In partnership with
Welcome Byte
This week marks a shift from demo to delivery. OpenAI’s DevDay unlocked apps in ChatGPTand a proper agent toolkit, Google released a Computer Use model that actually clicks and types for you, and Slack/Notion pushed deeper into agent workflows. Below: what changed, why it matters, and copy‑paste prompts you can use today.
🚀 Top AI Updates
OpenAI DevDay ships: Apps in ChatGPT, AgentKit, Sora 2
OpenAI announced Apps in ChatGPT (buildable via a preview Apps SDK), AgentKit for production‑grade agents, and Sora 2 video generation in the API, alongside model updates like GPT‑5 Pro and low‑cost realtime/image minis. Read the DevDay page.
Why it matters: turnkey distribution inside ChatGPT + first‑party agent tooling reduces glue‑code and speeds time‑to‑value for internal tools and consumer workflows.
Google DeepMind launches Gemini 2.5 Computer Use
DeepMind released a fine‑tuned Computer Use version of Gemini 2.5 Pro that can navigate UIs—scroll, click, fill forms—and showed leading results on interface-control benchmarks. Available via the Gemini API/Vertex AI and demoed on Browserbase. Google blog · VentureBeat overview.
Why it matters: this moves agents from API calls to doing work across real websites and web apps—useful for QA, ops automations, and internal admin panels.
Meta will personalize recommendations using your AI interactions
Starting December 16, Meta says engagement with its AI features will inform ads and content recommendations across its apps, with new controls to review or disconnect those signals. Meta newsroom.
Why it matters: marketers get richer signals; privacy teams should review settings and update consent language before the change takes effect.
⚡ Quick Hits
IBM × Anthropic: Claude integrates into IBM’s new AI‑first IDE; early internal users report ~45% productivity gains; IBM publishes an MCP‑based Agent Development Lifecycle guide. IBM release.
Slack expands AI platform with context‑aware apps, a real‑time search API, and an MCP server so agents can safely use Slack data. Salesforce newsroom.
Figma app in ChatGPT turns conversations into FigJam diagrams (flow charts, Gantt, more). Figma blog.
AWS Bedrock ships an open‑source MCP server for AgentCore and introduces cross‑region inference routing (with Claude Sonnet 4.5 support). AWS news · Bedrock ML blog.
Microsoft taps Harvard Med for Copilot health queries to improve accuracy and reduce hallucinations in medical searches. Reuters.
Cisco unveils a chip to link AI data centers across long distances with lower power and higher bandwidth. Reuters.
Google AI Plus expands to 36 more countries; Google also broadened access to Opal, its no‑code AI mini‑app builder. Google AI roundup.
How Canva, Perplexity and Notion turn feedback chaos into actionable customer intelligence
Support tickets, reviews, and survey responses pile up faster than you can read.
Enterpret unifies all feedback, auto-tags themes, and ties insights to revenue, CSAT, and NPS, helping product teams find high-impact opportunities.
→ Canva: created VoC dashboards that aligned all teams on top issues.
→ Perplexity: set up an AI agent that caught revenue‑impacting issues, cutting diagnosis time by hours.
→ Notion: generated monthly user insights reports 70% faster.
Stop manually tagging feedback in spreadsheets. Keep all customer interactions in one hub and turn them into clear priorities that drive roadmap, retention, and revenue.
🧠 Expert Prompt: Ship a ChatGPT App in One Afternoon
Use the new Apps in ChatGPT to deliver a simple, branded tool to your team or audience.
"You are an App Builder. Create a ChatGPT App called that:
- Authenticates with Google (Sheets + Drive) and Slack.
- Takes a CSV upload, validates columns, and posts a summary to Slack.
- Produces a Google Sheet with cleaned data + a QA checklist tab.
- Exposes a single 'Run Clean' button.
Deliver: required App manifest (permissions + OAuth scopes), UI copy, and step-by-step install docs for my org."
Follow‑up prompts:
1) "Generate the Apps SDK boilerplate with my manifest and file handlers."
2) "Package a quickstart README for non‑technical users."
3) "Add a log panel and export logs to a new Sheet."Built on OpenAI’s DevDay features. See DevDay page.
⚙️ Productivity Gem: Context‑Aware Agents in Slack
Pair Slack’s new real‑time search API + MCP server with your agent to answer questions grounded in channel history and files.
Stand up an MCP server exposing search_messages, fetch_file, list_channels.
Grant least‑privilege OAuth scopes (read‑only to start).
Route agent replies back into the originating thread with citations.
Docs: Salesforce newsroom.
🏥 AI-Enabled Health Tip: Turn your wearable data into a weekly plan
Use an LLM (ChatGPT or Gemini) as a lightweight coach that adapts to your Apple Health / Google Fit trends.
Export (iPhone) — Open Shortcuts ► search/create Get Health Samples ► add actions to pull HRV, Resting HR, VO₂max, Sleep, Steps ► add Create Note or Save File (Files/Drive) ► run to save your last 7‑day summary.
Coach — Paste the summary into your AI with the prompt below (attach a screenshot if easier). The model returns a 7‑day plan tuned to your numbers.
Automate Sundays — In Shortcuts ▸ Automation create a Sunday trigger ► actions: Open App (ChatGPT/Gemini) + Paste Text (your saved prompt). Or create a recurring Calendar event titled “Paste Health Summary to ChatGPT/Gemini.”
Android note: Use Health Connect to export the same stats, then save to Drive/Notes; schedule a weekly reminder with Google Calendar (or a routine app) to paste into your AI coach.
"You are my endurance + strength coach. Using the attached 7‑day summary (RHR, HRV, VO₂max, sleep hours, steps), create a 7‑day plan with:
- Zone‑2 targets (minutes + HR range),
- 2 strength sessions (compound lifts, sets/reps),
- 10‑min daily mobility,
- Recovery guidance.
Flag overtraining if HRV ↓ ≥15% vs baseline or sleep < 6.5 h; reduce volume by 20% and add an extra rest day if flagged."Re‑run weekly; the plan adapts as HRV/VO₂max change. Keep the export to <1 KB by storing a compact table in Notes.
🧒 AI for Kids: Build a mini app with Google Opal
Have your kid design a "+Homework Helper" in Opal (Google’s no‑code AI mini‑app builder). Scope: snap a photo of a worksheet, generate hints (not answers), and create 5 practice questions. Ship it, then demo to family.
Getting started: see Google’s September AI roundup for the Opal access update and new education tools. Google AI roundup.
That’s the signal for this week. If you try the ChatGPT Apps prompt or a Gemini Computer Use workflow, hit reply and tell us what shipped—we’ll feature the best ones.
— PromptHacker