prescriptivist

I want to say this with the caveat that I am generally a person who always contends with the contradictions of living in a capitalist-imperialist country and my own distaste for it. So this doesn't come from a place of American exceptionalism writ large, but I am a firm believer the we did get this part right:

Public lands and culture of the ability to access wild places, whether for hunting, fishing, hiking, camping, and just generally an affordance of access to wilderness that is codified into the laws of the country. In Europe they have the concept of "Right to Roam" which is a powerful concept that I appreciate (and in ways is superior to our systems for just walking in the woods) but it is also fundamentally different than the almost legalistic systems we have in this country towards public lands.

My surface understanding of China is that there is no such broad remit given to the people of China and there aren't designated places where the people of China can just go and exist in wilderness. Such places might exist by convention but they don't have the sort of legal framework that we have in America to recreate in these places.

I'm a principal engineer, been working on the same set of codebases for almost 10 years. I handle the 20% or so of my time that constitutes inbound faster than ever and I know because that inbound volume has clearly increased and yet I have, for the first time ever, begun chipping away at the "nice to have" backlog. My biggest time sink now is interviewing and code reviews -- the latter being directly proportional to the velocity increase across the teams I work with. Actually that's my biggest concern -- we are approaching a breaking point for code review volume.

Sorry I don't have DX stats or token usage stats I can share, but based on the directives from on high, those stats are highly correlated (in the positive).

[edit] And SEV rates are not meaningfully higher.

Congrats on that.

> We can in fact just NOT do things!

I agree with you on that. Not just on AI but a lot of things that suck about this world, and in particular the United States. But capital is too powerful. And these tools are legitimately transformative for business. And business pays our bills and, more importantly, provides the healthcare insurance for our families. The wheel is a real fucking drag isn't it?

I don't see anything short of a larger revolution against capital stopping or even stemming this. For that to really happen we would need a lot more people and interests than just those of software practitioners. Which may come yet when trucking jobs collapse and customer service jobs disappear. I don't know. I do know that I'm taking part in something that will potentially (likely?) seed the end of my career as I know it but it's just one of many contradictions that I live with. In the meantime the tools are impressive and I'm just figuring out how to live with them and do good work with them and as you can probably tell, I'm pretty convinced that's the best we can make of the situation right now.

> Is there no longer any need to think about new software paradigms, build new frameworks, study computer science, because the regurgitated statistical version of programming is entirely good enough?

All I'm saying is you're gonna have to figure out how to do this with an agent. It's not that I don't see value in the craft; it's just that value is less important. As far as the new paradigms, the new frameworks, new studies in computer science -- they still exist, it's just that they are going to focus on how to mitigate heisenbugs, performance regressions and security holes in agent written code. Who knows.. in five years most of the code written may not even be readable. I'm not saying it's going to be like that, but it's entirely possible.

In the meantime, there's nothing stopping you from using the agent to write the code that is every bit as high quality as if you sat down and typed it in yourself. And right now there is a category of engineers that exclusively use agents to create quality software and they are more efficient at it than anybody that just does it themselves. And that category is growing and growing every day.

I may be out a job in five years because all of this. But I am seeing where this is going and it's clear and so I'm going to have to change with it.

I don't think that people who don't want to use these tools or clean old ways are incurious. But I think these developers should face the fact that those skills and those ways they are reticent to give up are more or less obviated at this point. Not in the future, but now. It's just that the adoption of these tools isn't evenly distributed yet.

I think there's a place for thoughtful dialogue around what this means for software engineering, but I don't think that's going to change anything at this point. If developers just don't want to participate in this new world, for whatever reason, I'm not judging them, but also I don't think the genie is going back in the bottle. There will be no movement to organize labor to protect us and there be no deus ex machina that is going to reverse course on this stuff.

Glad to hear this. I actually went down this path based off of guidance from multiple LLMs (Anthropic, OpenAI, etc.), so I wasn't sure if it was just some kind of weird hallucination they all had or if they were regurgitating a very small amount of knowledge on this topic, because it was kinda hard to find stories where people had success with these strategies. Thank you for the link to the paper. I will definitely be reading it.

