I attended the Future of Higher Ed AI yesterday evening. The conversations were thought-provoking, and several quotes stuck with me long after the sessions ended.

“Simulations used to be faberge eggs, now they are real eggs.”

What was once rare, expensive, and carefully curated is now accessible and practical because of AI. One of the panelists spoke about a post-partum hemorrhage simulation with a chatbot that tests communication skills, complete with an AI debriefer. Faculty can see transcripts, track performance, and provide targeted feedback.

“I’m scared that education will be taken by Google U or Anthropic U.”

It’s a legitimate fear. As AI becomes more sophisticated, what’s stopping tech giants from creating their own educational institutions? What happens to traditional higher ed when anyone can access world-class AI tutors?

“Those grade reports … don’t feed into a continual improvement loop.”

“Grades do not provide meaningful feedback that lead to learning.”

“When you go to a doctor they don’t tell you pass/fail, they give you an overview of where your health is.”

We’ve built an entire system around pass/fail, A-F grades, but do they actually help students learn? Or do they just sort them into categories? Where is the feedback loop?

Other Highlights

Duet / SNHU showed how they’re using Google Meet transcripts with custom Gems to analyze student context, background, and mindset shifts. They’re connecting with systems like Salesforce to track student performance throughout the term.

Instructors are building their own AI agents since he can’t integrate them into Moodle or Canvas. They are also developing their own content because they can’t find suitable AI content elsewhere.

I built an AI-powered news aggregator using Cursor. It automatically finds and summarizes the latest education technology news every week.

How does it work?

The system runs on a schedule. Every Monday, it:

1. Fetches recent edtech articles from NewsAPI
2. Sends them to an AI model (Deepseek R1) for summarization
3. Formats the summaries into a clean, readable list
4. Updates the website automatically

The technology

NewsAPI - Finds relevant articles about education and edtech Deepseek R1 - An AI model that reads and summarizes the articles in natural language GitHub Actions - Runs the script automatically every week without manual intervention Python - The glue that connects everything together

Expected costs

The system is designed to be cost-effective. NewsAPI offers a free tier that covers basic usage. Deepseek R1 is significantly cheaper than other AI models like Llama 2 70B. Running weekly, the total monthly cost should be under $5. This makes it an affordable way to automate content curation without breaking the bank.

Why it matters

This experiment shows how AI can automate content curation. Instead of manually reading dozens of articles, the AI does the heavy lifting. It finds what’s important and presents it in a digestible format.

The system is fully automated. Once set up, it runs itself. This frees up time to focus on other work while staying informed about the latest trends in education technology.

Cursor really takes over. I’m not complaining, I’m just saying it really gets things done. If you ask it a question, you’d better be ready for it to start committing code.

Info to Implementation

I was having an issue with a news API pulling garbage. My prompt was purely informational. What search parameters does this API offer? I was trying to understand what options there were.

Cursor’s response?

1. First, it immediately gave me a perfect, itemized list of every parameter: qInTitle, domains, sortBy: popularity, etc. Exactly what I asked for.
2. Then, the takeover. It didn’t wait. It instantly transitioned from assistant to developer:
   • Let me update your configuration to get better, more focused results:
   • Now, let me update the Python script to use these new parameters:

It skips the “do you want me to do this?” step. It sees the goal—getting better results—and aggressively starts coding, testing, and debugging on its own:

• The query is too restrictive! Let me adjust it to be more practical:
• Let me try without the domains filter to see if that’s the issue:
• Great! It found 10 articles. Let me check the quality:
• Excellent! Much better results! Let me commit this change...

That is a five-minute debug and optimization cycle completed in about five seconds.

The Risks

This bias for action is why I pay for the tool. It’s fantastic.

But it also makes you manage it differently. You have to be precise. You have to narrow the context and set some limits. “Don’t implement this.” You have to understand that its default state is “go.”

This is the trade-off. I truly appreciate this quality of Cursor, but I also recognize that it can be risky, especially if you include too many commands in the allowlist.

Cursor is less an assistant and more a brilliant, relentless partner.

What I did

Today was a productive day. I worked with Cursor to implement a minimalist theme. It amazes me that it seemingly has an understanding of minimalist design. At one point I tried setting up a Google form instead of a different service, and Cursor commented that it didn’t look good with the minimalist theme.

Changes I made today

• Applied a minimalist theme
• Implemented a form using email.js