OpenAI has been busy lately – in the past week they've dramatically expanded Sora, their AI video model + sharing platform, and introduced Agent Builder, a new way to visually connect a range of tools to create an AI workflow.
We have detailed lessons on these coming up for AI Consultancy Project students, and as I prep for that, I thought I'd share my initial reactions with you as well.
Before we dive in, remember that we're opening enrollment for the next cohort of AI Consultancy Project students later this month.
Alright – let's look at OpenAI's newest features and tools...
Sora: OpenAI creates its own creepy TikTok
If you've been in my orbit for more than a few minutes, you probably know that I am not the biggest fan of TikTok (or Instagram Reels or YouTube Shorts). Even the fully-human-created versions of the "infinite scrolling video feed" apps are basically just built to rot your brain.
Which is why it's kind of odd that OpenAI – a company whose goal is to create higher levels of intelligence – would foray into what I'd describe as the lowest level of internet engagement.
The company released its excellent new Sora 2 Video Model alongside a mobile app that lets you easily create short videos from text prompts (as well as some other fun features, like adding internet celebrities' likenesses to the videos, with their permission).
Like Google's Veo 3, the new Sora 2 video model creates hyper-realistic videos and is a huge step forward in this branch of AI-powered tech. It will eventually transform how all filmmakers create video, making custom animation and what we used to call "CGI" dramatically less expensive and more accessible to everyone.
In contrast to the impressive new video model itself, the mobile app is basically a creepy, weird version of TikTok / Shorts / Reels.
It works really well as a promotional tool, since it allows me to easily share videos that I generate with my (very small number of) Sora "friends." It also clearly has incentivized people with existing Influencer businesses to start posting on the new app, similarly to how Meta's Threads picked up a lot of initial traffic because people who make money by "influencing" are always eager to get in the ground floor of a new platform.
That being said, I think it is clear that early reports of OpenAI building a social network (April 2025) were off-base – and in fact, the Sora mobile app is mostly a cheeky way to promote the video model, rather than anything that will stand alone or become a serious social-networking business. It's a cheap (vibe coded?) knock-off of Instagram/TikTok, and there appears to be minimal value or staying power beyond the few minutes of "haha, look at this one" you get during your initial session.
As a launch stunt or gimmick, the Sora mobile app is fun. But I think we're seeing two clear trends:
Social networking apps are the tech of the past; it makes no sense for an innovative AI-first company to go down this road as a business model. Time spent on social media is now trending downward.
There are few if any consumer use cases for AI video models, which is why the only way to showcase them is via silly memes. These are mostly tools that will be used by professionals to speed up existing filmmaking jobs. Great for consultants and businesses, mostly just a throwaway toy for the average person.
A final thought on AI video – my wife and son are already bored of it and roll their eyes when I show them all the weird stuff I created. (Fortunately, playing with AI and making my family roll their eyes are both important parts of my job.) This is a significant departure from the prediction that realistic fake video would take over the world and confuse everyone... instead, people are already saying, "Oh, that video is just AI? Boring."
Agent Builder: Visually Connect ChatGPT to More Outside Tools
OpenAI's Agent Builder Interface
This was supposed to be the year of "Agents," and while a lot of progress has been made, I think most AI enthusiasts would say that we haven't seen anything truly groundbreaking in the world of autonomous AI agents yet. (However, I do use ChatGPT and Claude as research agents extensively, enough to make it worth paying $200/mo for the super-duper versions of each.)
Most of the progress for businesses has been coming in the form of thoughtfully-designed custom workflows, which have "Agentic" properties, but these are being built as bespoke projects by consultants (like the people in our program) rather than being distributed in a "one-click, easy-install" format like what you'd get with old-school software. In other words, there is no Shopify or Hubspot analog for AI agents yet.
Agent Builder seeks to make things a little bit easier by integrating many of the features of our favorite automation tools (Zapier, Make and n8n) directly into the OpenAI platform. Like many new releases, it's a bit buggy and imperfect (see this video where the YouTuber actually just ends up showing us that the thing he wanted to build doesn't work yet).
However, it shows that OpenAI has noticed that the existing automation/orchestration tools – which we teach extensively inside the AI Consultancy Project – have become extraordinarily popular and successful because of AI. I've mentioned before that Zapier was basically a boring, B-rate tech company in 2022, and that the sudden ability to integrate AI into Zapier automations has reinvigorated what was once a sleepy, basic business. (Notable, Zapier seems very threatened by Agent Builder.)
One of the large, unanswered questions for businesses is how the pricing, privacy and data ownership will work across new automation tools in the long term. If OpenAI can provide high-quality systems that move fast and satisfy serious data privacy requirements, they can easily compete with Zapier, Make and n8n among folks who are building complex automations.
Then again... Google and Amazon could easily become heavyweights in this category too, if they see that there's a clear business opportunity there.