lol, what? You’re gonna hold 20 people hostage on the bus until some enforcers navigate a busy city to ticket a person who is likely to wipe their ass with the ticket? What country is that exactly?
Seriously, other than law enforcement what else can you do to someone who brazenly refuse to follow the rules? Even law enforcement (at least in the US) highly depends on where you live. In left leaning states and cities, DAs are not very likely to prosecute such small crimes like not paying a bus fare because they know it’ll make them unpopular next election. I live in a very left leaning county and state and it swings between center and left every 4 years or so. The swing is always “look how awful that guy was. He prosecuted vulnerable people for petty crimes for no reason”. Cops don’t wanna have to deal with all the paper work to book a guy for a couple of nights before they get released and do it all over again. If they know the person will not get prosecuted because there is no political capital to do so, why bother with the theatrics and all the paper work of arresting them? Brazenly refusing to pay the bus fare and getting in a verbal altercation with the driver and everyone on the bus is a fun afternoon for some people.
You end up with an outcry from the rich “liberals” (for lack of a better word), who never take the bus in the first place, complaining about how enforcing fares on buses is harming the poor who can’t afford transportation and pushing people away from public transportation.
It’s pretty infuriating. I started biking to work 2 years ago and try to bike almost anywhere I can. Mostly to lose weight but also put my money where my mouth is. I voted for every levy and prop to improve bike-ability and public transportation of the city in the last 10 years and figured I’m a hypocrite if I expect others to bike and take the bus and I never do. My tolerance for the homeless on buses has been dropping as I have to deal with them more and more. I was always “It’s our failure in not helping them. If I can’t help, least I could do is let them be” kind of person. Now every other week I end up with a negative interaction with someone on the bus or at a bus stop. Every time I air my grievances with people I know (who never take the bus) I always have to find myself on the defensive somehow.
For me, 2023 was an entire year of weekly demos that now looking back at were basically a "Look at this dank prompt I wrote" followed by thunderous applause from the audience (which was mostly, but not exclusively, upper management)
Hell man, I attended a session at an AWS event last year that was entirely the presenter opening Claud and writing random prompts to help with AWS stuff... Like thanks dude... That was a great use of an hour. I left 15 minutes in.
We have a team that's been working on an "Agent" for about 6 months now. Started as prompt engineering, then they were like "no we need to add more value" developed a ton of tools and integrations and "connectors" and evals etc. The last couple of weeks were a "repivot" going back full circle to "Lets simplify all that by prompt engineering and give it a sandbox environment to run publicly documented CLIs. You know, like Claude Code"
The funny thing is I know where it's going next...
I can't take anyone seriously who uses prompt engineering unironically. I see those emails come through at work and all I can do is roll my eyes and move on
Assuming you mean "context engineering" as in "engineering the context for LLM prompts" - 0.
I take particular issue with the usage of the word "engineering", in this context, as in my practical experience what I witnessed was more akin to "try random things until it somewhat works" than anything I would associate with "engineering". But hey, it's a free country and people can use words whichever way they want. Just shouldn't be confused if noone keeps on listening ;)
But did it work? This is the sticking point with me now. I've seen slides, architecture diagrams, job descriptions, roadmaps and other docs now from about a dozen different companies doing AI Agent projects. And while it's completely feasible to build the systems they're describing, what I have not seen yet is evidence of any of them working.
When you press them on this, they have all sorts of ideas like a judge LLM that takes the outputs, comes up with modified SOPs and feeds those into the prompts of the mixture-of-experts LLMs. But I don't think that works, I've tried closing that loop and all I got was LLMs flailing around.
It hasn’t really worked so far. Pretty much exactly what you’ve described. I don’t even really work on that team, but “a judge LLM” low-key triggered me just because of how much I’ve been hearing it over the last couple of months.
