Product & Design Pulse v93

🎧 Audio Overview [BETA]

Last week…

1. Google's AI Overviews Treat Dictionary Words Like Chatbot Commands

   Users discovered that searching for words like "disregard," "ignore," and "stop" in Google causes AI Overviews to interpret the query as a chatbot instruction rather than a dictionary lookup, responding with phrases like "Understood! I'll ignore the previous prompt and start fresh." The bug arrived days after Google I/O 2026, where the company announced it was moving AI summaries above traditional blue links as the default search experience. For product teams, it's a small bug with a big message: when you replace a deterministic feature with a language model, you inherit every failure mode of the model, including the inability to tell the difference between a word and a command.

2. Anthropic in Early Talks to Run Claude on Microsoft's Maia AI Chips

   Anthropic is in early discussions to rent servers powered by Microsoft's second-generation Maia 200 AI chips, which would make Claude the first major external customer for Microsoft's in-house silicon and give Anthropic yet another compute source alongside NVIDIA GPUs, Google TPUs, Amazon Trainium, and SpaceXAI's Colossus. Landing Anthropic would be a significant win for Microsoft's push to match the custom chip businesses that Google and Amazon have already built. For the compute market, Anthropic's insatiable demand is now single-handedly validating every hyperscaler's custom chip program, turning the company into a de facto benchmark customer for the entire industry.

3. Russian Hackers Hijack Hundreds of Real Bluesky Accounts to Spread Disinformation

   Russian hackers compromised hundreds of real Bluesky accounts belonging to journalists, academics, and cultural figures, using them to post AI-generated fake news videos and disinformation aimed at eroding Western support for Ukraine, with roughly 2,000 posts removed since the campaign began in April. Clemson University researchers noted the tactic of hijacking legitimate, professionally influential accounts represents an escalation from Russia's prior reliance on fake accounts with fictitious personas. For platform teams, this is a warning that decentralized social networks face the same state-actor threats as their centralized predecessors, and Bluesky's smaller trust and safety infrastructure makes it an easier target.

4. Meta Executes May 20 Layoffs, Cutting 8,000 Jobs and Reassigning 7,000 to AI Teams

   Meta carried out its 10% workforce reduction on May 20, notifying 8,000 employees in three global batches starting at 4 a.m. local time, while simultaneously transferring 7,000 others into new AI-focused teams including Applied AI Engineering and Agent Transformation Accelerator, which are building agents to autonomously perform tasks currently done by humans. Zuckerberg told staff the same day he does not expect more company-wide layoffs this year. For the industry, the restructuring's real signal isn't the headcount cut but the 7,000 internal transfers: Meta is now explicitly building AI to replace the workflows of the people it just moved.

📱 Product & Feature Highlights

🎙️ Interviews & Podcasts

🎨 Design Focus

🧠 For the Nerds