August 12, 2026
TechCrunch reported that Amazon will start training AI models on Twitch streamers' content by default, requiring creators to actively opt out if they don't want their broadcasts used. What makes this newsworthy isn't the policy itself -- plenty of platforms have quietly done the same -- it's that Twitch's own Chief Product Officer, Mike Minton, said the honest reason out loud on a livestream: if this were opt-in, nobody would choose in. That's not a talking point, that's an admission that the product has no consent-based market at all. When a company tells you the truth about why it's taking the coercive option, believe the truth and ignore the framing around it.
For any business that publishes user-generated content, livestreams, support tickets, or customer chats, this is a preview of a fight coming to your own terms of service. The default-on, opt-out model works because most people never touch a settings menu -- that's the entire business case for it. If you're building products on top of a platform that trains on your users' data by default, it's worth reading your own data handling commitments against something like a proper trust center before a customer asks you the same question Twitch streamers are now asking Amazon.
TechCrunch also reported that Cognition, the AI coding startup behind Devin, is reportedly in talks to raise at a $40 billion valuation -- months after closing $1 billion at $26 billion. That's not a growth curve, that's a parabola. Meanwhile Lovable confirmed a new $13.3 billion valuation on another $400 million raise, on the back of hitting $500 million in annualized revenue, and Blacksmith's valuation jumped almost tenfold in under a year as its revenue grew more than tenfold too. Put those three together and you get a real signal buried in the froth: AI coding tools are one of the only categories in this boom where revenue is actually keeping pace with valuation multiples, not just narrative.
That doesn't mean the prices make sense. A $40 billion mark for Cognition would put it in rarefied company built almost entirely on the bet that AI-written code becomes the default way software gets made, not a novelty for prototypes. I think that bet is directionally right -- the shift toward AI-assisted and AI-native building is real and accelerating -- but the specific multiples being paid right now assume near-flawless execution and no serious competitive disruption for years. Anyone who has watched a hot category before knows that's a lot to assume. For businesses evaluating whether to build in-house or lean on these tools, the more durable question isn't which startup wins the valuation race, it's whether the app you end up with can actually run your operations, not just impress a demo audience -- which is exactly the tradeoff we lay out in Buy vs. Build vs. ViibeStack.
The most useful headline of the day might be the least flashy: TechCrunch covered a debate at Ai4 where Geoffrey Hinton, Fei-Fei Li, and Andrew Ng argued over regulation, open source, and how the U.S. should compete as China advances its own AI capabilities. What's notable is who's on this stage. These aren't fringe accelerationists or doomers -- they're three of the most credentialed people in the field, and they still don't agree on whether openness makes AI safer or more dangerous at scale. That disagreement, at that level, should tell you the industry hasn't settled anything, no matter how confidently any single vendor talks about its own safety posture.
For a business leader, the practical takeaway is to stop waiting for consensus before making decisions about which AI tools and vendors to trust. Consensus isn't coming soon. What you can control is how much visibility you have into your own stack -- which is why vendor security posture and incident response planning matter more than whatever any one expert says on a panel.
Which of today's stories worries you more as a business decision-maker: platforms defaulting to using your data, or valuations in AI coding running well ahead of any company's ability to prove it out?
Sources