Pinned Tweet
I was showing @monoscope_tech at a conference this week and realised, damn! We really cooked.
6
23
183
15,935
Anthony Alaribe {···} retweeted
@tonialaribe Fair enough. What's the worst one you've caught that looked fine at first glance? I want to see whether my approval request would have shown it.
1
1
1
40
Anthony Alaribe {···} retweeted
We’re live for Day 2 of Sailsconf 2026! 🔴
Come watch with us, join the chat and meet the folks building with Sails, Node.js, Rust and AI.
Enter the stage: sailsconf.com/2026/stage
Now: @chinedu_10 on building IoT things with Rust as a web dev
Bring a friend. There’s still sconf
Come watch with us, join the chat and meet the folks building with Sails, Node.js, Rust and AI.
Enter the stage: sailsconf.com/2026/stage
Now: @chinedu_10 on building IoT things with Rust as a web dev
Bring a friend. There’s still sconf
1
4
11
1,033
Anthony Alaribe {···} retweeted
@tonialaribe @michaelpisaac Close enough to look right, wrong enough to break a contract - AI
0
1
1
68
Anthony Alaribe {···} retweeted
@tonialaribe @michaelpisaac Man, I had to put a hard cap of 10 files per PR. It introduces friction, but it saves me headaches.
1
1
1
106
Anthony Alaribe {···} retweeted
I keep saying this, this idea of just vibing and accepting everything that is written by the LLM will cost us later, what we are seeing now is nothing compared to the damage that will happen in the future when most systems has just being vibe coded
This week's incidents are a reminder about why "human in the loop" is important. Ie, read your damn code. Understand your systems. Know whats in there and what can go wrong.
No LLM advancement will replace human understanding.
3
1
6
343
Anthony Alaribe {···} retweeted
@tonialaribe Agree. And the less code there is, the easier it is to actually understand it. Half the reason I keep my stuff simple and boring :D
1
1
1
114
Anthony Alaribe {···} retweeted
in the end, it shouldn't matter whether you use AI to make something. LLMs are tools. that's why I dislike the idea of "vibe coding".
if you're making software for humans, you should take responsibility for its quality, reliability and safety.
if all of that is part of the vibe, then all good.
if you're making software for humans, you should take responsibility for its quality, reliability and safety.
if all of that is part of the vibe, then all good.
3
3
6
256
Anthony Alaribe {···} retweeted
That last sentence will not age well.
This week's incidents are a reminder about why "human in the loop" is important. Ie, read your damn code. Understand your systems. Know whats in there and what can go wrong.
No LLM advancement will replace human understanding.
1
1
0
149
Anthony Alaribe {···} retweeted
@tonialaribeecially with the last line. I have a slightly different opinion about reading the code, though.
reading the code does not necessarily prevent incidents because humans make mistakes too.
however, humans should verify the code produced by LLMs and understand how our systems can fail.
the direction should be towards building systems of verification that can keep up with the scale at which LLMs produce code.
reading the code does not necessarily prevent incidents because humans make mistakes too.
however, humans should verify the code produced by LLMs and understand how our systems can fail.
the direction should be towards building systems of verification that can keep up with the scale at which LLMs produce code.
1
2
1
334
Anthony Alaribe {···} retweeted
@tonialaribe the scope of a single human understanding their entire system will continue to shrink.
It'll be impossible to do just reading the code as they're generated. It's unfortunately not optimism. It's coming from a place of concern and trying to figure out how to appropriately adapt to this new setup. We outsource knowledge all the time. packages, libraries, frameworks, security tools etc. High level stuff. This would be no different unfortunately.
It'll be impossible to do just reading the code as they're generated. It's unfortunately not optimism. It's coming from a place of concern and trying to figure out how to appropriately adapt to this new setup. We outsource knowledge all the time. packages, libraries, frameworks, security tools etc. High level stuff. This would be no different unfortunately.
1
1
2
66
Anthony Alaribe {···} retweeted
@tonialaribe human in the loop equipped with frontier intelligence is the way
1
2
5
428
Anthony Alaribe {···} retweeted
it is pure unadulterated cope if you think we won't have 0days or CVEs in the future after AI has written all the code in the universe.
5
11
71
2,351
Anthony Alaribe {···} retweeted
@tonialaribeagreed on LLMs being tools. What I'm alluding to are the harnesses and sandboxes. They will get really advanced. We built the large systems we have today on private keys with just the letter "4" in them. We ran production systems on PHP with remote execution vulnerability for decades. These were run by companies and governments with a lot to lose.
As much I understand the underlying concerns, I fear that we are giving too much credit to our ability as humans to continue to spot and address severe issues at scale.
Heck, GCP in europe-west2 was knocked offline for 8 hours because of a power outage in London. multi-region deployments be damned - Revolut was knocked out cold.
So yeah, I see a world where the focus shifts to designing great sandboxes around programming outputs from agents, than about code that goes into those sandboxes.
I fully expect handcrafted and human reviewed code to become a niche space for small experiments. One cannot simply do it at competitive scale.
As much I understand the underlying concerns, I fear that we are giving too much credit to our ability as humans to continue to spot and address severe issues at scale.
Heck, GCP in europe-west2 was knocked offline for 8 hours because of a power outage in London. multi-region deployments be damned - Revolut was knocked out cold.
So yeah, I see a world where the focus shifts to designing great sandboxes around programming outputs from agents, than about code that goes into those sandboxes.
I fully expect handcrafted and human reviewed code to become a niche space for small experiments. One cannot simply do it at competitive scale.
1
1
4
444
Anthony Alaribe {···} retweeted
@tonialaribeilway founder bit was a bad look.
The timing couldn't have been worse. The PHP issue also appears to have been known for some time. In all of these cases, the humans did nothing.
I unfortunately see a world where these models just write and fix the code regardless and human resign to it due to pressure from competition, output volume etc
The timing couldn't have been worse. The PHP issue also appears to have been known for some time. In all of these cases, the humans did nothing.
I unfortunately see a world where these models just write and fix the code regardless and human resign to it due to pressure from competition, output volume etc
1
1
1
146