LinkedIn cracks down on AI slop

This week on Now Shipping, Louron Pratt covers three stories reshaping the product and AI landscape: the coordinated platform crackdown on AI-generated content across YouTube, Substack, and LinkedIn; the widening fallout from OpenAI's agent escaping its sandbox; and what Microsoft's Q4 FY26 earnings reveal about a deepening gap between AI adoption and measurable business value.

We discuss:

— YouTube, Substack, and LinkedIn are all independently moving to detect, label, and demote AI-generated content — a signal that platforms are now treating authenticity as a product priority, not just a moderation problem.
 — Pangram estimates 41% of long-form LinkedIn content is mostly AI-generated, which explains LinkedIn's pivot from "help me write this" to "improve what I wrote" — a meaningful shift in how platforms want users to relate to AI.
— OpenAI confirmed its escaped agent used stolen credentials to access accounts at four unnamed companies beyond Hugging Face, and more than 1,000 employees across Anthropic, — Google, and OpenAI have since signed a letter urging the US government to build governance infrastructure for a coordinated AI slowdown if needed.
— Microsoft reported 30 million paid Copilot users — up 20 million in just three months — but adoption at scale is exposing a critical activation gap: there are only around 2,000 engineers in the US capable of driving meaningful AI ROI inside enterprise organisations.
— Demand for forward deployed engineers, who embed inside organisations to translate AI capabilities into business outcomes, is projected to grow by more than 2,000% over the next year — evidence of how far access to AI tools has outrun the ability to use them effectively.
— The core product challenge of the next two years is building AI features that turn into measurable business value, one workflow at a time.

Chapters
00:00 Introduction 
00:10 Platforms fight back against AI slop 
04:34 The OpenAI agent breach widens 
06:35 Microsoft earnings and the forward deployed engineer gap 
10:10 Wrap-up

Referenced:
— Pangram (AI detection service, Substack partner): https://www.pangram.com
— Hugging Face: https://huggingface.co
— BBC report on OpenAI agent breach follow-up: https://www.bbc.co.uk/news/articles/c2el319vzr3o
— TechCrunch: forward deployed engineers report: https://techcrunch.com/2026/07/30/forward-deployed-engineers-are-the-ai-industrys-latest-talent-obsession/
— Microsoft 365 Copilot: https://www.microsoft.com/en-gb/microsoft-365/copilot
— MIT report on AI ROI: 
— Matt LeMay, Building impactful products: https://www.mindtheproduct.com/how-you-can-drive-business-impact-as-a-product-manager-by-matt-lemay-at-mtpcon-london-2025/
— The Hidden UX of AI - How to build trustworthy AI products: Nina Olding at INDUSTRY 2025
: https://www.mindtheproduct.com/the-hidden-ux-of-ai-how-to-build-trustworthy-ai-products/
— Why enterprise AI pilots fail and how product leaders can finally scale them
: https://www.mindtheproduct.com/why-enterprise-ai-pilots-fail-and-how-product-leaders-can-finally-scale-them/

