HN user
Building software for US immigration
awais@post.harvard.edu
Quality and speed are not diametrically opposed. A great engineer does well on both axes by building the minimal thing needed now in a way that is easy to extend in the future.
I have also seen projects go badly because the eng was trying to be perfect upfront. Whereas quickly getting to an MVP and then iterating tends to go better.
I’m curious about this scale vs speed distinction.
Every codebase includes parts that are more experimental, and parts that are more core. My sense is that AI can help on both of these fronts (I.e building rapid prototypes on the fringes and hardening the core with better test coverage).
Engineers seem to think business people don’t know what they are doing, but if your post were true, then companies would add slack to outperform their competitors.
The broken system likely doesn’t have enough business impact to justify the investment to maintain it.
When his supervisor sent him a paper to read, Bob asked the agent to summarize it. When he needed to understand a new statistical method, he asked the agent to explain it. When his Python code broke, the agent debugged it. When the agent's fix introduced a new bug, it debugged that too. When it came time to write the paper, the agent wrote it. Bob's weekly updates to his supervisor were indistinguishable from Alice's.
In my experience, doing these things with the right intentions can actually improve understanding faster than not using them. When studying physics I would sometimes get stuck on small details - e.g. what algebraic rule was used to get from Eq 2.1 to 2.2? what happens if this was d^2 instead of d^3 etc. Textbooks don't have space to answer all these small questions, but LLMs can, and help the student continue making progress.
Also, it seems hard to imagine that Alice and Bob's weekly updates would be indistinguishable if Bob didn't actually understand what he was working on.
My prediction is that we'll start to see a whole new layer of abstraction to help us write high quality code with LLMs - meaning new programming languages, new toolchains, stricter typechecking, in-built feedback loops etc.
The slop we're seeing today comes primarily from the fact that LLMs are writing code with tools meant for human users.
The EV tech seems to be good enough already in China
Doesn't the conversation.history permission let a Slack bot extract all messages? https://docs.slack.dev/reference/methods/conversations.histo...
People were saying the same last year, and then Anthropic launched Claude Code which is already at a $2.5B revenue run rate.
LLMs are useful and these companies will continue to find ways to capture some of the value they are creating.
I was wondering this too. I assume it’s because child care costs are lower when one parent isn’t working(?)
They also did an MRI scan on Honnold and found that he doesn't have the usual fear response. It's not clear if this was trained away, or if it's something innate.
https://nautil.us/the-strange-brain-of-the-worlds-greatest-s...
They wrote:
If you optimize below 1487 cycles, beating Claude Opus 4.5's best performance at launch, email us at performance-recruiting@anthropic.com with your code (and ideally a resume) so we can be appropriately impressed and perhaps discuss interviewing.
That doesn’t seem snarky to me. They said if you beat Opus, not their best solution. Removing “perhaps” (i.e. MAYBE) would be worse since that assumes everyone wants to interview at Anthropic. I guess they could have been friendlier: “if you beat X, we’d love to chat!”
It seems like the biggest downside of this world is iteration speed.
If the AT instagram wants to add a new feature (i.e posts now support video!) then can they easily update their "file format"? How do they update it in a way that is compatible with every other company who depends on the same format, without the underlying record becoming a mess?
I enjoyed hearing Claude Code creator Boris Cherny talk about "latent demand"[0], which is when users start using your product for something it was not intended for. When that happens, it's a great signal that you should go build that into a full product.
Cowork seems like a great application of that principle.
Claude can also write and debug tailwind for you! :)
It’s more about cognitive load, and abstraction level. If you’re trying to make an object spin, it’s much easier to use the tailwind class than it is to remember css keyframes.
Sure, when debugging a complex issue, it’s worth knowing the low-level, but CSS is not a great abstraction for day-to-day work.
People like tailwind because it feels like the correct abstraction. It helps you colocate layout and styling, thereby reducing cognitive load.
