jon-wood

I also thought I naturally didn't enjoy exercise and had resigned myself to at best forcing myself to go to the gym and grind out a session on a rowing machine or something. I then, in one of the best impulse decisions I've ever made, decided I'd go check out the climbing gym nearby and it turns out I just hadn't found the exercise that clicked for me.

On topic for the wider thread of gamifying doing things that are good for you, bouldering in particular works perfectly with my brain. Every few weeks there's a whole set of new things to try and climb to the top of, and its often as much about working out a sequence of moves that works for as it is about brute strength.

I've never really been able to make exercise stick for an extended period but this time round I'm on about 18 months of consistently making it to the gym twice a week or so and pulling on plastic holds for a while. Its (mostly, unless I have a bad session) fun while doing it, the combination of exercise and problem solving means my system is flooded with endorphins and dopamine afterwards, and I'm fitter than I've ever been.

I have memory turned off for everything, context management is the fundamental technique in working with LLMs, and I absolutely do not want random additional things being added to that context without my knowledge because it results in nonsense like using Claude to do work on the industrial IoT SaaS product I work on professionally and being told I could just use the Home Assistant instance which manages my flat to deal with that problem instead.

I’ve been trying to nudge agents (both Claude and GPT) into a red/green/refactor TDD loop and I’m increasingly unconvinced it’s worthwhile. TDD works for humans because it forces us into a pattern of doing the simplest thing that could work, then thinking about how to refactor that. An LLM can be prompted to do that without the skeleton of a failing test, and I plan to experiment with other methods of pushing them towards the same ends.

Hooks should (in my opinion) be deterministic. As an example I’ve also noticed Claude writing Python scripts to extract fields from JSON when jq is available to it. They almost always contain the same patterns, and having seen this I’m going to write a hook which triggers on those and fails the turn telling it to use jq instead.

It’s entirely possible if the application developers have made the choice to intentionally avoid tracking and tying personal data to user identity, which is exactly who this product is aimed at.

iCloud Private Relay (available on every macOS and iOS device in recent years) uses OHTTP to obscure your identity from sites you visit.

Most people are not competition ranked players, they’re people who like to play a few hours of a game now and again.

TypeSafe AI are the providers for Jev. This complaint is like saying it’s a red flag that Claude Code sends lots of information to somewhere called Anthropic.

> Except now LLMs let you, with very little technical know-how, plaster a ‘programmable’ interface on top of unstructured data/interface meant for humans, and because that's what we actually want, of course everybody does that.

I don't think that is what everyone wants, or at the very least its not what I want. My ideal software has a well designed and thought through user interface for the primary interactions, allowing me to think about the goal I'm trying to achieve rather than how I'm manipulating the underlying data model.

Additionally to that please also provide access to the underlying data. Ideally in the form of a locally accessible API but I'll accept a remote one if necessary, provide a CLI on top of that as well if you like.

There's a whole set of fallbacks built in to the standard, including Bluetooth, local network connections, and going via a relay server. All of them eventually end up with your device signing something and handing that back to the browser on the other device to complete the authentication flow.

Any time you've got hardware and want to attest that it hasn't been tampered with before allowing it to interact with something like an API endpoint.

At work we deploy industrial IoT gateways, these are very much not end-user devices. We are actually concerned about the device's identity, and more specifically about being able to attest that the device is in fact the one we thought it was and it hasn't been tampered with. By putting the key for TLS client certificate in the device's TPM, locked behind attestation that what's been booted is what we expected to boot, we can have a reasonable degree of confidence that we're communicating with the device we thought we were rather than just someone who managed to copy the private key off disk.

The first time I went for a diagnosis I was made to sit the same QbCheck test, which was then used as the entire justification for why I couldn't possibly have ADHD despite every other aspect saying I did. Apparently 30 years of having to sit at a computer and do a task in order to eat and pay rent means that I can sit down for 20 minutes and do a task at a computer.

I'm still quite bitter about this because it then took me another two years to work up the motivation to go get retested (this time somewhere that doesn't use snake oil), followed by an additional year to work up the motivation to sort out coordinating with my GP to get on meds. That's three years I could have been a functioning human being.

I don't think this is designed as a journey planning tool, it's a data visualisation showing the extent of your local bus routes and doesn't really need to be any more than that.

Instinctively this feels like a two phase problem - start with some machine translation into a single spoken language and index that, then when people are querying do the same thing. When returning search results show them in the original language.

It’s Google turning Google Pixel phones into servers, they were able to call in the hardware design team to work out what was essential and how to strip the devices down, and then the firmware team to be able to boot a standard Linux kernel on them. This is a very different situation to most people’s experience of trying to do out of the ordinary things with a phone.

They're projects where there's not a payment page to buy a copy sure, but VSCode especially has I'm sure indirectly led to a ton a of profit - its the natural home of Copilot, integrated with Github (including in their paid product Codespaces), and has I'm sure led to a ton of standardisation and productivity increases across the company.

There are few, if any, countries which have taken capitalism to the same extremes the US has. The concept of a country which is so at the mercy of capitalists that people will beg you not to call an ambulance for them when they’re injured is incredibly foreign to me despite living in the UK, a country renowned for its financial sector.

