Michael Khait

🇺🇦🇮🇱 Investing in the family. Building software. Growing teams. CTO @trustpointai. ex- @thoughtworks, @edcast, @ebay

[Michael Khait @mkhait]

@hsaab310 @serenaa_ge Yes I too am very curious: where does composer 2.5 stand against the big guys?

[Michael Khait @mkhait]

@petergyang It's stable. Never had issues with gateway. Observing memory sytem works very-well too.

[Michael Khait @mkhait]

@lennysan @trustpointai
Remote/contract
DM me

[Michael Khait @mkhait]

@Markmanson Nope

[Michael Khait @mkhait]

@MatthewBerman yeah. sticking with it for now. Great product. There is something about seeing code and being able to change it, when I feel like it.

my .openclaw folder is a Cursor project

[Michael Khait @mkhait]

I'm an avid user of #JournalApp on the #ios . I think it's a great product and a very smart component that Apple has developed.

I wish more app developers would use the API.

https://t.co/jYuSZ4lia7

[Michael Khait @mkhait]

I have things to say, I suppose I should be saying things more.

[Michael Khait @mkhait]

@Jason 💪

[Michael Khait @mkhait]

@rowancheung @memdotai Mem it

[Michael Khait @mkhait]

We are going to flatten a stack of roles in an org. First, we asked designers to write html, now we have VC augmented by AI managing multiple hands on roles.

[Chamath Palihapitiya @chamath]

I own a company - @Hustle - that makes enterprise software. I recently added the responsibility of being their CEO (I'm still CEO of @socialcapital ). I did this because I felt I was losing the specific details of how to build well-run companies in 2023. I was becoming too much of a talking head and not enough of a hacker/doer.

On Tuesday, I met with some of our team to tackle a problem that was leading to increasing volumes of customer support needs. So I offered to try and tackle this using AI.

By the close of business yesterday, I had been able to complete the following steps: initially using OpenAI, then redoing it in Anthropic, and finally writing some basic Python in Replit in a crawl, walk, run sprint:

1) Initially configured a new GPT with OpenAI and added some specific knowledge that helped it learn edge cases, allowing us to create a simple way of helping our customers avoid the problem upstream. This should improve their experience with our tool and impact the error rate in a way that decreases the OpEx to serve these customers (TBD and will confirm when I know for sure) - this seems applicable to many companies.

2) Moved this to Anthropic to expand how/what it could learn (ie does increasing tokens add to accuracy/value to us?). Also TBD.

3) Built it in Replit so that our eng team could fork it, make it less janky and actually integrate it into the product. Replit is nice.

Here are my observations over this past week as someone who was never a great coder, didn't have a dev environment that had evolved beyond vi and had only talked about AI up until now without actually doing anything with it:

1) The abstracted way to create functionality is really impressive. I wrote and edited the "PRD" in a text file in human english. This feels profoundly disruptive by increasing the top of the funnel of those who can now build products.

2) Adding knowledge to the model was easy, but once you hit a threshold, fine tuning is the only way. I haven't started this process yet so will have to report back on how hard this is.

3) Every single tool I used was excruciatingly slow and unusable in any reasonable production setting. I know it’s early, but man, it’s slow. Super slow. Did I mention how slow it is?!?! Whoever can help solve this will be a big winner...

Will report back as I learn more and try new things with AI to help this company grow…

h/t to @sundeep for showing me that i could do this

[Allen Holub. https://linkedIn.com/in/allenholub @allenholub]

User stories are supposed to be unclear. They're anything from a word to a sentence long with no implementation details. (They describe the user's work, not yours.) They are a reminder to have a conversation just before and during implementation. Fill in the details during Sprint Planning or equivalent��just enough detail to start, not finish. Collect additional details with more conversations as you work. Collecting lots of detail up front is just a waterfall requirements-gathering phase. No agility there.

[Michael Khait @mkhait]

Who is going to play @sama in the upcoming @netflix psychological thriller about last weekend?

[Michael Khait @mkhait]

Is @bluesky still a thing?

[Michael Khait @mkhait]

Chamberlain has been historically very bad and closed as they address technology. Here’s a proof. https://t.co/GDzJATrpmy

[Michael Khait @mkhait]

But what if I don’t follow any ports and don’t want to see it in my top navigation for Apple News? Maybe a setting option to turn it off?

[Michael Khait @mkhait]

I still do this 20 years later to verify if I'm connected to the internet. Still the best way 🤣

[Michael Khait @mkhait]

My latest brisket cook on ⁦my Masterbuilt Gravity 800

Data-Driven Cooking (DDC)⁩

[Michael Khait @mkhait]

@heyabhishek @memdotai Mem it

[Michael Khait @mkhait]

@carlvellotti Remind me of an early days of the @AppStore and how various interactions were tried and failed. Now, guidelines and interactions are pre of less established.

[Michael Khait @mkhait]

@lobanovskiy @dribbble It’s like a next BMW model in a release year. It’ll grow on you.