Is there anything substantial in his list ("agents, subagents, their prompts, contexts, memory, modes, permissions, tools, plugins, skills, hooks, MCP, LSP, slash commands, workflows, IDE integrations") that Claude Code or Cursor don't already incorporate?
I empathize with his sense that if we could just provide the right context and development harness to an AI model, we could be *that* much more productive, but it might just be misplaced hope. Claude Code and Cursor are probably not that far from the current frontier for LLM development environments.
Residents of every city claim that they have the craziest drivers or toughest streets to navigate in. London isn't really that materially harder to drive in than San Francisco.
But I've lived in both places and London is very different than SF. I'd say the UK has better drivers on average (and much more strict licensing requirements), but driving in London is much more challenging due to the tiny roads you have to navigate. There is no road in SF that is as hard to navigate as the average suburban London two-way traffic single car width road with parking on both sides.
An I'm not saying London is "the worst" by any means. It's nothing like driving in Vietnam or India. But it is very different to SF.
In some sense I think the "harder" roads might actually be easier since they just result in everyone moving slower and maybe worst case it gets stuck and requires manual control. While on "easier" roads when something goes wrong people are immediately killed.
Driving in London is kind of weird these days in that you feel almost stationary. It's typically wait at lights for 2 min, drive at 15-20 mph for 40 seconds, repeat. I've mostly given up on it and use an ebike instead.
That is just nonsense, sorry. San Francisco is a modern layout (grid) and London is an ancient city with road networks based on thoroughfares that are 500+ years old.
It may be true that all major cities have their own quirks, but London has significantly more complexity than San Francisco or any US grid based city with super wide roads.
Also, the US bought into the ‘Car is King’ idea whereas that’s never really been the case in the UK outside of a few places like Birmingham. It’s generally harder to be a driver in the UK.
Whether that causes significant problems for Waymo, who knows? But I am also of the opinion that if it works in London then that’s a pretty powerful tell that they’ve got it right. We’ll, at least for places where drivers generally stick to the rules.
This specific bet is very targeted, but we do absolutely have commerciallly available self-driving cars in 2025 in several cities, and the list of cities is rapidly expanding.
An 8-10 year delay from expectations is not too bad all things considered.
Tourists get this impression by staying in the city centre. However, if you live in an outer neighborhood, what would be a 15 min car ride can easily become a 40 minute journey on public transport if you have to walk to the train station, take the train, and perhaps take a bus for the last mile. Now imagine you have a kid in tow or an elderly parent.
Yes, public transit is not a necessity here like in the States, but it's a nice convenience to have, and plenty of people are wealthy enough to pay for it.
I'd love to see someone build Duolingo with a single system prompt. By HN standards, no one would be paying for Character.ai, Cursor, Windsurf, or dozens of others tools because they could just call the ChatGPT API themselves.
He literally says in his post "It might look antiquated but it makes Cursor, Windsurf, Augment and the rest of the lot (yeah, ours too, and Copilot, let's be honest) FEEL antiquated"
The scams and fake job postings on the market are also increasing rapidly, outpacing the growth of overall job listings. This trend is significantly affecting the accuracy of employment and job availability tracking.
Most billionaires cannot extract most of their wealth without crashing the price of their remaining assets. No doubt the gulf is still great, but it's a little smaller than the raw numbers imply.
One wonders how they can buy private jets, luxury cars, and mansions with their modest liquidity, right? It's as if they could extract their wealth over time and not all of it at once, so the remainder doesn't lose its value.
They can access credit by giving poetuons of their wealth as collateral. Do you think they sustain lavish lifestyles and have an outsized influence by crashing the stock value of their companies?
It's much easier to do this than many people think as wealthy individuals frequently borrow against their non-liquid assets. This circumvents the price drop associated with flooding the supply of some stock.