Which I guess is to say yes, it is the culture of a country, and the culture of the US strikes me as money above all else.

I kind of agree with what you're saying, in that screenshot and share functionality is increasingly overloaded with options and things that just don't make any sense in that context. I disagree that this isn't because a product designer lacks vision, Apple control the entire stack on iOS, and there's nothing stopping them from rethinking things so that "Share" isn't a weird synonym for "Junk drawer menu".

There is no justifiable reason that Safari's find in page option is under the share menu, hidden underneath a View More button. Finding within a page has nothing to do with sharing, and even if somehow a user guessed that it might be in that menu without searching the web for it, they're probably going to dismiss it because they haven't expanded the view more menu.

iOS and macOS are riddled with illogical decisions like this. The menu bar is an unusable mess of icons that things have added, a problem that Windows (kind of) solved about a decade ago by allowing me to select which icons appear by default. The Dock seems to choose at random whether an app remains pinned or not, with particular confusion brought by apps updating themselves. Shortcuts is a wonderful idea in principle while in practice anytime I try to do something with it some part is missing.

I used to be a huge proponent of macOS in particular, and I would like to be one again, but currently its incredibly hard to do that.

Really? You'd be ok with me putting a proxy server on your home network then, which anyone with a few bucks can use to attach your IP address and subscriber details to anything they choose to request from the internet? How about a Tor exit node?

Its incredibly obvious to anyone applying any thought at all to this that its a malicious to sell a product that labels itself as a TV streaming stick which is in fact a paid for relay server with the money made from providing the internet connection to a random third party unrelated to the person who bought the thing without ever telling the customer.

The typical consumer has no idea what they're buying, and they shouldn't have to because the retailer selling the product should have done some basic due diligence before stocking the thing. People aren't going to some clearly shady Chinese website and buying a device labelled "cheap TV streaming stick, will sublease your internet connection to criminals", they're putting "FireTV" into amazon.com and somehow being presented with these things alongside the Amazon FireTV they expect to find, or maybe "streaming stick" which really shouldn't be surfacing clearly malicious products.

Amazon even have big "people commonly return this product" warning on some product pages. Anywhere halfway sensible would maybe reconsider stocking a product worthy of that but because they've set themselves up as a middleman without any of the risk they can just churn junk out of their warehouses.

I really wish Ruff would introduce something similar to Nix’s stateVersion, which is used to determine the set of defaults that will be applied. Updating Ruff at a scale beyond a single repo is a bit of a crapshoot currently, with every version introducing a bunch of new default rules which you then have to deal with immediately (either by turning them off or fixing them).

I’m aware we could have an allowlist in place which specifies all rules that are turned on but I’d much rather have a simpler config file, and be able to bump the state version at a good time for everyone to spend a few hours fixing new violations.

Are you really saying that in 2026, a year when only the nuts are still trying to claim climate change isn’t a thing, the most intelligent approach to the problem of powering data centres that are mostly being built for the purpose of juicing share prices is gas turbines at an industrial scale? The most intelligent approach is to not build the things. The next most intelligent approach is solar and batteries nearby. Way down the bottom of the list is burning gas to power them.

There is absolutely a reason for propagation to wait an hour or a day, and you've set that value in the TTL field for the record in question. It stands for Time To Live and specifies how long downstream resolvers should cache the last value they saw before querying the authoritative server again, if you set that value to a few seconds then almost every DNS resolver in the world will get updated values within a few seconds as they expire the cached value.

> Delays are hopefully minimal on modern systems

The delays are almost always self inflicted by people who've blindly put 36400 in the TTL field like some sort of magic charm rather than considering how long they want the cache expiry to be. Long ago I was one of them and then I had the company greybeard point out that DNS record updates are a thing we had control over - just drop the TTL to 30 seconds or whatever a day before a planned change of servers and (barring the odd stubborn resolver which thinks it knows better than you do) the time it takes for those caches to refresh goes from 24 hours to 30 seconds.

No. No you can't. The specific example of preparing a playlist sure, but Carplay goes way beyond that. You can preload the navigation route, tweak it a bit, then when you sit down and plug your phone into the car its on the navigation screen. You can make phone calls using the microphones in the car. In a sufficiently integrated car you can have turn by turn directions pop up on the dashboard or a heads up display rather than having to look down into the center console while trying to work out which lane you should be in.

At least half the people firing off LLM generated PRs will have left the "Coauthored-by: Claude" line on it allowing automated rejections.

Why are you optimising for token use rather than getting whatever tasks you're trying to do done? Max effort Opus will happily sit there talking to itself for five minutes fixating on every tiny detail while Sonnet or even Haiku will just blast through the problem and be ready for the next one in seconds.

Oh this was in 2000, when virtualisation was only just becoming accessible so I can get of get the justification. It still made the entire exercise in writing some software feel pointless when I knew it would never get executed by anyone but myself.

If the focus of this is truly on one-per-person personal domains then you don't need to allow transfers and reselling. (Although you'll probably get a grey market of people just repointing DNS to someone else anyway, because if there's money to be had someone will take it)