gravypod

I have already seen multiple CLs from people who are both senior and junior where I know they did not look at the code. I thought, at first, they worked and upon review it looked good but when I synced these changes I found either big bugs that a human would never make or a test would never catch or lots of code repetition.

One example was an engineer who was refactoring some code that ended up doing this:

    def execute(jobs):
      for job in jobs:
        asyncio.create_task(compute(job))
        yield await compute(job)

This is very simplified, it as actually broken up into two separate loops and hidden behind multiple nested calls but at some point there literally was a `asyncio.create_task` where the result was not being used.

I looked at this code because we were hitting some quota limits very early for some reason and it turned out we ran 2x the reads we needed to. I refactored this code, 1/2ed the execution time, fixed the quota issues, and took it from ~300 lines to 80 lines.

This was code from a *senior software engineer* with 15 years in the industry. What is very interesting is I see similar bugs from juniors who do or don't use AI.

I am not saying AI can't be useful. On weeks I have had clear tasks set out, while the rest of my team was OOO, I tackled probably 5x the work our team normally accomplishes (this was after all the work was identified, just working). My skip actually said "Wow, we had a very productive week!" so multiple layers noticed the productivity. I think what made this possible was:

    1. I fully understood the **entire** task and the end-user needs.
    2. The code base was structured "fine" with enough decoupling between components that I wasn't hitting merge conflicts with myself.
    3. I self-reviewed all the work before sending anything to other people (opened all the changes in my IDE, read all the tests).
    4. If something seemed too complicated I refactored it manually. 
    5. I left the AI chugging for long periods of time on objectively measurable tasks.

I don't think the practice of engineering software is dead. The architecture of your software now has measurable impact on productivity. I don't think thinking about performance is outdated. If you're running code no one has reviewed but is functional you wasting cycles / money. Having domain knowledge still improves your velocity.

Because of these reasons I think there is still marginal value to programmers. Companies which maintain internal talent pools and build tooling to scale the impact of people will probably beat out smaller companies that just vibe code.

An ex of mine mentioned to me that she was going to join a local community group trying to prevent development on an empty lot near her apartment. I asked why she would want to do that? Her concerns, and the concerns of the people she was showing me on facebook, were:

1. "They didn't do an environmental review" 2. She didn't want to hear construction noise. 3. She didn't want the construction to cause rats to leave the lot and go to buildings. 4. She also said she didn't want her rent to go.

I believe it was this lot: https://web.archive.org/web/20250821002539/https://www.nytim...

I think it is common for people to organize, even in the most urban and educated areas in America, against their own interest.

Have you thought about ways to include the sessions / reasoning traces from agents into this storage layer? I can imagine giving an rag system on top of that + LLM publications could help future agents figure out how to get around problems that previous runs ran into.

Could serve as an annealing step - trying a different earlier branch in reasoning if new information increases the value of that path.

Doh! That makes a lot of sense. I don't know how I missed that. Thanks for the links, they are great reads.

I thought the US primarily got it's LNG supply from fracking. Is this incorrect?

This is very interesting. I wonder if someone could create a future-sight benchmark for these models? Like, if given a set of newspaper articles for the past N months can it predict if certain world events would happen? We could backtest against results that have happened since the training cutoff.

> He’s also noticed the interview process getting longer as companies get “more picky,” he says. One recent company required eight rounds, and he still didn’t get the role.

No one can truely understand how grueling this is until they see this happen. My girlfriend recently had a hard time finding a new gig. Entry level office jobs are doing >7 rounds. Phone, recruiter, HR, manager, manager + team member, manager + skip level, ... etc.

Whats even worse is all the people in your life who are older can't understand this at all.

I tried to explain getting ghosted, ghost jobs, online applying, multi-round interviews to my grandma. It just does not make sense. It has never been like this, why is it like this now?

This was in the news recently: https://investors.micron.com/news-releases/news-release-deta...

I think the arguments for this would possibly be, if AI continues to be useful (generation demand skyrockets): Meta would possibly have a positive ROI for these investments which would lead to others copying the investment strategy and building more nuclear. If that happens a large portion of AI demand would become green(-ish) energy.

If AI demand lowers (generation demand plummets): Meta would have subsidized a bunch of nuclear reactors which would likely continue to produce power for 10 years - 50 years.

A big reason I have heard for lack of nuclear build out is the lack of starting capital but after they are built they are generally stable and maintenance is predictable.

An example of this: https://en.wikipedia.org/wiki/Beznau_Nuclear_Power_Plant. It may be turned down eventually but a 60 year runtime is pretty impressive for 60s engineering!

Does anyone have any ideas for if some of the original seats home assembly manuals exist? I'd love to take a look at those plans and can't find any records.

If these ai companies had 100x dev output, why would you acquire a company? Why not just show screenshots to your agent and get it to implement everything?

Is it market share? Because I don't know who has a bigger user base that cursor.

I've heard this mentioned a few times. Here is a summarized version of the abstract:

    > ... We conduct a randomized controlled trial (RCT)
    > ... AI tools ... affect the productivity of experienced
    > open-source developers. 16 developers with moderate AI
    > experience complete 246 tasks in mature projects on which they
    > have an average of 5 years of prior experience. Each task is
    > randomly assigned to allow or disallow usage of early-2025 AI
    > tools. ... developers primarily use Cursor Pro ... and
    > Claude 3.5/3.7 Sonnet. Before starting tasks, developers forecast that allowing
    > AI will reduce completion time by 24%. After completing the
    > study, developers estimate that allowing AI reduced completion time by 20%.
    > Surprisingly, we find that allowing AI actually increases
    > completion time by 19%—AI tooling slowed developers down. This
    > slowdown also contradicts predictions from experts in economics
    > (39% shorter) and ML (38% shorter). To understand this result,
    > we collect and evaluate evidence for 21 properties of our setting
    > that a priori could contribute to the observed slowdown effect—for
    > example, the size and quality standards of projects, or prior
    > developer experience with AI tooling. Although the influence of
    > experimental artifacts cannot be entirely ruled out, the robustness
    > of the slowdown effect across our analyses suggests it is unlikely
    > to primarily be a function of our experimental design.

