San Francisco, CA
- Tweets 30,616
- Following 2,856
- Followers 3,407
- Likes 26,970
Talk to your Mac. LoudInk writes - and acts.
I've been building Loudink for the past few weeks:
🎙️ Dictate clean, formatted text into any app
🤖 Run multi-step tasks by voice — "open Gmail and show me my social tab," "open Calendar in month view"
🔒 100% on-device — your voice never leaves your Mac
New today: I've trained a sub-1GB model that matches our larger default one on both writing and computer use (called "Lite"). Smaller, faster, but just as sharp.
Free to try → loudink.io
2
6
877
Only if you're completely locked in to the Apple ecosystem. Otherwise this doesn't work
5
403
25,617
so what are the negative repercussions? becoming insanely viral on X seems to only be helping him gain popularity and likely more money
5
162
14,509
It's only going to get worse as developers start relying on AI and never actually learn anything about how to build good software
7
3
141
10,012
People do still need to know DOS commands. Might be a small set but it's not nothing 😅 same with mainframes and cobalt
4
111
2,760
Agreed. It was very underwhelming and they should have just released this as a blog post
1
116
23,096
Start from scratch and clone neovim config from GitHub. Good chance to remove all the fluff you no longer use
3
111
8,537
And you'll still be net negative because your fancy AI takes more than 1M a month to operate 😂
3
108
7,970
I disagree. You will always need people editing code, there will inevitably be things the AI will not get right
8
44
12,341
Correct. I've seen way too many "senior" and "staff" engineers be far worse than new grads. Generally titles are meaningless and are usually only influenced by who you know and who in your management chain likes you.
1
3
41
4,722
Yeah he looked very disappointed. I think to make a presentation like this stand out with the content they had, you need the right people to present.
1
37
2,420
"New opportunities to try our product" == I want a new MBP and I don't have any other justification
1
28
2,005
Yes JavaScript is faster than Python. But the rest of this benchmark doesn't reflect reality at all
1
23
3,117
Yes because for complicated software they are useless. You spend more time babysitting wrong answers than getting work done. The waste so much time and effort, delaying you shipping code.
1
18
1,114
I had a three hour conversation about this the other day.. I'm shocked how many engineers don't know how to use the basics of a debugger and they waste countless hours on a slow iteration loop with logs and print statements..
21
1,550
@AmpCode one shots a complex problem in a multi-language repo while all other tools can't get a successful implementation after 10 attempts.. @Sourcegraph is cookin !
1
1
20
536
Why do people think rebasing is so hard? I don't want all the junk history on your dev branch to come into the main branch. Learn to rebase and stop this idiotic merging.
14
261
Not that quickly because companies are still cognizant not to overhire. It'll be faster than 2024 😂 but that's not really saying much
13
4,756
So... it's just going to be a UI wrapper around Chrome? Just like vercel is a wrapper for AWS?
10
1,147
Codex sucks. @AmpCode is by far the best CLI tool for agentic coding
2
9
219
World class engineering 👏
Anthropic's API is down, but @AmpCode is still 100% up because it falls back to using GCP Vertex for Claude inference (absolutely identical model and quality) during Anthropic outages.
Happy coding!
11
401
Because it actually works 😂 good luck maintaining languages like Rust at that scale. When a lot of investment has been made over decades, it's not easy to replace something
1
8
579
Sorry to disappoint but your sample set is clearly biased. If an engineer is making 2 code changes a month and always using AI to write their code it means (1) they aren't working on anything challenging and (2) they will be performance rated out of their job.
2
8
3,051
Hahahaha they are scared and have nothing else to turn too. Typical democratic coping mechanism -> blame others
1
8
722
This is the most underappreciated skill to learn because people typically want to work on shiny new greenfield projects.
I think it would be useful if AI can help understand the hacky workaround that Bob added 7 years ago which hasn't been touched since 😅
9
518
Hmmm not too hard to imagine it accidentally spending all your money and leaking your banking info
9
1,417
Yes! Don't use AI if you are first learning how to code. Actually learn the concepts and get good at the fundamentals.
2
9
1,151
Everyone must use @TigerBeetleDB's TigerStyle #zig. I do not accept any alternatives. @jorandirkgreef will definitely agree.
1
9
302
Yes in many cases they can outperform. Search recommendations is a good example of where tree based algorithms still do very well in industry.
1
8
3,399
go learn something from @PlanetScale - it's a great post
These are our guiding principles for engineering the most resilient cloud database.
planetscale.com/blog/the-pri…
8
373
@samlambert I've worked on many systems that handle a million+ TPS. Netflix could have done way better. This is a solved problem.
1
7
1,049
The reverse is also true. If you're in AI/ML please learn proper software engineering. At the end of the day, productionizing AI/ML will continue to be the majority of the work. Fancy ML without proper platforms and infrastructure are completely useless.
7
828