My AI Brief Needed a Page

Prepare my News Beat briefing in this conversation, using today’s local date in America/New_York. Independently research current public primary sources. Do not depend on dewy, Spaces, Pages, archives or other tasks.
Cover three sections:
1. Northeast AI and agent events: prioritize Connecticut, Boston, NYC and the wider Northeast. Verify event dates, local times, venues, registration and cancellations with organizers.
2. Agentic AI: releases, research, enterprise platforms, orchestration, context, memory, harnesses and evaluation.
3. Major model providers, including OpenAI, Anthropic, Google and xAI: models, capabilities, API access, pricing and deprecations.
Prioritize the past 24 hours. Label older background and upcoming events clearly. Distinguish publication dates from event dates, previews from general availability, and analysis from facts. Never invent news, availability or metrics.
Deliver concise plain text with all three sections. For each item give what changed, why it matters, Action or Awareness, any important caveat and a direct primary-source link. State honestly when there is no material update or research is blocked. Do not create Pages, Sites or interactive UI, and do not wait for another task.

Create my interactive News Beat briefing in this conversation, using today’s local date in America/New_York. Independently research current public primary sources. Do not depend on dewy, Spaces, Pages, archives or other tasks.
Cover three sections:
1. Northeast AI and agent events: prioritize Connecticut, Boston, NYC and the wider Northeast. Verify event dates, local times, venues, registration and cancellations with organizers.
2. Agentic AI: releases, research, enterprise platforms, orchestration, context, memory, harnesses and evaluation.
3. Major model providers, including OpenAI, Anthropic, Google and xAI: models, capabilities, API access, pricing and deprecations.
Prioritize the past 24 hours. Label older background and upcoming events clearly. Distinguish publication dates from event dates, previews from general availability, and analysis from facts. Never invent news, availability or metrics.
Present a mobile-readable interactive UI with category navigation, expandable story details and clickable original sources. Each story should include what changed, why it matters, Action or Awareness, dates, caveats and relevant event logistics. Lead with a short overview. State honestly when a section has no material update.
Use a supported interactive surface and verify rendering when possible. If this scheduled-task environment cannot deliver interactive UI, provide the complete briefing as readable text with direct sources and explain that limitation. Do not claim a widget was delivered when it wasn’t. Do not create private Pages or Sites or wait for another task.

Dewy, create News Beat: my personal news and events briefing, available as both a ChatGPT Site and an interactive ChatGPT extension, with a scheduled workflow that researches and publishes a fresh edition every day.

Build the complete working product, including the initial research, interface, persistent content, archive, deployment, extension, and daily schedule.

My event region is [YOUR CITY OR REGION—for example, Connecticut, Boston, NYC, and the wider Northeast].

My timezone is [YOUR TIMEZONE—for example, America/New_York].

Run the daily research and refresh at 7:00 AM in that timezone, including weekends.

1. Purpose and content

I want a concise, trustworthy daily briefing that helps me understand what changed, why it matters, and whether I should act.

Organize it into three beats:

Local and regional AI events

Find relevant conferences, meetups, workshops, research gatherings, and networking opportunities in my specified region. Include particularly worthwhile online events.

Prioritize events about AI, agents, enterprise technology, and building AI products.

For each event, verify the organizer, date, local time, timezone, venue, format, registration link, published price, eligibility restrictions, and any registration deadline.

Check for cancellations, postponements, or meaningful changes. Distinguish check-in time from program start time where necessary.

Do not assume tickets are available simply because a registration page exists. Clearly label anything you could not verify.

Agentic AI

Cover consequential developments in:

- Agent orchestration and durable workflows.
- Context, retrieval, memory, and state.
- Agent harnesses and developer tools.
- MCP and tool integration.
- Human approvals and oversight.
- Evaluation, observability, reliability, and security.
- Enterprise agent platforms.
- Research with practical implications for builders.

Explain what is useful or different and what it could change for someone building AI applications.

Major model providers

Cover OpenAI, Anthropic, Google, and xAI, plus other providers when there is a significant development.

Track model launches, capability changes, API access, pricing, multimodal features, tool use, deprecations, and migration deadlines.

Distinguish consumer product features from API capabilities, research from released products, and previews from general availability.

2. Research and editorial standards

Research current public sources independently for every edition. Prioritize the previous 24 hours and record an explicit research cutoff.

Prefer primary sources: official announcements, documentation, changelogs, papers, and event organizer pages.

Use secondary reporting for discovery or additional context, while linking important factual claims to original evidence whenever possible.

