ChatGPT Alerts are here... what does it mean for Agentic AI in 2025?

You're receiving this email because you registered for one of our workshops. You can unsubscribe at the bottom of each email at any time.

Hey there! Welcome to 2025 🙂 

After a holiday hiatus, we’re back to posting about exciting updates in the world of AI and MindStudio.

This release is special—and I’m about to spoil part of what our team has been working on over the past few months. It’s going to change how you use MindStudio forever. MindStudio Architect (our workflow generator) was just the first step toward making MindStudio ready for the agentic era and mass adoption.

In other news, OpenAI ended their 12d of releases with a complete shift to the consumer market with ChatGPT in WhatsApp, via phone calls, and several releases that only work within ChatGPT. Mistral released Codestral, supposedly better than all other coding models its size. NVIDIA announced the cheapest computer able to run big models locally. Microsoft & Google both entered the consumer market through their workspace solutions.

Continue reading to learn more!

As a reminder, we’re now welcoming partners who want to build AIs for their clients. Sign up for extra support, training resources, and more here.

🗞️ Industry news

ChatGPT Alerts, and what it means for OpenAI

After the disappointing end of the OpenAI days, the company released ChatGPT scheduled tasks, enabling ChatGPT flows to run asynchronously and return the result at specific times.

Basically, this is a scheduled task that can use AI to perform actions. From my testing, it seems able to search the web and uses GPT-4o as default model, but there are very little official details about what’s going on in the background.

Users on X and Reddit go from very excited to sorely disappointed all they can show for their AGI ambition is a glorified to-do app. I’m somewhat in the middle:

• ChatGPT Tasks is rarely useful at this point in time. Some use cases people are showcasing are not only uninspiring, but frequently don’t even need AI at all. Do you really need AI to remind you to work out or eat breakfast? Is it just another todo app you’re going to ignore?

• However, it shows where OpenAI is going. Some even spotted they bought a domain similar to “Operator”. The point of ChatGPT Tasks is not what they are now, but what they show us about the future of ChatGPT. And what they show is an agentic future where ChatGPT stops being a chatbot and starts executing actions on your behalf throughout the day, without any human input

• It’s worth noting Scheduled tasks are only available to plus and pro subscribers. If you’re on a plus or pro account, choose the new model from the dropdown and ask it to schedule a new task for you. You can review all active tasks in your profile dashboard.

This is another step into mass adoption for OpenAI after the announcement of o3, the most intelligent model yet. While they move toward AGI with a model that resembles human intelligence, they need a way to keep the ship afloat. ChatGPT is their revenue machine, rather than the models, which is why they’re focusing so heavily on making it useful for…well, everyone.

But they’re not alone.

Other minor recent releases include being able to call ChatGPT, message it on WhatsApp, and use the desktop app to interface with more applications on your screen.

NVIDIA announces Project DIGITS, the mini computer for AI

Project DIGITS, powered by the new GB10 Grace Blackwell Superchip, delivers a petaflop of AI performance in a desktop system, letting researchers, data scientists, hobbyists, and students prototype, fine-tune, and run massive AI models locally.

The system runs on standard power, includes 128GB of unified memory, and can handle up to 200B-parameter models. Two units can even be linked to run 405B-parameter models, which means you can run a GPT-4o or Claude 3.5 Sonnet model locally.

Why is it interesting? Because price starts at $3,000, shipping in May.

What seems like a lot for personal computing is actually the best deal for the average consumer to run these kind of models. For the price of a specced MacBook Pro you will be able to buy a machine able to run the smartest computer programs in human history.

And, like most things tech, the price will continue to go down. In a couple of years, we might be running models like GPT-4o in machines that cost less than $1,000 - something that most companies will be able to spend for each employee.

It comes shortly after the release of Jetson Orin, a tiny computer, think a big Raspberry Pi, specifically designed to run AI models at home. It costs only $249 and should be able to run most small models (1b to 8b parameters). Perfect for AI enthusiasts that want to play with models locally but can’t afford more expensive machines!

