Weekly Links: AI Euphoria, AI Dislocation, and Upgrading AI Voice Conversations

This week: OpenAI's first hardware device was not what you expected, and we found an exoplanet with an atmosphere.

On to the main stories:

• Charts of the Week: Software’s Selective Sell-Off. A16Z is not an unbiased source of information on the AI economy, but their regular chart post is interesting. This week covers the software sell-off (it's dramatic) and has several graphs that argue that AI might be helping accelerate junior hiring again. It's not obvious this is cause and effect (AI beneficiary companies are also growing the fastest, so would hire more). The most interesting chart, though, is the one showing that open-weights token used is outpacing closed-weights AI.
• How Tech Workers Actually Feel About AI in 2026. Podcast conversation that deepdives into a 6000 person survery of US tech workers. The survey covers different roles from engineering to product, sales, and marketing, but all at tech companies. This is certainly a group that is experiencing AI in a slightly different way to the rest of the working population (faster adoption, more change, but potentially more agency to deal with that change). The results are not super surprising: some of the experience is euphoric (powered by the ability to do much more), and some is highly anxious (not just due to job risk, but overwork and uncertainty). I'm not sure I agree with the overall narrative that tech is "splitting". My experience is more that many people experience all the positive and negative emotions in cycles. So it's a roller coaster. This one warrants some reflection on the part of companies rolling out AI, though. We are going super fast, and there are structural problems (a slowdown in learning a craft, but also a huge pressure to learn new tools almost hourly) that are long-term problematic.
• State of AI Agents // Databricks 2026. Databricks released a lengthy AI adoption report based on 20,000 survey participants and customers. There are some interesting points, but I would critique two of the headlines on the report. One is "Enterprises are transitioning from single chatbots to multi-agent systems, which grew by 327% in less than four months" - these are almost certainly not really multi-agent workflows; they are single agents coordinating other agents' workflows – single agent toolcalling where the tool is an agent (that's fine as well, but it's not a great industry trend to call everything a multi-agent system just because more than one agent is involved). The second is "Companies that use evaluation tools get nearly 6x more AI projects into production. For those using AI governance, it’s over 12x more." We are all in on evals with spec27.ai, but surely there is some reverse causation here.
• I Got Slopped. AI writing is starting to turn into books landing on Amazon in the form of biographies. The quantity of these books is rising rapidly, and they now at least look credible, but certainly aren't accurate or readable. The story follows a prolific author who seems like an engaging character. The long-term implications of these changes are hard to get your head around, however. Getting a book published is now both mechanically easy (print on demand) and intellectually easy (very little writing required). The books will likely also soon cross into recognizably coherent content. What happens when the number of books multiplies 10 or 100x? They presumably become economically valueless without some curation and filtering mechanism. Welcome to your new AI-powered reading curator.
• ChatGPT’s upgraded voice mode is better at shutting up. OpenAI spent the week taking heat for the changes to it's desktop app with users confused how hidden chat is. Sadly burried by that was the launch of a new voice mode which is much more like a back and forth interuptable conversation that any other AI. If you haven't tried this, you can activate it in the ChatGPT app and it feels much more natural. At first this might feel like a natural upgrade but I suspect it will make a huge difference to general user adoption. The naturalness really changes the nature of interactions. I'm not sure how much more expensive this system is for OpenAI to run but it's a big leap forward. Hopefully OpenAI is able to offload a lot of the voice processing to the local device otherwise they'll be incurring a lot more cost for million of free interactions.

Wishing you a great weekend.