Techno-optimist, long time AI/ML guy, love augmenting human capabilities with AI software.
Currently Director of Eng/AI building experimental AI products at Datadog (skunk works). Previously founded a small AI startup called Augmend (acquired) and love to stay involved helping AI startups in any way, reach out.
Previously led teams/orgs in the ML engineering and product space for: Meta (PyTorch), Amazon (Alexa AI and AWS Games/Simulations), Microsoft (Cortana and XboxOne spoken language understanding), Lockheed (stuff), and various other less interesting gigs before that.
"eval startups have a hard time finding customers, because clients have to be technical developers who want to build with APIs, but also not technical enough to run their own evals"
To add to this, they have to be developers who aren't already using a fullstack observability solution, since it's fairly straight forward to add the eval startup featureset to an existing observability solution, and easier (plus cost effective) to just keep it all in one place.
I’ve always wanted to start a company that builds automated underwater swarms of “probes” that just search and return info and carry out small exploration tasks but over long amounts of time and space.
Do it right and you can send the first underwater explorers to Europa.
Hard to find the right way to monetize in the early stages though. SpaceX had a variety of options.
Yep. I love open source but there isn’t a model that comes close still to the closed source options like Opus 4.8 and that’s obvious from most people I see across the software industry as well. There are at least another few models after Opus from OpenAI and Anthropic most would go down the list using before any of the Chinese models at this point.
Press x to doubt. Moving slower is never the better way to create shareholder value unless you’re implying a million other things that are going wrong in your definition of moving faster.
“Move fast with stable infrastructure” as they say.
I’m relatively certain it’s just this at the end of the day. Everything I see people doing in their custom built TUIs or claude/codex CLI can be done, likely even easier, in a simplified IDE or easier to scan UI, but it feels nice/cool/cyberpunk/work-like to look like you’re doing more.
Everyone will have a “reasonable” explanation though for why they have to stay in the terminal even when they aren’t really coding anymore and it wouldn’t be hard to have a window next to your terminal if you really have to, but live and let live. Whatever makes you happy as be all become managers.
I too like a cyberpunk interface even if it’s last the need :)
None of the important projects or FANGs build with frameworks like that, sorry. Project management frameworks are all larp and mostly used by slow moving firms from the past.
1. Full carrying cost of an employee is much more then their salary so this math is not as straight forward if you’re just cutting time and salary to account for that time.
2. You should assume most people aren’t counting hours in places like Meta, reducing to a 4 day week imho will start making people think more about counting exact hours they’re working. It’s partially why the “4 10s” concept is also a bad idea that permeates the defense contractors.
3. Staying focused 5 days a week for one person probably has better compounding effects for that week than a few people working part time and taking longer to get the work done with longer breaks in between “sessions”. Harder to measure of course but it’s one thing I’d be worried about. Easier to think about if you say each person works 2.5 days a week for half their pay, I’d rather just have one person.
you're telling me dispatchagents.ai :) (open to new names if anyone has cool ones, didn't expect anthropic to start using dispatch with their agents, naming is way too hard)
Pretty neat but as someone who commutes every day on the New York subway I hope it’s never “cracked” here. Phone usage without headphones is already annoying enough and I greatly appreciate the various people trying to take calls eventually lose service.
If I recall correctly, a prior interview about claude plays pokemon stated they purposely chose pokemon as a use case that was not meant to be trained/finetuned on. That's what makes it an interesting problem, so hopefully they aren't.
Not wanting to use Vercel is honestly a good enough reason. If you’re a heavy Vercel user you probably aren’t their target market since they’re aiming at enterprise types from what it looks like.
Currently Director of Eng/AI building experimental AI products at Datadog (skunk works). Previously founded a small AI startup called Augmend (acquired) and love to stay involved helping AI startups in any way, reach out.
https://twitter.com/diamondbishop
[email protected]
Previously led teams/orgs in the ML engineering and product space for: Meta (PyTorch), Amazon (Alexa AI and AWS Games/Simulations), Microsoft (Cortana and XboxOne spoken language understanding), Lockheed (stuff), and various other less interesting gigs before that.