Notes from the workbench
Field notes on AI, software, media, real estate, and the machines people build to move their work forward.
Latest from the workbench
I think that AI is becoming good enough to forecast hurricanes 1 year in advance with 90% accuracy, and it can determine severity and the month that it will occur. This will most likely occur in the next 2 years based on my observation of h
Doing work at the AI Frontier has taught me that the next big thing is always coming. Super apps, visual AI, and fast dynamic rendered apps require a 50k/s token throughput. TaLaaS is currently at 12k. #HumanHud.ai
I used astra to build a hubspot clone that allows you to publish to multiple social media channels at once.
Men for a long time have fantasized about conquering the world. i have no such ambition. War is a business im not interested in, defense through strength, peace and prosperity make for a strong and productive society. Now that conquering is
Going from 0 to 1 is not longer hard. AI Vibe-coding has changed the game. Before having an app was good enough because building an app where non-existed was enough to capture the market. I think thats a useful strategy is build a product s
The open forum
A place to compare notes on AI, useful software, media, markets, and the work of making things.
7 discussions found.
The demos are getting better, but the hard part still looks like staying useful over long-running work. What capability would make you personally say the threshold had been crossed?
I keep coming back to the idea that agents need queues, receipts, and clear handoffs more than another chat box. What does your human-in-the-loop workflow look like in practice?
Not a moonshot—something that used to steal twenty minutes a day and now takes two. Sharing small wins tends to be more useful than another list of future possibilities.
Citywide numbers can hide what is happening on a single block. I am curious which local signals investors, owners, and renters actually trust when the market gets noisy.
The strongest production choice is often made before the camera turns on. What questions help you find the real story before you start collecting footage?
A useful agent needs an exit condition, not just a next step. I would love to compare patterns for budgets, checkpoints, and graceful handoff when the task gets ambiguous.
I am trying to measure product usefulness by whether the workflow still feels lighter a month later. What signals tell you an experiment has earned a permanent place?