Things are changing so fast and so chaotically with this technology. I'm also writing everything now using Claude code, and I've been thinking a lot about what this means for my work moving forward. One thing I've noticed, is that I will just keep hammering and hammering on my work until I force myself to quit. Even on the weekend I feel the pull to go work on it. I'm just less sort of mentally exhausted by work, I suppose, but I don't think that's particularly healthy if it leads to me working way more than I should. On one hand, I think that's a reflection of how powerful and exciting this technology is, but on the other hand, I think that it triggers some different kind of reward function in my mind that I'm not used to.

In any case, I think if one wants to continue to have a career in this industry for years to come, it's basically table stakes to become fluent in using these tools.

A missing link right now is automated high-quality code reviews. I would love an adversarial code review agent that has a persona oriented around all incoming code being slop, that leverages a wealth of knowledge (both manually written by the team and/or aggregated from previous/historical code reviews). And that agent should pull no punches when reviewing code.

This would augment actual engineer code reviews and help deal with volume.

Anthropic makes it kind of clear in all of their statements that they are not opposed to working with the surveillance state, with the military industrial complex, etc. Their central philosophy, it seems, is not incongruent with working with entities, public or private, that can be construed as imperialist or capitalistic or a combination of both. I actually appreciate their honesty here.

They exist within the regime of capital and imperialism that all of us who are American citizens exist within. This isn't a cop-out or cope. It's just the reality of the world that we live in. If you are an American and somehow above it, let me know how you live.

I have an agent system analyzing time series data periodically. What I've landed on is the tools themselves pre-process time series data, giving it more semantic meaning. AKA converting timestamps to human dates, additionally preprocessing it with statistical analysis, such as calculating current windows min/mean/max value for the series as well as a the same for a trailing window and surfacing those in the data. Also adding a volatility score, and doing things like collapsing runs of similar series that aren't particularly interesting from a volatility perspective and just trying to highlight anomalous series in the window in various ways.

This isn't anything new. It's not particularly technical or novel in any way, but it seems to work pretty well for identifying anomalies and comparing series over time horizons. It's even less token efficient on small windows than piping in a bunch of json, but it seems to be more effective from an analysis point of view.

The strange thing about it is that it involves fairly deterministic analysis before we even send the data to the LLM, so one might ask, what's the point if you're already doing analysis? The answer is that LLMs can actually find interesting patterns across a lot of well presented data, and they can pick up on patterns in a way that feels like they are cross-referencing many different time series and correlate signals in interesting ways. That's where the general purpose LLMs are helpful in my experience.

Breaking out analysis into sub-agents is a logical next step, we just haven't gotten there yet.

And yeah the goal is to approximate those of us engineers who are good at RCAs in the moment, who have instincts about the system and can juggle a bunch of tabs and cross reference the signals in them.

Just Wispr Flow and a PTT key binding. It's very good for doing plans with Claude Code because I can just ramble and ramble. As long as I just convey the details of what I want over a sufficiently long string of text, it will work even if it has errors in speech-to-text or I have slight contradictions in my framing of the prompt.

If I need to explicitly reference files in the plan prompt, I just manually annotate them into the prompt at the end.

> People on here act like we don’t know if AI will be useful. And I’m sitting over here puzzled because of how fucking useful it is.

Yes, it's very strange to read AI threads here because the general tone is so different than, say, at the company I work at, where hundreds of engineers are given enormous monthly token budgets and are being pushed to have the LLMs write as much code as possible. They're not forced to, and no one is reprimanded for not adopting Claude Code or Codex or Cursor. But there's been a strong tonal shift in technology leadership in the last month that basically implies that this is how it is going to be done in the future whether one likes it or not.

As for me, I've been writing all of my code via Claude for a while now, and I don't think I will ever go back to working in an editor writing code the way I did for most of my career. Nor do I want to.

This is a funny point that you're making (for me, anyway), because prior to early December, probably 5% of the lines of code I wrote in a week were AI-generated by cursor. Then I started using Claude Code. Fast forward to today, I would say 98% of the code that I've shipped in the last three weeks has been written completely by Claude Code.

Prior to three weeks ago, I had used speech-to-text to do accomplish approximately 0% of the work I've done in my 20 years of coding. In the last three weeks, well over half of the direction that I've given to Claude Code has been done with speech-to-text.

Yeah, if this is going to be a nuclear bomb for software developers, imagine what it's going to be like for people in customer service, account managers, etc.

I would say the vast majority of people in this thread don't believe that this is related to AI at all, other than as a pretext. It's kind of incredible.

