This is good advice and I wish I had read it before today. I use the state of the art model (Opus 5.5) and telling it with words how to interact with my issue tracker and my repo still doesn’t work. Mostly because of its overly restrictive auto-mode guardrails and brittle command whitelisting, but also a lot of “oops, I didn’t read the doc that you clearly tell me to read at the top of agents.md, I’ll make an ineffective note so this never happens again (it keeps happening)” The solution I’m using now is various scripts that encapsulate almost everything the agents usually need to do and just whitelist those.
I've been working on our custom PR agent at work. The agent phase is minimal, read-only/no bash (except for the .review dir). Most of the effort is in deterministic context gathering with scripts (before), and then posting the lines from a comments.jsonl file (after). I do have to work a something back into the agent phase, such that it can validate the comments.jsonl, because it seems to be generating bad payloads.
It's an iterative take-give, take away as much as you can so the bot can focus on the core task, then give some back to improve the agent output as you find repeated mistakes.
Same. Opus 5.5 was my first exposure to a finally smart-enough model and I was blown away at how it reverse engineered everything I pointed it at. But after a few weeks, the undecipherable prose and "safety-first" everything are making it increasingly unpleasant for my toy project. I really don't care if it breaks anything, just merge the damn PR and stop roadblocking me. Oh and I guess I need to start being less curt with it, or else Anthropic may decide I'm verbally abusing their token generator and ban me?
I'll use the best tool for the job, but I am eagerly awaiting for others to catch up so I can stop giving money to those weirdos. I'm proactively going to start shifting my usage to OpenAI, and then diversify further into other providers as they get good enough.
My impression is Claude is trained to work in production systems and behaves that way instinctively. Some copypasta/context I often feed it is that I am the only user, no production data exists, any data that exists is mine and I accept responsibility for breakage, there is no public release yet, avoid compatibility shims, etc. Stuff like that. Claude will usually get right on board with having a window to make big changes now to avoid technical debt. But you are correct its instincts are very conservative.
> I'll use the best tool for the job, but I am eagerly awaiting for others to catch up so I can stop giving money to those weirdos.
The worst part here is that there is no guarantee that others will catch up in the near to medium term. Their bleeding-edge research culture and strong focus on software engineering above everything else are two competitive advantages that no other frontier labs are even trying to do the hard work in these areas required to displace.
> Everything below L4 is _incredibly stressful_ and unpleasant to drive, because you have to be constantly aware and alert without having anything to do.
Hasn't been my experience for the past year or so, give it a try sometime. Incredibly relaxing and frees up my mind to focus on potential hazards further than just one car length in front of me. Sometimes (don't hate me) frees up my hands/eyes to handle other things that need some temporary attention. Traffic jams, getting cut off by impatient drivers, being stuck behind slow drivers, none of it bothers me. I'm chilling out and talking to my kids and enjoying the scenery.
So basically, you're treating your Tesla as a Level 3 or Level 4 autonomous vehicle, even though its driving system is only Level 2. If you were endangering only yourself, that would be one thing. But you're also endangering other drivers sharing the road.
At this point if you’re able to afford FSD and you choose not to for politics, you’re just making your life worse and less safe for silly reasons. Fine by me as I get driven around in my tank, except for the fact I have to be inconvenienced by human-caused wrecks and my insurance is higher to subsidize them.
The bright minds here are busy posting Trump inauguration photos, trying to “gotcha” free speech advocates, and talking about prison time for tech CEOs. Quite the imagination these guys have!
It was literally today that we heard about TikTok banning the account for the Musk movie because it’s politically inappropriate. This is hardly coming out of nowhere.
lol. 16 years after the first protests started being coordinated over social media, Meta, the largest social media company in the world whose tech folks are probably paid in the millions at the highest level, still hasn't figured out a way to avoid trampling over legitimate dissent for purely 'innocent, technocratic reasons'?
Yeah, I have the same question. I'd like to incorporate cheap & fast models like DS into my workflow to reduce my reliance on OpenAI/Anthropic, but after trying several such models and having to have a frontier model review/fix/redo everything multiple times per task, I gave up and simply have the frontier models (Sol/Opus) do everything. All of this fits pretty comfortably in my monthly subscription.
Maybe the equation changes if I exceed my subscription usage, and need to fill the gap with something else. But it seems like a huge waste of your time to go out of your way to use a cheap model if you have the tokens to spare in your subscription.
Financial flop, maybe. I put down the pre-order fee on day one, then paid foundation series price for delivery and 2 years later it's still objectively the best car I could have for my family (space for toys, self-driving, safest truck on the road; not every family has these requirements). If they stop making it for financial reasons, that's fine, I'll have gotten my years of use out of it and hopefully someone else will have figured out how to put FSD on a production car.
Yeah I’m rushing to reverse engineer some of my favorite games now when I have tokens to burn. Document the addresses, functions, break the encryption, unpack all assets. There’s no way opus keeps this functionality for long. Then I can have a dumber model continue the work or a smarter model with guardrails reassemble the pre-extracted logic and assets.
That’s what I’ve been doing. Managing this stuff by hand is awful and I’m not a sysadmin who is good at it. I just added single sign on to my home network for a wide variety of various services. Some of them didn’t even support SSO but a quick AI patch fixed that. There is no way I would even attempt that by hand let alone succeed.
I saw statements to the effect of “this would be nice but it’s a lot of work” several times in the post. I don’t understand why that has to be a blocker anymore. All this tedium is gone.
I see people say the same things in threads discussing open source agent harness repos that are MIT licensed. I don't understand how it isn't obvious to them at that point haha. I don't think people realize they can just... do whatever they want to
My homelab has many fancy new things like a self-hosted gym app and many more metrics and dashboards in my home assistant instance, both of which I use on a daily basis. It also now runs a linux VM that is an always-on agent workspace, and various emulators & decompilers and their MCPs. I, like apparently everyone else, am also reverse engineering games and mashing them up for my own personal enjoyment. I've modded a few games for fun. I've done a security and utilization audit of my server and saved hundreds of dollars in server upgrades I would have bought if I didn't have AI identifying waste and optimizing things. I could have done all of those things before but I didn't, because of how annoying it is to have to dive so deep into the weeds to do small things.
People who have the technical skills and free time for these hobbies, and aren't having the times of their lives right now, are missing out.
Abstaining from using LLMs isn't doing anything to actually oppose them, it's just denying yourself the stuff you could produce with them.
The game RE stuff is also NUTS right now; the kind of stuff you can do in terms of porting, modding, emulating, etc, is a whole new frontier in personal game ownership.
Like, if you have a copy of a game downloaded, you can prevent it being taken away now. Including by devs turning off remote servers.
reply