Every story should include:

- A clear headline.
- A concise explanation of what changed.
- Why it matters.
- Action or Awareness.
- A practical next step when appropriate.
- Publication, release, update, or event dates as relevant.
- Availability and access restrictions.
- Important caveats.
- Clickable original sources.
- The time the information was verified.

Separate facts from analysis. Label older material Background and future events Upcoming.

Do not invent stories, event details, prices, availability, dates, metrics, or quotations. Unknown information should remain unknown.

Deduplicate coverage of the same announcement. Recognize repeated upcoming events as previously listed or rechecked unless something material changed.

Do not pad the briefing to reach a fixed number of stories. If a section has no material verified update, say so honestly. If research is blocked or incomplete, describe the limitation.

Choose the top story based on significance and freshness. Keep background labels visible even when older research deserves attention.

3. Visual identity

Give News Beat a clean editorial identity with generous spacing and a clear reading hierarchy.

Create a simple newsbeat. wordmark:

- “news” in dark text.
- “beat” in blue.
- A small warm-colored final dot.
- A compact icon of three colorful vertical bars beside it.

For the light theme, use a very pale cool-gray background, white content panels, soft borders, dark headlines, muted supporting text, and blue interaction accents.

Use rounded corners, restrained shadows, a clean sans-serif typeface, and comfortable line spacing.

Use small uppercase labels for categories, publishers, and section headings.

For the top story, create a wide abstract banner with a blue-to-violet-to-coral gradient and subtle overlapping circular shapes. Build this with lightweight CSS or SVG.

Keep decorative artwork separate from factual evidence. The product should feel like a polished daily publication.

4. Homepage layout

Build the homepage around this structure:

At the top, show the wordmark on the left, with the edition date and a System / Light / Dark theme selector on the right.

Below that, provide category navigation:

- All beats
- Events
- Agentic AI
- Providers

Give the active category a subtle blue selected state. Category changes should update the visible content without reloading the entire page.

Use “Your daily briefing” as the main heading. Beneath it, show the edition date and actual last successful update time.

Include a small Saved edition indicator when showing a persisted edition.

Add a utility row containing:

- A search field labeled “Search this edition…”
- An Edition archive date selector.
- A refresh button with a clear accessible label.

On desktop, use an approximately two-thirds/one-third layout:

- The wider left column contains the featured story and remaining news.
- The narrower right column contains upcoming events.

Include a short overview of the edition without overwhelming the first screen.

5. Featured story and news cards

The featured panel should contain:

- A TOP STORY label.
- The abstract gradient banner.
- Category and publisher.
- A prominent headline.
- A short summary.
- A Read the brief ↗ button.
- An article-level ＋ Add to context control inside the ChatGPT extension.

Below it, display remaining stories under NEWS, with a story count.

Use compact cards with consistent typography and metadata. Make the headline and reading action easy to find.

Show Action, Awareness, Background, or other relevant status labels clearly and consistently.

Search and category filters should work together. Provide a useful empty state and a way to clear filters.

6. Upcoming events panel

Label this panel UPCOMING EVENTS / [count].

Each event entry should include:

- A compact month-and-day date tile.
- Event name.
- City or venue.
- Local time and timezone.
- A Read the brief action.
- An individual Add to context control in the extension.

Sort upcoming events chronologically, with the soonest first.

The full event brief should contain the verified logistics, registration link, published price, eligibility requirements, and any caveats.

Keep event dates distinct from the publication date of an announcement. Do not present past events as upcoming.

7. Full article reading experience

Clicking a headline or Read the brief should open an accessible detail dialog or reading panel.

On desktop, use a centered, rounded dialog with a dimmed backdrop and a comfortable reading width. On mobile, adapt it to a full-width or nearly full-screen view.

Include:

- Category and headline.
- Full briefing text.
- Why it matters.
- Action or Awareness and any suggested next step.
- Availability, when relevant.
- Event logistics, when relevant.
- Important caveats.
- Sources & evidence.
- Verification timestamp.
- Add to context inside the extension.

Each source should show its title, publisher, relevant dates, clickable URL, and a short explanation of the claim it supports.

Provide the complete saved briefing rather than only repeating the card summary.

Support keyboard navigation, Escape to close, proper focus management, and returning focus to the originating article.

8. Mobile, themes, and application states

Make the Site and extension work well on desktop and phones.

On smaller screens, stack the columns, keep navigation usable, make controls fit the available width, and prevent horizontal scrolling.

Use readable text and comfortable touch targets.

Implement light, dark, and system themes. Persist the user’s preference and respond to system-theme changes when System is selected.

Include clear states for:

- Initial loading.
- No stories in a category.
- No search results.
- Missing editions.
- Refresh in progress.
- Failed updates.
- Older content being shown.