This seems like a real coarse and not particularly accurate binary, but even if it were true, the thing about Claude Code and agentic coding like this is the cost of making a mistake or the cost of not being happy with a design and having to back it out is getting smaller and smaller.

I would argue that rapidly iterating reveals more about the problem, even for the most thoughtful of us. It's not like you check your own reasoning at the door when you just dive head first into something.

I have. 3 is fine, 3.1 is good. But they are terribly slow. Quality is fine but the the only thing they have going for them is flash pricing. Their response performance sucks.

Assuming these plans are based on Gemini, Google is doing these users a favor, frankly.

Honestly? Yeah.

I've been writing code for 25 years.

A year ago my org brought cursor in and I was skeptical for a specific reason: it was good at breaking CI in weird ways and I keep the CI system running for my org. Constants not mapping to file names, hallucinating function names/args, etc. It was categorically sloppy. And I was annoyed that engineers weren't catching this sloppy stuff. I thought this was going to increase velocity at the expense of quality. And it kind of did.

Fast forward a year and I haven't written code in a couple of weeks but I've shipped thousands LOC. I'm probably the pace setter on my team for constantly improving and experimenting with my AI flow. I speak to the computer probably half the time, maybe 75% on some days. I have multiple sessions going at all times. I review all the code Claude writes, but it's usually a one shot based on my extensive (dictated) prompts.

But to your identity crisis point, things are weird. I haven't actually produced this much code in a long time. And when I hit some milestone there are some differences between now and the before days: I don't have the sense of accomplishment that I used to get but also I don't have the mental exhaustion that I would get from really working through a solution. And so what I find is I just keep going and stacking commit after commit. It's not a bad thing, but it's fundamentally different than before and I am struggling a bit with what it means. Also to be fair I had lost my pure love of coding itself, so I am in a slightly weird spot with this, too.

What I do know is that throwing myself fully into it has secured my job for the foreseeable future because I'm faster than I've ever been and people look to me for guidance on how they can use these tool. I think with AI adoption the tallest trees will be cut last -- or at least I'm banking on it.

> there _might be_ new bits but nothing meaningful as far as I know

If you didn't need backwards compatibility with older rubies you could use Ractors in lieu of forks and not have to IPC between the two and have cleaner communication channels. I can peg all the cores on my machine with a simple Ractor pool doing simple computation, which feels like a miracle as a Ruby old head. Bundler could get away with creating their own Ractor safe installer pool which would be cool as it'd be the first large scale use of Ractors that I know of.

I think fibers (or, rather, the Async library) in Ruby tends to be fetishized by junior Rails engineers who don't realize higher level thread coordination issues (connection pools, etc) equally apply to fibers. That said this could be a pretty good use case for fibers -- the code base I use every day has ~230 gems and if you can peel off the actual IO bound installation of all those into non-blocking calls, you would see a meaningful performance difference vs spinning up threads and context switching between them.

I used this in earnest yesterday on my Zillow saved listings. I prompted it to analyze the listings (I've got about 70 or so saved) and summarize the most recent price drops for each one and it mostly failed at the task. It gave the impression that it paginated through all the listings, but I don't think it actually did. I think the mechanism by which it works, which is to click links and take screenshots and analyze them must be some kind of token efficiency trade-off (as opposed to consuming the DOM) and it seems not great at the task.

As a reformed AI skeptic I see the promise in a tool like this, but this is light years behind other Anthropic products in terms of efficacy. Will be interesting to see how it plays out though.

That's awesome and I have a few similar conversations with Claude. I wasn't quite an AI luddite a couple months ago, but close. I joined a new company recently that is all in on AI and I have a comically huge token budget so I jumped all the way in myself. I have my choice of tools I can use and once I tried Claude Code it all clicked. The topology they are creating for AI tooling and concepts is the best of all the big LLMs, by far. If they can figure out the remote/cloud agent piece with the level of thoughtfulness they have given to Code, it'd be amazing. Cursor Cloud has that area locked down right now, but I'm looking forward to how Anthropic approaches it.

