qazxcvbnmlp

Humans decide what to remember based on their emotions. The LLM’s don’t have emotions. Our definition of good and bad comes from our morals and emotions.

In the context of software development; requirements are based on what we wanna do (which is based on emotion), the methods we choose to implement it are also based mostly on our predictions about what will work and not work well.

Most of our affinity for good software development hygiene comes from emotional experiences of the negative feelings felt from the extra work of bad development hygiene.

I think this explains a lot of varied success with coding agents. You don’t talk to them like you talk to an engineer because with an engineer, you know that they have a sense of what is good and bad. Coding agents won’t tell you what is good and bad. They have some limited heuristics, but they don’t understand nuance at all unless you prompt them on it.

Even if they could have unlimited context, window and memory, they would still need to be able to which part of that memories is important. I.e. if the human gave them conflicting instructions, how do they resolve that?.

I eventually think we’ll get to a state where a lot of the mechanics of coding and development can be incorporated into coding agents, but the what and why we build will still come from a human. I.e. will be able to do from 0 to 100% by itself a full stack web application, including deployment with all the security compliance and logins and whatever else, but it still won’t know what is important to emphasize in that website. Should the images be bigger here or the text? Questions like that.

Cad in general isn’t good at modeling spacial relationships between parts as a graph.

Some specification exists as formal constraints. Ie: c code will or will not compile.

However some specification only exists in natural language. IE: make this page optimized for a smartphone. The task of turning that vague direction into formal requirements is work in and of itself. The more you can have the llm help with that — the more time it will save you.

My mental model is that ai coding tools are machines that can take a set of constraints and turn them into a piece of code. The better you get at having it give its self those constraints accurately, the higher level task you can focus on.

Eg compiler errors, unit tests, mcp, etc.

Ive heard of these; but havent tried them yet.

https://github.com/hmans/beans

https://github.com/steveyegge/gastown

Right now i spent a lot of “back pressure” on fitting the scope of the task into something that will fit in one context window (ie the useful computation, not the raw token count). I suspect we will see a large breakthrough when someone finally figures out a good system for having the llm do this.

I think this was the missing piece from the article. The simple behaviors enabled interest in a market, but at the end of the day, the original Roomba didn't work well. It's cool to romanticize the simple behaviors, but like they miss things.

I messed around a bit and found it annoyingly rigid. In reality I would have already established clear communication and rapport with both the ic and the vp.

When we setup our sprint goals we would have built in time to handle random requests aannddd I would have kept in good grace with the vp so that he knows were on the same team even if I say no to the request.

Also, how I respond depends on why the vp decided to backchannel me. If they did because they didnt care about our teams goals (and only theirs) then I might need to escalate to their management to set clear boundaries. If they did so because in the past I forgot about their requests, then I probably need to not forget about their requests.

If they are clearly not looking out for our shared interest (or that of the company) and instead only worried about themselves, maybe slightly narcissistic then Im going to respond in a different tone than if they made a genuine mistake.

Without commenting if parent is right or wrong. (I suspect it is correct)

If its true, the market will soon reward it. Being able to competently write good code cheaper will be rewarded. People don't employ programmers because they care about them, they are employed to produce output. If someone can use llms to produce more output for less $$ they will quickly make the people that don't understand the technology less competitive in the workplace.

We humans use timestamps in conversations to reference a persons particular state of reference at a given point in time.

Ie “remember on Tuesday how you said that you were going to make tacos for dinner”.

Would an llm be able to reason about its internal state? My understanding is that they dont really. If you correct them they just go “ah you right” they dont say “oh i had this incorrect assumption here before and with this new information i now understand it this way”

If i chatted to an llm and was like “remember on Tuesday when you said X” i suspect it wouldn't really flow.

and varies person to person

especially based on their prompting skills

The difference is how you relate to the job providing accommodations. If you know certain employers are more willing and able to provide accommodations then you can consciously weigh that piece of information when considering a job/career field.

By consciously accepting who you are and how you work with the world, it lets you navigate better in it. For some people that is just feeling it out and ending up in a career that fits them. For some people, it might be getting a diagnosis. The end result might be the same.

This is an important distinction. Indeed, many behaviors (ie attachment issues) that are maladaptive for an adult can be adaptive to a child and it is important to not change those without taking into account the environment a child is in.

An 8th grader may not have control over how their time is spent, but an attuned response from the people around them will help the child adapt.

A child experiencing: 'everyone around me can take this test, but i cannot, I must be dumb'

vs

A child experiencing: 'everyone around me can take this test, but i cannot, I must be dumb' and a caring figure in their life explaining to them 'you show traits of ADHD, this commonly makes it harder for you to focus on things like a math test. it really hurts when you fail the test and you wanted to get an A. Why don't you try again at the same problem at home, I believe in you. And maybe I will talk to your teacher about some extra time for the next test. We can't always get this, and even if you don't pass the test it's ok.'

Well put.

> A lot of modern, aspiring-middle-class and online culture

Theres also a pernicious way of identifying with the struggle. Instead of I have trouble focusing in certain situations, so maybe I should find ways to spend my time (careers, hobbies) that work well with that. We instead go to 'I have ADHD' and my 'job' should make special accommodations for me.

Regardless of whether a job should or should not make accommodations. It's not a very helpful construct to think they should. It removes agency from the person experiencing the struggle. Which in turn puts them farther from finding a place that they would fit in well.

For the vast majority of behaviors (ADHD, attachment issues, autism, etc) they exist on a continuum and are adaptive/helpful in certain situations. By pathologizing them, we(society) loose touch for what they mean in our life. It also makes discourse hard because the (this is causing me to truly not be able to function) gets mixed in with the (this is a way that my brain behaves, but I can mostly live a life).

