julianharris.bsky.social
- Ex-Googler cutting through the BS about AI agents https://makingaiagents.substack.com
- Passionate about the climate crisis.
- Also 2 sons & a muso
842 posts
1,339 followers
3,611 following
Regular Contributor
Active Commenter
comment in response to
post
Bluesky needs growth, not polls.
comment in response to
post
hey jp you called it
comment in response to
post
My measure of AI coding agent performance is inverse to the number of swear words I dole out to them in a chat.
Interestingly, with Sonnet 4 Max on Cursor my profanities have been surprisingly sparse. But then also it cost me $80 from one day of usage š±
comment in response to
post
Same
comment in response to
post
Ah I have a sibling version of that.
comment in response to
post
Interesting! Doesnāt seem to touch on schema migrations though (I could have missed it?)
comment in response to
post
Thereās at least one company now moving from literal retrofits to entirely new cars inspired by yesteryear designs.
comment in response to
post
I watched a 15 minute YouTube video on this so Iām something of an expert on the topic. Biggest issue is retrofitting batteries: you canāt really sneak them into crevasses and theyāre really heavy so itās better to have them low, flat and centred.
comment in response to
post
This is all I can see on mobile
comment in response to
post
Chrome has a neat mobile device test mode.
comment in response to
post
Is there any decent ice cream in the US? Seems like the US palate has a stronger preference for sweetness than Iāve experienced elsewhere.
comment in response to
post
Nice. Could you add mobile support? Itās very hard to use on mobile currently.
comment in response to
post
Why canāt the ground truth be built in?
comment in response to
post
Seems like the message is āyou cannot fully automate evals yet, human in the loop is still neededā ā correct?
comment in response to
post
Mortgage free house?
comment in response to
post
10 like tickets please ššššššššš
comment in response to
post
Funny guy
comment in response to
post
Yes, if we look at the conventional axes of faster cheaper better, then what does that look like? And is there an equilibrium point where you canāt really get faster cheaper better. Whatās the CD audio moment? What is the retina display moment?
comment in response to
post
Yeah strange slightly breathless feeling isnāt it when the whole solution is implemented in seconds correctly despite having spent maybe hours coming up with the prompt.
comment in response to
post
source?
comment in response to
post
So much fun. Are we witnessing a renaissance in the game industry? I saw a post that suggested this early raps innovation is how it was in the good old days of gaming.
No Mac version is that because Steam?
comment in response to
post
Had a UPS guy try and not collect a critical parcel recently. My security camera detected movement and I caught him scarpering off āI did ring the bellā. Camera says otherwise mate.
comment in response to
post
This is amazing. Was this hand-coded or did you use AI to generate any boilerplate?
comment in response to
post
If you want low risk, buy a lottery ticket. High risk: make a profitable software business.
comment in response to
post
I see an interesting question too about *precision*: ATM, based on prompts Iāve seen, thereās a vague notion that you can get high accuracy output from low-precision input which seems ā¦ optimistic.
(Quite aside from the difficulty of high accuracy output from high precision input.)
comment in response to
post
Nice: does this cover the new skills a manager needs to manage hybrid teams of human and AI agents? And cases where itās just AI agents? What are some resources youād recommend to read while we wait in eager anticipation until October?
comment in response to
post
I think itās more a case of they donāt care
comment in response to
post
Hope. Messages of hope are always valued, and sometimes can be the tipping point for someone out there who is struggling.
comment in response to
post
I tried it this morning and for small research tasks I found it to be quite good. But there is a limit after which it just gives up and says āno Iām not gonna do it and here are placeholdersā
Iām going to share insights vs other Deep Research solutions in my newsletter makingaiagents.substack.con
comment in response to
post
Or ā¦ any ā¦ Iād argue š
comment in response to
post
Try Superwhisper superwhisper.com
comment in response to
post
I make a few attempts at it in my newsletter šhttps://makingaiagents.substack.com
Each issue it charges a bit ā¦
comment in response to
post
Oh wow, nice! I found a video on it here youtu.be/pWZw7hYoRVU?...
comment in response to
post
Which part? Nothing I could see here at least github.com/pydantic/pyd... what am I missing?
comment in response to
post
Iāve seen research that shows that people consume more content when the stuff you really want is hidden a bit because you end up browsing more.
They are almost purely driven by attention and retention so Iād assume that filtering reduces it in some mysterious wayš¤
comment in response to
post
Come on itās never about the money this is a massive redistribution of wealth and power. Those two guys are not remotely interested in whatās good for America or anyone else. They envy dictators like OrbĆ”n and the others and thatās that they are working on.
comment in response to
post
You missed āsolve an important urgent problem your customers will pay forāā¦
comment in response to
post
I cycle between VS Code / Windsurf / Cursor as I routinely lose patience with all of them for different reasons šš±š¤¬
comment in response to
post
Charges are so much based on usage so I couldnāt comment on that. I think everyoneās trying to find out the cheapest way of doing this stuff. I think the absolute cheapest way is to get the code generated by fixed subscription like Claude and copy and paste. (Itās how I started 2 years ago.)