Agree. I'd add that a aha moment to skills is AI agents are pretty good at writing skills. Let's say you have developed an involved prompt that explains how to hit an API (possibly with the complexity of reading credentials from an env var or config file) or run a tool locally to get some output you want the agent to analyze (example, downloading two versions of python packages and diffing them to analyze changes). Usually the agent reading the prompt it's going to leverage local tools to do it (curl, shell + stdout, git, whatever) every single time. Every time you execute that prompt there is a lot thinking spent on deciding to run these commands and you are burning tokens (and time!). As an eng you know that this is a relatively consistent and deterministic process to fetch the data. And if you were consuming it yourself, you'd write a script to automate it.

So you read about skills (prompt + scripts) to make this more repeatable and reduce time spent thinking. At that point there are two paths you can go down -- write the skill and prompt yourself for the agent to execute -- or better -- just tell the agent to write the skill and prompt and then you lightly edit it and commit it.

This may seem obvious to some, but I've seen engineers create skills from scratch because they have a mental model around skills being something that people must build for the agent, whereas IMO skills are you just bridging a productivity gap that the agent can't figure out itself (for now), which is instructing it to write tools to automate its own day to day tedium.

Skills have a lot of uses, but one in particular I like is replacing one off MCP server usage. You can use (or write) an MCP server for you CI system and then add the instructions to your AGENTS.md to query the CI MCP for build results for the current branch. Then you need to find a way to distribute the MCP server so the rest of the team can use it or cook it into your dev environment setup. But all you really care about is one tool in the MCP server, the build result. Or...

You can hack together a shell, python, whatever script that fetches build results from your CI server, dumps them to stdout in a semi structured format like markdown, then add a 10-15 line SKILL.md and you have the same functionality -- the skill just executes the one-off script and reads the output. You package the skill with the script, usually in a directory in the project you are working on, but you can also distribute them as plugins (bundles) that claud code can install from a "repository", which can just be a private git repo.

It's a little UNIX-y in a way, little tools that pipe output to another tool and they are useful in a standalone context or in a chain of tools. Whereas MCP is a full blown RPC environment (that has it's uses, where appropriate).

Another thing I'd suggest: look into and use non-coding AI tools that improve productivity. For example:

Zoom meeting transcriptions and summaries or Granola. A lot of context is lost when you take manual notes in meetings. If you use a tool that turns a meeting into notes automatically, you can use those notes to bootstrap a prompt/plan for agents.

Unless there is not something I am seeing, people aren't racing to move to rural New England. Maybe it's retirees, red to blue state migrations, or remote workers. But I haven't seen a ton of evidence of that. People didn't really migrate out here before covid and I don't think enough people have to justify the rise in prices.

Personally I think people that otherwise would be selling are sitting on their homes because of the interest rates and this is causing a strange feedback loop of low turnover causing low supply which in turn causes new buyers to accept the prices (probably with a hope that interest rates will come down and they can re-fi in the years to come). I also think a non-trivial number of houses that on the market due to the owners passing or going into retirement homes are sitting there on the market because prices are so high but the only money the family is out is taxes. Or they are being turned into rental units, since rental prices are out of whack in these areas too.

My point I guess is where I live we haven't seen a big influx of population (probably the opposite) or significant job or wage growth to make sense of the increase in housing prices. I guess at the end of the day people are just stretching themselves further and sending more money to the banks in the form of interest to get into homes that were literally half the price in 2019. Strange times.

There's something fundamentally strange about how prices have spiked and inventory has tightened since Covid. Where I live in rural New England, prices are up 50–100% in five years. And this is on pretty poor quality homes. Yes, low interest rates led to a surge in buying and bidding wars that spiked the baseline, but when people say "the real problem is there isn't enough housing" that feels incomplete to me. Of course supply has been an issue for a while, but home prices nearly doubling in five years doesn't look like a normal supply story -- it's not as if we suddenly created 20-50% more qualified buyers in that time. I guess the lack of churn, with people hanging onto those sweet 3% mortgages much longer than usual, is probably part of it. But I really don't have an answer for the current state of home buying. I make great money but if I was to buy a house the quality of the house I got in 2018 with the same % down payment I would be looking at over 40% of my take home going to a mortgage, PMI and taxes.

I mention mean-time to decision making and that's one of the rationales for the mcp. A skill could call a script that does the same thing -- but at that point aren't we just splitting hairs? We are both talking about automated repetitive thinking + actions that the agent takes? And if the skill requires authentication, you have to encode passing that auth into the prompt. MCP servers can just read tokens from the filesystem at call time and don't require thinking at all.