accumulate plugins, MCP servers, and skill files over months, context bloat silently degrades performance and inflates token costs; /optimize gives developers a first-party tool to audit and trim that overhead.

• Google: Gemini Drive AI Overviews expand to Android and iOS with cross-file search summaries (13 July 2026). Field workers and mobile-first teams can now get cross-file knowledge retrieval from Drive on their phones without opening individual documents, extending enterprise search AI beyond the desktop.

• Microsoft: Microsoft is testing ‘PC Insights’ in Windows Copilot — real-time CPU, RAM, GPU, and storage diagnostics via AI (13 July 2026). PC Insights positions Windows Copilot as an ambient system-awareness layer for end users and IT support workflows, though the early rollout and resource-usage concerns mean IT admins should monitor before broad deployment.