I certainly can’t keep up! I missed the last couple of weeks. I feel like I’m running just to catch myself! To try to catch you up, here’s a review of major events as concisely as I can manage (still long, since so much happened!).

The AI Value Gap

The most important news in the past few weeks is a study on the value of AI from BCG. You may remember the report from MIT that said that only 95% of AI pilot projects fail to make it into production. Although the survey size was small and some of the specific conclusions were overstated, it was directionally correct.

Now there’s more evidence of the large gap in value. Many companies are struggling to get ROI from AI projects, but a few are seeing value, and the value is huge. Boston Consulting Group released a more robust report (1,250 companies) and found much the same thing. 60% of companies lag. 5% of companies are seeing huge value. And most importantly, for the companies that are getting strong returns, their gains are accelerating! Specifically, companies leading in AI adoption are showing 1.7x more revenue growth and 1.6x more EBIT margins than laggards. That means the leaders are getting further ahead, and the laggards are falling further behind.

“This creates a cycle that is virtuous for some and vicious for…companies that started to experiment but struggled to scale…and stagnating companies that have taken little or no action.”
– BCG Build for the Future 2025 Global Study

---

That’s why this is the most important development of the past few weeks. Unless you’re already one of the leaders in adopting AI, you’d better get moving. If you don’t, you risk falling behind your competitors who are, and you may never catch up.

---

But don’t go it alone. Even though your needs aren’t the same as everyone else, t’s better to leverage outside tools than to try to build everything from scratch. “Strategic partnerships succeed twice as often as internal builds.”
– MIT, The GenAI Divide: State of AI in Business 2025

Note: this survey encompassed all kinds of AI, from classical AI to generative AI to the overhyped but still-nascent agentic AI. But the rapid gains measured over the last year are primarily from generative AI, and the companies that have adopted agentic AI fall solidly into the AI leaders.

Targeting AI in the Enterprise

The data we have says that Anthropic currently has the lead in generative AI within the enterprise, mostly with coding. But there’s money to be made in the enterprise, and the other companies are taking aim at businesses for the next step: agentic AI.

OpenAI released apps in ChatGPT, a way of accessing third party apps from directly within ChatGPT. Right now this is consumer-facing, but they said they will soon release this ability within ChatGPT for Enterprise. This is clearly a play to be the go-to interface for doing everything you need at work, powered by OpenAI’s models.

Google announced Gemini Enterprise, yet another platform for building and running AI agents.

“Gemini Enterprise brings the best of Google AI to every employee through an intuitive chat interface that acts as a single front door for AI in the workplace.”

Reading their announcement closely, they seem to be focused on the consumer side (customer service) and companies that target consumers. Despite their “no-code workbench” they only talk about developers building agents, not business users. They claim they’re the only one with a full-stack platform, one that includes:

• Gemini models
• No-code workbench
• Pre-built agents
• Connectors
• Governance
• Open ecosystem

AI Is Taking Jobs

The degree to which it’s really happening and the degree to which companies are using it as an excuse to cut staff isn’t clear, but the trend continues. There are fewer jobs available in certain roles, especially in some markets like consulting and customer service, and especially at the entry level.

Lufthansa announced that it was cutting 4,000 jobs, primarily due to AI. 

Accenture Is doing the same, cutting more than 11,000 jobs in the past three months with layoffs continuing through the end of November.

“We are exiting on a compressed timeline people where reskilling … is not a viable path for the skills we need.”
– Accenture CEO, Julie Sweet, per the Finanical Times

Also per the article, generative AI is hitting consulting and investment firms hard: 

• Roughly 20% of this year’s Harvard Business School class did not have jobs at graduation (they didn’t say what the normal rate is)
• PwC said it will hire one third fewer people over the next few years.

Other Stuff

It’s been less than 3 years since ChatGPT and we’ve reached the point where an amazing new frontier model falls into the…yawn…other stuff category! That’s crazy, but here we are:

Anthropic has long been recognized for having the best model for code. That was challenged by the latest Grok model from x.AI (although it’s doubtful it gained any market share), but no more. Anthropic released Claude 4.5 Sonnet, and is back on top. They claim it is able to operate continuously on coding projects for up to 30 hours (no idea how many human-hours that is).

Amid all these software advances, where are the robots? They’re coming. Figure debuted their newest robot, Figure 03. Robotics is having a moment because of generative AI and the rapid progress here is remarkable. They’re pushing hard for a household humanoid robot, and seem to be making progress faster than Tesla’s Optimus robot. Here’s a 6-minute promo video of what the Figure 03 can do.

Meanwhile, don’t forget the battle raging for an AI-oriented internet browser. Perplexity released their AI-powered browser, Comet. Their goal is to transform the experience of the internet to one that is AI-centic; the launch describes built-in AI that understands, organizes, builds (websites), emails, creates, and (yikes!) shops.

Also, OpenAI introduced parental controls for ChatGPT. It’s a good attempt to protect youth by giving parents the ability to restrict and observe how their kids interact with ChatGPT. But since they don’t own the delivery platform, it’s pretty easy for kids to get around it (all they have to do is get their own account).

And OpenAI (tired of them yet?) announced ChatGPT pulse, where an agent proactively researches information based on your interests and past conversations, and updates you when you log in. Instead of you starting the conversation with a question or request, it starts the conversation with you!

Video, Video, and more Video

Not critical for business, but big news on the AI-generated video front:

Grok launched an upgraded version of their video generation model, Imagine 0.9.

Meta launched Vibes, a video generation model designed to create short-form video for social media feeds. It’s geared towards generating short, engaging video clips ala TikTok. Reception has been two-sided:

• It was immediately called the “infinite slop machine” (if you haven’t heard, slop is the term for AI-generated content that is either low-quality or simply lacking in value…including when it’s from one of your coworkers, then it’s called workslop.)
• It got massive uptake, with social media feeds on Meta/Facebook/Instagram quickly filling up and overflowing with content that is the epitome of eye candy…fun to consume, but no nutritional value.

But most importantly, OpenAI released Sora 2, the update to their video generation model – write a prompt, get a short video of absolutely anything you want. All the usual improvements in video quality, object persistence, etc. Most notable:

• It includes sound (Sora 1 was video only). This makes production of realistic videos much easier.
• It introduced “cameos,” a way to insert yourself into a video. This is clearly aimed straight at content creation for the social media crowd.
• It’s also a social media app. That’s right folks, TikTok, but 100% AI-generated videos. Can anything good come from this?

The Sora app immediately shot up to #1 on the app store. OpenAI claims they’re concerned about all the harm that social media and doomscrolling causes, that could be exacerbated by AI slop. To avoid repeating the sins of social media, their app is supposedly more friends-oriented and not geared to maximize engagement. Do I believe them? Maybe. Maybe not (they don’t have a good track record).

“…our only current plan is to eventually give users the option to pay some amount to generate an extra video…”
– OpenAI, Sora 2 is here

Sound a bit like a drug dealer?

Also, in classic Sam Altman style, OpenAI originally told copyright holders they had to request copyright protection…in other words, if you’re Disney and you don’t want people making videos with Cinderella or the Lion King, send us a note and we’ll block those requests. The copyright holders freaked out (as they should), causing another Sam Altman about-face, so that OpenAI will block everything that’s copyrighted. But by the time OpenAI started blocking, there were already hundreds of thousands if not millions of videos based on such characters. Disneyslop, I guess.

That’s a wrap, just a normal few weeks in the life of AI. See you next week. Hopefully.