I think the reason of the recent pivot is to “keep the human in the loop” more. The current thinking is they tried to remove the human too much and were getting bad results. So now they just want to make the interaction faster and let the human be more involved like how we (developers) use Claude code or copilot by checking every interaction and nudging it towards the right/desired answer.
I got the sense that management isn’t taking it well though. Just this Friday they gave a demo of the new POC where the LLM is just suggesting things and frequently asking for permissions and where to go next and expecting the user to interact with it a lot more than the one-shot approach before (which I do think is likely to yield better results tbh) but the main reaction was “this seems like a massive step backward”
I think long-term just having a single LLM responsible for everything will win out compared to brittle and complex subagent hierarchie. Most use of "subagents" today are just workarounds for LLM limitations: lack of instruction following, context length, non- determinism, or "hallucinations".
All of these are things that will need to be solved long-term in the model itself though, at least if the AI bubble needs to be kept alive. And solving those things would in fact materially improve all sorts of benchmarks, so there's an incentive for frontier labs to do it.
I think this is why you have the back-forth pattern that GP mentioned. You start with a single model doing everything. Then you find all sorts of gaps that you start to plug ad-hoc, and decide that breaking it into subagents might help fix things. This works for a while but then you realize that you lose out on the flexibility of a single-model having access to the entire context, so you starting trying to improve communication between subagents. But then a new model drops that fixes a lot of the things you originally had to workaround, so you go back to a single-model setup. Rinse and repeat. It's a great VC- bubble funded employment program though.
The general pattern seems to be that LLM+scaffolding performs better than LLM. In 6 months time a new model will incorporate 80% of your scaffolding, but also will enable new capabilities with a new layer of scaffolding.
I suspect the model that doesn’t need scaffolding is simply ASI, as in, the AI can build its own scaffolding (aka recursive self-improvement), and build it better than a human can. Until that point, the job is going to remain figuring out how to eval your frontier task, scaffold the models’ weaknesses, and codify/absorb more domain knowledge that’s not in the training set.
You are talking about context management stuff here, the solution will be something like a proper memory subsystem, maybe some architectural tweaks to integrate it. There are more obvious gaps beyond that which we will have to scaffold and then solve in turn.
Another way of thinking about this is just that scaffolding is a much faster way of iterating on solutions than pre-training, or even post-training, and so it will continue to be a valuable way of advancing capabilities.
You don't. Envoy is great if you programmatically configure it, or if you have very small and simple configs. It can't be maintained by a human. But if you have tools that generate it programmatically based on other config, you can read through it.
My understanding was always that this is a way to monetize a text editor. How else do you monetize dev tools? Developers are used to very high quality free tools. You’re either one of the few old guards (like JetBrains, Microsoft or maybe Oracle) that can sell IDEs and other dev tools because 25 years ago open source dev tools were far from beginner friendly.
But how do you monetize a programming language, a text editor, a build system, a terminal emulator, etc in 2025? The examples are deno, bun, mojo, nextjs, zed, earthly, warp, etc. all know they can’t monetize the actual tool. You monetize services that you build around the tool. Like a cloud/workers/deployment (basically compute), or a sharing service or an AI service, etc. once you have critical mass on your platform, you can find other easy services to offer. Like if Zed has a critical mass of users, maybe the offer “in editor chat”. A small startup with just 3 devs working together can replace slack with zed. Maybe they offer an uptime check service. Why not? Maybe a file sharing service. Maybe a small wiki service, etc. all things that have million other solutions. But if you have critical mass, someone will pay for those things.
I'm honestly all for it. As AI keeps poisoning all aspects of online content and online interactions, people who care about that sort of thing will have to move back to in person interaction more.
The person you're physically interacting with might be using AI in their workflow, that's fine. I use AI too. I just don't want to "build a relationship" with AI. I don't care for AI "content". Art, blogs, articles, advertisements, even detestable things like sales and marketing are all forms of human relationships to me. It's fine if you wanna autogenerate it. I'm even "in the market" for autogenerates stuff as I use AI bots too, but you can't try to sell me a 100% automated-burger when I have the Fabricator 3000 too.