Biden Administration proposes export restrictions on AI chips

The Biden administration is proposing new AI chip export restrictions, aiming to limit China’s access while maintaining U.S. leadership in AI. The plan impacts 120 countries, including Mexico, Israel, and Switzerland, restricting their ability to acquire advanced AI chips for data centers. However, key allies like the UK, Japan, and Germany remain exempt.

Industry leaders, including NVIDIA and the Semiconductor Industry Association, are pushing back, warning that these rules could fragment supply chains and hurt U.S. competitiveness. Critics also argue the restrictions could limit chips already used in gaming and consumer tech.

The plan includes a 120-day review period, meaning the incoming Trump administration will decide whether to enforce or revise it.

AI will definitely be key to the future of superpowers like the United States, but here’s what we now for now:

• Chinese companies released DeepSeek v3 for pennies on the dollar, and the model is outperforming industry leaders like GPT-4o for some use cases. The model is partially open source and is starting to get traction

• Other Chinese companies are releasing video models that are as good, if not better, than the existing options in the US

• NVIDIA will strongly oppose the legislation, for obvious reasons. If this doesn’t pass, they’re likely to become the most valuable company in the world in the next few months

• There’s nothing much going on in the EU other than Mistral, which is lagging behind in terms of adoption and innovation.

🔥 Product Updates

Starting next week, MindStudio is going to add integrations with services like Google Sheet, Google Drive, Google Docs, Google News, Google Trends, Twitter/X, LinkedIn, YouTube, Email Finders, Enrichment Providers, and more.

This is the start of something bigger, and it’s something we’ve been working on for the past months: MindStudio as your one-stop for Agentic AI workflows.

In order to let you build useful, 360° business workflows, we couldn’t stop at AI models, fine tuning, and RAG. That’s cool, but we got so many of you asking us… now what?

How do I create a Google Doc once the article is ready?

How do I push data from LinkedIn into the AI to tune it?

How do I research live data from YouTube to send a summary on Slack?

Now, we have an answer.

Now, we have integrations 🎉

From next week, you’ll get access to:

• Post to X

• Search X

• Create Google Doc

• Create Google Sheet

• Fetch Google Doc

• Fetch Google Sheet

• Search Google News

• Search Google Trends

• Post to Slack (already available)

• Call MindStudio workers from Slack (already available)

But… this is just a tiny fraction of what’s to come. Our entire team has 1 priority right now: giving you what you need to succeed with more integrations and practical ways to get started building.

This is why you will see new integrations come up almost daily, and a full library of use cases, templates, and videos along with them.

We already published 10+ videos on YouTube to show how our internal team is building agents, and there are many, many more to come. We’re all building tools you’ll be able to remix and use in a dedicated library on the website.

Finally, this workers can’t live forever within a MindStudio account - right?

We’re working on something big. Something we’re sure all of you will be excited to try. We’re not ready to announce it yet, but it’s coming very soon. Stay tuned, it’s worth it!

💡 Tip of The Week

You can now generate code for custom functions and asset generation with AI. To do that, simply try to create a new function and click on “Generate” in the bottom left corner.

But… what do you ask it to do? After all, if you’re not a coder, you might not even know where to start.

Here’s a tip: use an aggregator like Apify or SerpAPI, and learn the ropes just ONE time for many integrations at once.

Learning how to sync to Google Docs is very hard (and you won’t have to do it on your own very soon), but calling one API is easy.

So, familiarize yourself with a platform like Apify, find the scrapers you need (for example, a YouTube scraper), and ask the AI to handle it. Something like:

“I want to scrape YouTube videos. This is the API endpoint: [add it here], this is the JSON body it’s expecting [add it here]. These are the variables that should be in the configuration file: [add them here]”.

That’s it. All you need is the endpoint, the JSON body to send, and the variables you want the user to type in, like their API key and some dynamic values.

While it requires a bit of technical knowledge, it’s much better than asking the AI to “figure it out”.

This also works in the MindStudio architect when generating whole workflows. Make sure to always specify everything I mentioned above, and/or paste in the whole documentation article from the provider on how to use their API.

🌯 That’s a wrap!

Stay tuned to learn more about what’s next and get tips & tricks for your MindStudio build.

You saw it here first,

Giorgio Barilla
MindStudio Developer & Project Manager @ MindStudio