If an update fails, keep the last successful edition readable and display its real date and update time.

At the bottom, show the research cutoff and use the footer line:
“A little signal for a busy day.”

9. ChatGPT extension

Create a matching interactive ChatGPT app or plugin extension using the currently supported capabilities.

The extension should share the Site’s visual identity, published content, archive, and durable data.

Support natural-language requests such as:

- “Open News Beat.”
- “Show me today’s briefing.”
- “Show me upcoming AI events.”
- “Find stories about agent memory.”
- “Open the edition for [date].”
- “Show me the sources for this story.”

Provide an actual embedded reading interface inside ChatGPT, with category navigation, search, article details, and edition selection.

Every article and event must have its own ＋ Add to context control.

Let me select multiple articles independently, see their selected state, and remove individual selections.

The context supplied to ChatGPT should identify the selected story and edition and include relevant briefing content, caveats, and original sources within supported host limits.

Selecting an article should prepare context for my next message. It should not automatically send a message or start a new assistant turn.

I should be able to select a story and then ask:
“What does this mean for the product I’m building?”

In the standalone Site, gracefully disable host-only context controls and explain that they are available inside the ChatGPT extension.

Use the official install/connect flow where required. Verify the actual embedded interface and context delivery when the host is accessible; describe any remaining limitation honestly.

10. Persistent content and archive

Create a dedicated supporting ChatGPT Space containing:

- The product and editorial brief.
- The daily research and publication instructions.
- An editions index.
- A dated Page for each completed edition.

Preserve complete article bodies, sources, research cutoff, verification timestamps, and corrections.

Use the Space as the canonical editorial source and supported durable server storage for validated snapshots consumed by the Site and extension.

Give editions and stories stable identifiers and retain links back to their source Pages.

The Site and extension should display the same published edition. Content updates should not require rebuilding the whole interface.

Persist editions across sessions and devices. Use browser-local storage for display preferences rather than as the only copy of the content.

Show only real saved editions in the archive. On first use, show today’s initial edition; build the historical archive as subsequent editions are published.

11. Daily scheduled research and publication

Create one enabled workflow that runs every day at 7:00 AM in my specified timezone, including weekends.

Use a named timezone so daylight-saving changes are handled correctly.

Each scheduled run should:

1. Determine today’s local date.
2. Research current sources across all three beats.
3. Verify claims and event logistics.
4. Deduplicate and select material developments.
5. Write the complete edition.
6. Save the canonical dated content.
7. Validate and publish the durable snapshot.
8. Read back the published result.
9. Update the latest-edition pointer and archive.
10. Make the same edition available in the Site and extension.
11. Record the actual outcome and completion time.

The workflow must run in the supported hosted environment without requiring my browser to remain open or my personal computer to stay awake.

Store sufficient instructions in the scheduled task or an accessible runbook so it can operate independently of this conversation.

Check for an existing schedule linked to this product before creating another. Avoid duplicate jobs.

Make retries safe: repeated runs for the same date must not create duplicate editions or erase corrections and human edits.

Only replace the latest published edition after the new one has been validated. If research or publication fails, retain the previous successful edition and mark its freshness accurately.

Verify the saved schedule’s time, timezone, enabled state, and next run when available. Distinguish an immediate successful test from a future unattended run that has not happened yet.

12. Refresh behavior

The normal refresh button should reload the latest saved edition.

If you provide an owner-only Research and update now action, label it distinctly and use the same publication workflow as the scheduled job.

Prevent overlapping research runs and duplicate publication. Display progress and the actual result.

Opening the Site should promptly load published content without starting a fresh research job for every visitor.

13. Build, verify, and deliver

Use the current supported Sites, Spaces, extension, storage, and scheduling capabilities available to you.

Proceed with sensible implementation defaults and ask only for essential missing information or required user actions.

Research and publish the initial edition now, build and deploy the Site, prepare the ChatGPT extension, and create the daily schedule.

Before finishing, verify:

- The Site is deployed and accessible.
- The initial edition contains real source-backed content.
- Each beat contains relevant coverage or an honest no-update statement.
- Search, filters, article details, and source links work.
- Event dates and logistics display correctly.
- The archive loads saved editions.
- Desktop and mobile layouts work.
- Theme preferences work.
- The Site and extension read the same edition.
- The extension displays its interactive interface.
- Context selection and removal work in the actual ChatGPT host when accessible.
- The daily schedule is saved and enabled.
- An immediate end-to-end test exercises the publication path.
- A failed update preserves the last successful edition.

Keep the new product owner-private initially.

Finish with the Site link, supporting Space link, extension install/open control, daily schedule details, a brief verification summary, and any specific action still required from me.