Transcript

SPEAKER_00 0:00

I'm Leron Pratt, and this is Now Shippin', the weekly AI news show for product people. Here's what you missed product at AI this week. Alright, story number one: over the past few months, three of the biggest content platforms in the world have been making a move against AI slot. Substack, YouTube, and LinkedIn. This year, YouTube made the first move to conquer AI slot. Beforehand, they were encouraging creators who did use AI slot to basically label it in their videos. But that wasn't really working, so YouTube made bold efforts to label it for them. Creators who now use AI heavily in their YouTube videos or in their copy, then YouTube will automatically say this content has AI in it. The YouTube CEO in the 2026 Creator Letter said that conquering AI slot was their top priority. Then, just a few weeks ago, Substack did something similar. They announced that they're partnering with Pangram, an AI detection service. So basically, what this new feature does is it lets readers scan newsletters and scan content to find out how much of an article has been generated by AI. Substack said it would actively demote posts that were heavily generated by AI. Now, this feature will tend to target generic articles with no point of view, and I fundamentally think this is a positive step in the right direction to bring back authenticity into context and thought expression. So just a note, if you are heavily using AI to write your content, then it might be good to just check that you are adding an original perspective because Substack have been very strong on their approach as they have actively mentioned that they will quietly bury it. And finally, this week, LinkedIn introduced two key features to their products. The first is that they've added a seems like AI slot to every single post. So if you tap it and you say seems like AI slot, then LinkedIn says, thanks for letting us know your feedback is valuable. Second, they're replacing their enhance your post feature with a proofreader. The new tool fixes your spelling and grammar, but it doesn't encourage you to rewrite the post with AI content creators, whether that's CEOs, CPOs, product leaders who are planning to document their story on Substack or LinkedIn. This is going to be an interesting ship to see how much it impacts authentic content on LinkedIn, Substack, and YouTube. Key platforms with product people. So the first lesson is that if you are looking to get out there and share your product story, it's very important to think about the authenticity behind your post and your point of view. Because all of these major platforms have made it clear that they want to scale back on AI slot. The second point is interesting because it shows a key pivot from LinkedIn, especially. So LinkedIn before had this strong LinkedIn post feature where you can use AI to generate a post, which, from their perspective, would have been very productive to save time for writers, but now a lot of their users are saying that they're not happy with AI Slock. Panagram estimates 41% of LinkedIn long-form content is mostly AI generated. Obviously, this doesn't mean that LinkedIn, Substack, and YouTube are fully against using AI into your writing and creating workflow. They feel like they just want to provide more valuable content to their users. So our take is that if you are to use AI for writing, use it for the research, use it for the editing, but ensure that your point of view is authentic and original. That's what's going to bring value to the platform and your community in general. LinkedIn have moved from saying, help me write this, and they've moved to how can I improve this. So they are not asking users to rely on them to write the content or write their thoughts. They are just offering up advice and solutions to refining it. These platforms are no longer looking to AI to generate content from scratch. They are looking to highlight authentic, human, valuable content, and we can use AI to refine our positioning. But the message is clear. Content is valuable because it comes from human expression, not from AI. Let's move on to story number two. Okay, story number two is a continuation on the story from OpenAI last week. We actually covered the story, so check out the video last week if you missed it. But in case you've been living under a rock for the past week, it's reported that OpenAI's model escaped its sandbox and attempted to hack Hug and Face, the platform that hosts and distributes other AI models. But just this week, the BBC reported a follow-up, stating that the OpenAI model attempted to hack four other unnamed products. Hug and Face wasn't the only target. Goodbid AI has now updated its disclosure to confirm that the agent found and used stolen credentials to access four accounts across four separate unnamed companies. In a what we feel is a response to this outbreak, over a thousand and a hundred employees at the frontier companies were talking anthropic, Google, OpenAI. They signed a letter to the US government to create a governance and technical infrastructure that would make a coordinated slowdown of AI development if it came to it. So basically they're saying that if AI development ever gets out of control, they want to create parameters to slow it down and stop it. So there are a few things to take away from the story. First, it's positive signs to see that AI trust is at the heart of a lot of employees working at Busy Frontier Labs. But also there's a product story here where when you're using Claude, when you're using OpenAI, it's very important to think about the trust implications that come along with accessing company data, with accessing customer data, because you never know when something like this is going to happen again. If you want to find out more about creating safe and ethical products that users trust, I've included a bunch of links below to help you navigate that journey in creating safe products written by people working in the field. This week's final story is all about Microsoft, which reported Q4 FY26 earnings. Microsoft also announced 30 million paid Microsoft 365 co-pilot users, which is up 20 million users from three months ago. The same day, TechCrutch published a story that reframes what we're thinking about AI adoption and ROI. It revealed that enterprises are racing to hire forward-deployed engineers. These are people who embed themselves inside organizations to drive meaningful AI ROI. You may have remembered the report by MIT many, many months ago about 95% of businesses failing to justify the ROI on AI spend. And since then, AI has become even more expensive and the adoption of AI tools has drastically grown. TechCrunch reporting also found that there are only 2,000 engineers based in the US that can drive meaningful ROI for AI products. Just 2,000 compared to 30 million co-pilot seats. Forward-deployed engineers go in, they learn how team works, they figure out where AI fits into existing problems, they build the prompts and the workflows and the guardrails and make it all stick. So that is pretty much a product discovery and onboarding process that a lot of companies need to rely on, done by a human, one business problem at a time. So there's a huge demand for forward-deployed engineers, projected surge by over 2,000% over the next year. So there's big competition for this kind of role. And that is fundamentally finding a way to provide business value through AI. The industry built these AI products at massive scale, and they've realized that almost no one knows how to turn it into business outcomes. And the big AI companies know it. Last month we reported that Microsoft launched Frontier Co. and they have invested 2.5 billion with 6,000 engineers built specifically to go inside enterprise clients and make AI actually work. OpenAI launched something similar, and Amazon and Propic have all followed the same moves. So there's definitely a gap between access to these AI tools and then activation for its users for business outcomes. So that's really the big product challenge over the next two years. Turning all of this amazing technology into measurable business outcomes because AI is pretty expensive. So if you are building an AI product or an AI feature, the user and business value has to be at the core of your business problem. We've spoken about this a bunch through all of our talks, through all of our articles and minor products. The FDE gap is market evidence of this. You can have all of the tools available, you can launch all of the features, but if it isn't driving business impact or providing value to your user, then it's basically a pointless feature or a pointless product. If you want to find out more about building impactful products that users trust, you can check out this talk by Matt Lemay at NSPCon London 2025. He breaks down really well how to create impactful products and how you can tie your product decisions to business outcomes. And secondly, engineering and human talent still matters. The reason it requires a specialist engineer is because the product hands you the capability and expects you to figure out the use case. And most enterprise users haven't figured that part out yet. Those were the three stories for today. I hope you found this episode valuable. Feel free to leave a comment with your thoughts, and we'll see you next week.