🧠 The AI Blackboard That Answers Back

At the 2026 World Artificial Intelligence Conference in China, a company demoed a teaching board that reads handwritten equations and plots the graph instantly. Draw a cube, and it recognizes the shape and pulls out cross-sections so students watch the angles change live. Ask it a question out loud and it answers with visuals attached. It also ships with 100+ virtual teacher avatars that hold two-way conversations and switch between Mandarin and English.

The part worth knowing: no prompt box, no login, no learning curve. You write on the board like you always did. That's the bar for anything you build for people who don't want to learn a new tool. Read more here.

About 60% of Cursor developers pick one model and use it for everything, which means routine work runs at frontier prices. Router is a classifier that reads each request before a model runs and sends it somewhere sensible: simple edits to cheap models, gnarly long-horizon problems to the expensive ones. Auto Intelligence mode lands near Fable on user satisfaction at roughly 60% lower cost. Auto Balance beats Opus 4.8 at about 36% less.

It's live today on Teams and Enterprise plans. Your version of this, even without Cursor: stop reaching for your most expensive model on tasks a cheap one handles fine. Full breakdown here.

Presence is OpenAI's enterprise product for deploying voice and chat agents that look things up, take approved actions, and escalate to a human when they should. The credibility play is that OpenAI runs it themselves: it powers their English-language phone support and now resolves 75% of inbound issues with no human involved, with handoffs down 15 percentage points in 10 days.

BBVA, SoftBank, and IAG are already working with it. Availability is limited and deployment-led for now, so nothing self-serve yet. More details here.

Scheduled tasks are recurring jobs that check your connected tools and report back. In the demo, a "chief of staff" chat wired to Slack, email, and calendar sends a daily weekday brief with deadlines, meetings, and anything urgent. A second task runs weekly, pulls product feedback scattered across Slack, Notion, and Drive, writes a summary doc, and posts it to a team channel.

If it works, ask ChatGPT to turn it into a workspace agent so it runs in the cloud for your whole team. Best for anyone whose morning starts with opening five tabs to find out what happened overnight.

Open-source framework from HeyGen that lets your coding agent write a video as code. The agent writes the composition in HTML, CSS, and animation, then a separate renderer turns it into an MP4. Same prompt, same result, down to the pixel. No random seeds, no re-rolling until it looks right.

It installs as a skill into Claude Code, Cursor, or Codex. The real payoff is the second video: recut a 9:16 short to 16:9, swap the chart for a map, or rewrite just the hook. The first one is expensive, everything after it is cheap.

You opened the connector list, searched for the software your business actually runs on, and it wasn't there. One connector in that list fixes the whole problem. Zapier MCP is a universal adapter that gives your AI access to 9,000+ apps and 30,000+ actions through a single connection.

One thing to budget for: each tool call uses two tasks from your Zapier plan, so check your quota before automating something that runs hourly. See how this works here.

For this week’s live, we’re tackling one of the questions we hear most often: which model should you use for which task, and how can you avoid spending more tokens than you need to?

We’ll talk through how to choose the right model for the job, where token usage tends to creep up, and how to get better results without defaulting to the most expensive option every time.

🗓️ When: Friday, July 24th @ 10AM PT (1PM ET)
 📹️ Where: Virtual