If I'm hungry and just want a burger, I'll get my Fabricator 3000 to generate one for me. If I'm in the mood for a human touch on food and a dining experience, I'll cook, go to a (reputable) restaurant or a friend's place who likes to cook. Maybe there is a market for you to run your Fabricator 3000 to generate a burger for me. maybe. I don't know why I'd buy it though when I can just get your prompt and feed it into my own Fabricator 3000...
Cute story, but do you think a massive company that specializes in loans and credit would move to the largest credit market in the world with a 100 years of very detailed financial data and models about loans, delinquencies, defaults, bankruptcies, etc but “it never occurred to them” to check that?
I mean, that was the literal whole point of the article, that they were culturally caught off guard. It happens! I'm going to try to find it again and edit this with a link.
I’m not doubting the existence of such article. I’m asking you to engage some critical thinking and not accept whatever cute, PR or otherwise “fun” narrative you read anywhere.
It’s “fun” to think that Starbucks struggles in Italy is because they are dumb and didn’t realize that Italians have a long cultural tradition with coffee that’s very different than Americans. “Oh dumb Americans and dumb Starbucks. How stupid are they to try to sell espresso in the land of espresso lol”. You need to engage just a little bit of critical thinking and realize that Starbucks knew that well in advance and thought the would give it a shot anyway. Maybe they could occupy a different cultural niche just like they did in dozens of other countries with similarly rich cultural traditions around coffeeshops and coffee or tea like the Middle East and Asia.
Critical thinking is long gone from most, and this site is now popular enough, and cs is main stream enough, that a good majority of this site even barely thinks beyond what fluff was put in front of them.
Status pages stopped being automated a long time ago because they are bad PR.
Often you’d have dozens if not hundreds of services on a status page. If you have a major networking outage for example, then everything is technically down. Someone screen shots the sea of red that your automated status page is showing and tweets “lol everything is down at [insert company]. Then you get a million imverysmart people posting about single point of failure or whatever.
As a result status pages, in every place I know, require a human to actually declare the outage there. Internal ones are usually automated, but if your service is down due to dependency on another service, you don’t mark yourself as down.
Also most places I know of have moved away from public status alerts anyway. You get a customized alert in your account or email if you happen to be impacted by a particular outage. The public ones are for the very very _very_ bad outages.
My understanding is that it's also a legal CYA. If you have SLAs in place, outages might mean you owe money. So companies tend to err on the side of underreporting.
CloudKitchen was holding plenty of software hiring events near me few years ago. All these companies developed software to streamline the process of listing thousands of stores on all delivery apps, receiving orders, organize and manage kitchens assembly lines like how the orders are received, and dispatched to cooks, etc. Also they integrate analytics, cost tracking, supply chain management, and other random things like that. Basically for any PE or a billionaire who wants to larp as the next McDonalds or Starbucks, they don’t want to build everything from scratch.
The kinda thing a regular restaurant is probably managing using a spreadsheet and a notebook.
Seriously, other than law enforcement what else can you do to someone who brazenly refuse to follow the rules? Even law enforcement (at least in the US) highly depends on where you live. In left leaning states and cities, DAs are not very likely to prosecute such small crimes like not paying a bus fare because they know it’ll make them unpopular next election. I live in a very left leaning county and state and it swings between center and left every 4 years or so. The swing is always “look how awful that guy was. He prosecuted vulnerable people for petty crimes for no reason”. Cops don’t wanna have to deal with all the paper work to book a guy for a couple of nights before they get released and do it all over again. If they know the person will not get prosecuted because there is no political capital to do so, why bother with the theatrics and all the paper work of arresting them? Brazenly refusing to pay the bus fare and getting in a verbal altercation with the driver and everyone on the bus is a fun afternoon for some people.
reply