AI Product Updates Daily - June 18, 2026

OpenAI turned scheduled tasks into a fuller proactive-work surface, GitHub shipped several Copilot agent features into general availability, Google connected Gemini to Classroom context, and Adobe pushed AI deeper into creative and marketing workflows. The day had no confirmed frontier-model launch from Anthropic, Google DeepMind, Mistral, or OpenAI; the practical movement was in product surfaces, routing, and workflow automation.

ChatGPT scheduled tasks moved from a narrow reminder feature toward a managed automation surface. The new Scheduled page sits in the sidebar and lets users view active tasks, see the next run time, pause, resume, edit, or delete tasks. OpenAI also says tasks are now faster and more reliable, can be scheduled for specific times or wider windows such as morning or evening, and can monitor the web or connected apps for changes before notifying the user. 1

The important product change is the Pulse sunset. OpenAI says Pulse is being sunset as proactive updates move into scheduled tasks; Pro users keep Pulse for 14 days from June 17, 2026. The replacement action is simple: ask ChatGPT to schedule a daily briefing based on interests and past chats. 1

Availability is limited to Plus, Pro, Business, and Enterprise. Active task limits vary by tier. Tasks cannot run more than once per hour, and unattended tasks may pause automatically after inactivity. 1

GitHub had the densest product cluster of the day. The Copilot app is now generally available on macOS, Windows, and Linux. It lets users start sessions from an issue, pull request, or prompt; run parallel sessions across repositories; review diffs; validate work in an integrated terminal and browser; and open PRs through existing team checks. 2

Since the technical preview, GitHub added canvases for shared human-agent work surfaces, cloud automations for scheduled agent work, and bring-your-own-model / tool connections through MCP servers. Business and Enterprise access depends on admins enabling Copilot CLI in policy settings. 2

Agent finder is the governance piece. Copilot can search a registry of available AI resources, then return ranked matches that the agent can pull in on demand. GitHub says the feature implements the Agentic Resource Discovery specification, developed with Google, GoDaddy, Hugging Face, and Microsoft. It can point at GitHub's public catalog or a private internal registry, respects managed settings, and does not auto-install resources. 3

Auto model selection is now GA in Copilot Chat on github.com and GitHub mobile. It routes among models such as Claude Sonnet 4.6, GPT-5.4 mini, GPT-5.4, and Haiku 4.5 based on plan, policies, availability, and task complexity. GitHub says paid subscribers get a 10% discount on token use when they use auto. 4

GitHub also shipped smaller but practical workflow changes: GitHub CLI v2.95.0 adds gh repo read-file and gh repo read-dir for inspecting public or private repositories without cloning, and generated release notes now credit the human requester beside @copilot when a Copilot cloud-agent PR is merged. 10 5

Google's Classroom app in Gemini gives Gemini access to class context when prompts need it. Teachers can draft communications and assignments, generate differentiated materials, summarize student progress, identify where students may need support, update draft assignment details across classes, make seating charts, and find old assignments. Students age 18 and older can ask for upcoming assignments, study plans, practice based on Classroom content, and recent announcements. 6

Gemini may infer that Classroom context is relevant, or users can explicitly type @Classroom. The feature cannot enter grades, provide private feedback directly, delete/archive/post assignments or announcements, or create rubrics. It can post to draft. 6

Rollout covers Rapid Release and Scheduled Release domains, with expected completion by June 18, 2026. It is available in English to Workspace customers who are 18 or older and signed in to the Gemini app. Education domains get the Classroom app on by default; other domains get it off by default. Google also notes that the Classroom app in Gemini does not currently support data regionalization, with support expected later this year. 6

Photoshop 27.8 expands Generate Image beyond Adobe's own Firefly Image Models. CG Channel reports that Photoshop can now use third-party models from Google's Gemini family, including Gemini 3.1 with Nano Banana 2, plus Black Forest Labs' FLUX.2 Pro. The same report says beta Custom Models can be trained on 10-30 images, with training using 500 generative credits and each generation using 20 credits. 7

Lightroom Classic 15.4 adds automatic duplicate detection, stacks duplicate candidates for review, improves people-focused culling through a Faces panel, improves subject selection in busy scenes, adds AI-focused metadata filters and Smart Collections, and syncs keywords across Lightroom desktop, web, and mobile. 8 Lightroom on the web also added AI-powered blemish removal for one or multiple people in a photo. 11

Adobe Brand Visibility is aimed at marketing teams rather than creators. Storyboard18 reports that the product combines Semrush visibility data with Adobe content optimization tools, giving marketers access to nearly 300 million real-world AI search prompts and tracking brand mentions, audience reach, competitive share of voice, and content gaps across AI surfaces such as ChatGPT, Google AI Mode, Microsoft Copilot, and Perplexity. 9

Snap's SPECS launch was dated June 16, but it is still relevant for product teams because the developer tooling is an AI-agent story. Snap introduced a developer preview of agentic development for building SPECS Lenses in Lens Studio, with Lens Studio agents and skills available inside Claude Code, Codex, and Cursor. The company also announced the SPECS Spatial Benchmark, a Migration Agent for porting projects from tools like Unity into Lens Studio, and a Native Development Kit for bringing C and C++ code into Lenses. 12

SPECS themselves are priced at $2,195 with a $200 refundable deposit, with shipment expected this fall in the United States, United Kingdom, and France. The glasses are standalone, use two Snapdragon processors, claim 7 ms motion-to-photon latency, and offer up to four hours of mixed-use battery life plus a case for up to 20 total hours. 13

No official page or changelog found a new Anthropic model restoration, new Mistral model, new Perplexity API launch, or new OpenAI frontier-model release inside today's verified sweep. Anthropic's Fable 5 / Mythos 5 access situation remains a watch item, but there was no confirmed product availability change suitable for this product-update digest.