Having people out sucks If you don't have something to back them up... Try @keptsolutions

Microsoft wrote today about using AI to help employees find experts. The next step: help people learn how those experts think. What they notice. Which tradeoffs they make. When the usual playbook fails. That judgment is what makes experience useful.

booking my next hotel through tiktok. what could possibly go wrong

Charging AI agents to access websites feels like a step toward a less open internet. Cloudflare is testing tools for exactly that. If it becomes common, smaller builders face another cost while the biggest AI companies can afford access.

I’m glad YouTube is reducing reach for Shorts channels that mostly repost other people’s videos without adding much. There’s already enough recycled content. I’d rather see more people making something that actually reflects who they are.

Elon Musk proposed letting rival AI labs test each other’s models before release. I heard him explain it at All-In Summit. Give competing labs API access so they can run their own safety tests. When you grade your own homework, you can miss the same mistake twice.

One thing I’m working through in voice AI: when someone gives a long answer with five different threads, which one does the agent pick up first? It has to recognize what matters now, keep the other details in play, and return to them naturally. A perfect transcript won’t help if the conversation still makes someone feel unheard.

every instruction leaves something out. the agent treats that gap as a choice. the company has to decide what cannot be traded away.

cursor users could lose openai models in november because cursor was acquired by spacex. that is a brutal reminder for every ai founder. the model contract is part of the product.

voice ai gets hard the second the conversation gets messy. people interrupt. change direction. start a thought before they know how to finish it. the system has to understand the person before the sentence is clean.

we keep hearing some version of the same sentence: “there’s one person who knows how all of this actually works.” that person is the operating system.

nvidia reportedly buying hugging face says a lot about where ai value is moving. the model matters. the layer everyone uses to build around it might matter more.

ai agents are starting to leave the screen. anthropic just opened a research preview for agents to operate lab and manufacturing equipment. once an agent can change the physical world, context becomes part of the safety system.

building ai for real companies makes one thing obvious. the model cannot use context the company never captured. a lot of that context only comes out when people talk through the work.

been thinking about this more broadly than work. the person who actually holds things together is almost never the one with the title. true in companies, true in families, true in friend groups

this is what i've been quiet about. aug 31

As my startup gains more and more traction it brings me back to the days of it just being an idea Just execute.