They want Bezos to be forced to sell off his assets and give the money over for government use. Sell it to who, though? Where did they get the money to buy it and why are they allowed to have that money in the first place? Why not just find out who wants to buy Bezos's assets, then take their money instead?
I hear this muddy defense of billionaires every time anyone tries to compare the wealth held by the working class vs. the ruling class, and it's pure BS.
A billionaire's net worth is measured in dollars. If that isn't a valid unit of measuringing their wealth, then find another way to measure it. The theoretical loss in stock price if they sell their holdings shouldn't factor in here, because it isn't universally true, and doesn't actually change the real-world value of their assets.
I too cannot liquidate my entire net worth without selling my house and car, cashing out my retirement accounts, and spending lots of fees in the process. Regardless, my net worth is still measured in the dollars those things are worth before I liquidate them, because that's how wealth is measured.
Musk spent $44B buying a website that has since plummeted in value, and his net worth has only skyrocketed since. It seems like wealth for billionaires does indeed work differently than wealth for the rest of us, and it is fact MORE forgiving for billionaires than it is for us.
2. Transaction costs for liquidating your wealth is materially different from selling enough to significantly affect the market for an asset. As an extreme example, large holders of a meme cryptocurrency cannot sell the majority of their holdings without crashing the value of their coin.
3. Borrowing works for smaller amounts if you can spread out the sales of your assets over a long period of time (or if you don't need to sell at all, e.g. if investing in something that gives you returns).
you fail to see that their net worth is dependent on everybody else’s actions: if the combined shareholders of their companies start selling for whatever reason, what happens to musk’s net worth?
the only way that rich people are rich is because other people want to get rich on their succes and that’s why they buy in. regardless of the real value produced, net worth is much like influencer reach.
No, this is true for most random people. My net worth is largely tied up in assets like my car and house - if everybody else decides these are worthless, because I live in a "bad neighborhood", or if they decide that they prefer new cars way more, I stand to lose huge proportions of my net worth.
This is no less true for the average person. If the real estate market in your area crashes and you own a house, your net worth would crash through other peoples collective action.
The only difference is that most of these billionaires are invested in a few specific companies, however there are plenty of people who (likely for bad reason) are also heavily invested in only a few companies.
There is still publicly available code and documentation to draw from. As models get smarter and bootstrapped on top of older models, they should need less and less training data. In theory, just providing the grammar for a new programming language should be enough for a sufficiently smart LLM to answer problems in that language.
Unlike freeform writing tasks, coding also has a strong feedback loop (i.e. does the code compile, run successfully, and output a result?), which means it is probably easier to generate synthetic training data for models.
> In theory, just providing the grammar for a new programming language should be enough for a sufficiently smart LLM to answer problems in that language.
I doubt it. Take a language like Rust or Haskell or even modern Java or Python. Without prolonged experience with the language, you have no idea how the various features interact in practice, what the best practices and typical pitfalls are, what common patterns and habits have been established by its practitioners, and so on. At best, the system would have to simulate building a number of nontrivial systems using the language in order to discover that knowledge, and in the end it would still be like someone locked in a room without knowledge of how the language is actually applied in the real world.
> Why would it benefit society to get less targeted ads as opposed to more targeted ones?
Because more targeted ads require a dangerous and abusive system of pervasive surveillance while less targeted ads can still be targeted without hurting as many people in the process.
If I am Tylor Swift fan I should get her merch advertising only when I am visiting swifties forum or group - but not when I am checking my fishing forum where I expect fishing gear ads. But nowadays I get adsg
Tracking stats of your customers using the data they willingly gave you and making marketing decisions isn't the problem. Neither is making sponsorship agreements with content producers who make relevant content.
Creation of global markets with interconnected networks to track and share detailed personal information about mental states, political alignment, sexual life and more is the problem.
I empathize with his sense that if we could just provide the right context and development harness to an AI model, we could be *that* much more productive, but it might just be misplaced hope. Claude Code and Cursor are probably not that far from the current frontier for LLM development environments.