With CSS you have to add meaningless class names to your html (+remember them), learn complicated (+fragile) selectors, and memorise low level CSS styles.
With tailwind you just specify the styling you want. And if using React, the “cascading” piece is already taken care of.
Agentic AI companies are doing millions in revenue. Just because agents haven’t spread to the entire economy yet doesn’t mean they are not useful for relatively complex tasks.
A philosophy of software design by John Ousterhout is the best software book I have read.
But, from your post it’s not clear specifically what you are looking for. If you think you will level up by learning how to apply numerical modelling techniques, then it’s probably best to focus on that.
I think this is a huge use case for vibe coding, and something I’m excited to do more in 2026.
There are so many apps I want, that companies are not incentivised to build for me.
I'd bet a lot of people are trying to optimize their codebases for LLMs. I'd be interested to see some examples of your ASI-unlocking codebase in action!
I predict people will want an ultra-thin phone, even if it’s just for weight reasons.
We’ll see what the sales numbers are like.
Yeah it's not clear to me how many of his shares he sold but it seemed to be a substantial chunk.
The board seemed to be pretty anti-Amelio already so it wasn't clear to my why Steve _needed_ to sell these shares.
It seems like agentic browsers will develop aa new set of core primitives (e.g. always ask for manual approval when spending money), and this flavor of security vulnerability will go away.
Web browsers didn't begin with the same levels of security they have now.
I fully agree. I for one am excited about a future in which we can take on bigger challenges. With (good) LLMs, we won't need to spend as much time thinking about "how" to get things done, and can spend more time lucidly deciding "what" we want done.
If anything, with the right tooling LLMs should improve the quality of our thinking. For example, how much better would your thinking be if you could ask the LLM to reliably figure out all the 2nd and 3rd effects of your ideas, or to identify hidden assumptions.
I'm pretty sure the job offer requirement is still in effect, and will remain in effect, since the O is an employment-based visa. What made you think it had changed?
I have lots of examples:
* By law, the US can only issue 140,000 employment-based green cards per year, and no more than 7% to one country. This means people from India or China can face a 100+ year backlog, even after they have proved they qualify for a green card. There's no cap on marriage-based green cards.
* Processing times for many green cards (i.e. for people who have already qualified, but just need the physical green card), are 12-24 months.
* USCIS still expects many applications to be sent by mail. Some applications (like O-1s, EB-1s) require hundreds of pages of evidence, and it all needs to be printed out on 8.5x11" paper, for USCIS to scan it in on B+W scanners. This means that there is no error checking (e.g. on fee amounts), and if you have made a mistake, you might not know about it for weeks. Also, it means your petition cannot include working hyperlinks, webpages, or videos - the USCIS officer judges the petition by scrolling through a 400+ page PDF.
* The 'standard' post-graduate work visa is the H-1B. It's entirely lottery-based, not merit-based, and typically there are 400,000+ people competing for 85,000 visas. Many qualified people are forced to leave the US each year because they didn't get selected in the lottery.
Wouldn't this person be put on a PIP and then fired if their performance didn't improve?
Even at companies with non-uniform salaries, it's difficult to down-level someone. Their morale will drop, the team's anxiety will go up, and (if they were genuinely bad for a long time), the team will wonder why they weren't fired.
It seems like it's worth taking some time to steel-man the AI argument, even if your CEO hasn't made it very well.
E.g. If you don't work on AI now, and AI models keep improving, how likely is it that a competitor who integrates AI well will eat your lunch? If it's >50%, it seems worth it to shift some focus to AI regardless of the series C round.
This post from a few days ago has some great tips on how to integrate AI _well_: https://koomen.dev/essays/horseless-carriages/
This is excellent! One of the benefits of the live-demos in the post was that they demonstrated just how big of a difference a good system prompt makes.
In my own experience, I have avoided tweaking system prompts because I'm not convinced that it will make a big difference.