HN commenter logic: 'if I don't understand it, this is clearly wrong'

'If someone doesn't spend money irrationally like I spend money irrationally its bad'

There is indeed a blind spot.

so many opportunities for shenanigans!

'watch this ad before we give you your boarding pass'

'no boarding pass for you until your group is called'

'ah ha! you have an iPhone 17 pro max instead of a $BUDGET_ANDROID, cool your seat fees will be 20% higher'

'oh you got your boarding pass on this device, but need it on a different device now? cool lets do a device transfer fee'

Reading the posters cv and experience, I suspect they have a skill gap in theory of mind. ie, understanding how they are perceived. Sure, the economy is hard, and finding a job is difficult. Questions I have

- What jobs are they applying for? - Do they understand the benefits they can bring to a team? - Are they showing up in interactions like they show up in this blog post? How can they take radical responsibility for the problem of finding job? Doing what you are told and not getting a job sure sucks but if that's all someone tells me about what they did, I am 100% not passing on a good recommendation. - Their resume needs work

Agree 100%

It's very hard to get a job right now, I don't doubt that. Also it's not very helpful in getting a job to look at macroeconomic trends: the relative change in the trends is much smaller than how you show up in the process.

The poster had consulting work, and 3 internships.. I sense a disconnect between what a potential employer needs (ie why they would pay you) and what they have to offer.

Its easier for the ego to go "man the job market it bad", ie if I don't get this job what does that say about 'my worth as a human' but its not very helpful in getting a job.

Brenda also needs to put food on the table. If Brenda is 'careless' and messes up we can fire Brenda, because of this Brenda tries not to be carless (also other emotions). However I cannot deprive an AI model of pay because it messed up;

I use Render, spend remarkably little time doing devops. Its fantastic

Case in point: I have several technical manuals that look like books but have no ISBN numbers.

> My experience has been that if you take the time to explain what the current state is, what your desired state should be, and to give information on how you want the agent to proceed,

I have a pet theory. 1. This skill requires a strong theory of mind[1]. 2. Theory of mind is more difficult in those with autism 3. The same autism that makes people really good at coding, and gives them the time to post on online forms like hn, makes it hard to understand how to work with LLMs and how others work with llms.

To provide good context to the llm you need to have a good understanding of (1) what it will and will not know, (2) what you know and take for granted(ie a theory of your own mind) (3) what your expectations are. None of this you need to do when you are coding on your own, but are critical on getting a good response from the LLM.

See also the black and white thinking that is common in the responses on articles like this.[2]

[1]https://en.wikipedia.org/wiki/Theory_of_mind [2]https://www.simplypsychology.org/black-and-white-thinking-in...

What's the round trip latency on this? ask question -> response. Do you parse the question word by word and feed into llm or wait for the whole question before feeling into llm?

This is the right answer.

Was this good/bad? Idk Room for improvement? Maybe? Clearer direction with the benefit of hindsight? Maybe. but the majority of the sentiment in the responses is coming from people not type rated in a 737.

Software engineers are the translators. We (as a metaphorical community are in denial). Read the 100s of comments on this post: either (AI code is wrong or this isn’t correct or it’s not effective or some other justification). At the end of the day it’s really hard to accept the change.

What these articles miss:

1) not all coding is the same. You might be working on a production system. I might need a proof of concept

2) not everyone's use of the coding agents is the same

3) developer time, especially good developer time has a cost too

I would like to see an article that frames the tradeoffs of AI assisted coding. Specifically without assigning value judgments (ie goodness or badness). Really hard when your identity is built around writing code.

the ratio of {solve interesting problem} to {remember how to compile foo bar with update 23.12.21938 on a full moon} is so much better now

Letting an agent figure out how to compile old projects is magical. What used to be multiple days of slog is now “compile this, make changes and download tools as needed” with 10 mins of git review to make sure it didn’t do anything stupid.

I agree:

Senior engineer: "you shouldn't have to use llm to do simple things" Their boss: "hey I need you to write this thing in Go for performance reasons" Senior engineer: "you shouldn't make me learn new languages, js is plenty performant for this task just install a new server. why are they making us use all these new tools? everything is new again, blah blah excuses excuses"

vs.

Boss: "hey can you write this in go" Vibe Coder: "write this loop to ignore the first semicolon in this response", and then proceeds to move on with life.

It seems like when people evaluate the merits of what a code generation ai agent should and shouldn't do, they leave out a lot of their implicit assumptions about how project should work and their own theory of mind around coding.

"I don't know Claude did that" isn't a bad thing in and of itself... If someone is spending a bunch of time on code that Claude could have done and easily verified it was correct, they are going to move slower and produce less useful things of value than someone who cares about reading every line of code.

usually cynicism is a coping mechanism for underlying/subconscious pain... If you resolve that, then there is room for happier/positive thoughts to emerge. However, it's just like any other addiction (alcohol), it's not resolved in a day and takes hard work.

Much more comfortable to think; (everything sucks, those in power are out to get me) than (wow I really had a high hope for what my life would be, and this sure isn't it).

Great promise; sometimes need to reference docs to build context.

I looked at the examples you posted and did a quick glance. For example

'''init?(exactly: Float80)'''

the tool converted it to

'''- [initexactly-63925](/documentation/Swift/Double/init(exactly:)-63925)'''

To achieve its goal I would be worried that it dropped the verbatim function signature. Claude still figured it out, but for more obscure stuff that could be an issue.