So what we can gather:

1. 16 people were randomly given tasks to do

2. They knew the codebase they worked on pretty well

3. They said AI would help them work 24% faster (before starting tasks)

4. They said AI made them ~20% faster (after completion of tasks)

5. ML Experts claim that they think programmers will be ~38% faster

6. Economists say ~39% faster.

7. We measured that people were actually 19% slower

This seems to be done on Cursor, with big models, on codebases people know. There are definitely problems with industry-wide statements like this but I feel like the biggest area AI tools help me is if I'm working on something I know nothing about. For example: I am really bad at web development so CSS / HTML is easier to edit through prompts. I don't have trouble believing that I would be slower trying to make an edit to code that I already know how to make.

Maybe they would see the speedups by allowing the engineer to select when to use the AI assistance and when not to.

I don't necessarily believe this is the best explanation but: We know the economy is doing pretty poorly and tech companies are consolidating. Amazon is losing it's two main drivers of revenue: Irresponsible startups with huge AWS spend and no pressure to optimize their stack and consumers buying treats online. Regardless if people are spending on AI, the only thing businesses are investing on is AI and analysts at AWS are likely signaling that many AI companies are not seeing a large ROI and model developers will likely build their own versions of successful products (Claude Code). AWS doesn't want to scale up it's GPU fleet and be left holding the hardware bag. Amazon can't juice numbers for consumer purchases since the rest of the economy is contracting, most people are losing jobs, etc. So the easiest way to for Amazon to juice their metrics is to offshore office work that can be done anywhere. They can claim they are using AI - but from conversations with friends who are working at Amazon this does not sound very realistic - and ride the AI bubble with no liabilities.

I would love having some first class support for monorepos at bigger organizations (ex: silos, vfs, etc).

Looking at the site there are comparisons to features between WordPress and other non-ai site builders. How does this compare to things like Lovable?

> The PhiCode runtime for example - a complete programming language with code conversion, performance optimization, and security validation. It was built in 14 days. The commit history provides trackable evidence; manual development of comparable functionality would require months of work as a solo developer.

I've been looking at the docs and something I don't fully understand is what PhiCode Runtime does? It seems like:

1. Mapping of ligatures -> keywords (ex: ƒ -> def).

2. Caching of 3 types (source content, python parsing, module imports, and python bytecode).

3. Call into phirust-transpiler which seems to try and convert things into rust code?

4. An http api for requesting these operations.

A lot of this seems to be done with regexs. Was there a motivation for doing string replace instead of python -> ast -> conversion -> new ast -> source? What is this code being used for?

I was just doing research and landed on this exact page last night! I was wondering if anyone knows how someone could mic a room and record audio from only a specific area. For my use case I want to record a couch so I can watch TV with my friends online and remove their speech + show noise from the audio. Setting up some array of mics and using them for beam steering would probably work but there's not a lot of examples I could find on GitHub with code that works in real time.

I really want to try MathAcademy.com. How quickly do you think someone doing light study could move from a Calc 1 -> advanced stuff using that site? In my case I could put in at least 30 minutes to an hour a day.

I would love to see a rendered timeline of the history of these forums and a year by year highlighted diff. It would be cool to see what the explanation was. I'm sure there were many interesting reasons for additional fields that we don't know of today.

> Have you ever had the need to read the source code of a tool because it doesn't behave the way it should, and its help and documentation prescribes?

I believe this varies wildly programmer to programmer. I'm often diving into code and reading that when debugging. I've done this with many code bases. It may be debatable if any of them are "good".

It could be that each article with js charts is a single repo.

Does anyone have any good tutorial series on FPGAs using some of these more affordable devices? Like someone showing how to setup a dev environment on linux and showing how to get started putting logic on devices?

Are those links connected to your comment?

> start by taking on voluntary leadership roles such as mentorship, or running initiatives within your company (setting up an informal team to codify source code management practices, that kind of thing).

This I have been successful with at multiple companies. Often bringing a complete shift in development practices / headcount allocation strategies. I've setup multiple teams and initiatives at my current FAANG employer as an L3/L4. My most recent one was a library which is being used by extremely high priority projects with VP-level visibility.

> Make sure your own management know that you are interested in this path.

This is where I have been extremely unsuccessful. I explained that at some point in the future I wanted to be at the Director or VP level. One of the higher level employees that I was discussing this with laughed at me (she also did the same about my library and other initiatives only later to applaud when our VP thanked me for landing them).

My manager and skip level manager are advising to "take your time" and "you may not actually want that" and similar things. They say to focus on L+1 which I do understand as a priority but what I am really trying to communicate is:

1. Long term I'd like to be higher level leadership

2. I don't know if the skills I am learning right now will put me on this path

3. I want to in the short term work on high impact projects and in the long term build the muscles needed so that when I say "I'd like to be a director" no one laughs at those ambitions

I'd love to hear about the other direction. What kind of skillsets or prep can a problem solved do to transition into more leadership roles? Is it primarily getting an MBA? Getting lucky?

Many people who are unhappy with this tax are people who live in NJ where cost of living is significantly lower and it is much more affordable to own a car. Many people who live in NJ work in the city.

Where are you getting round trip tickets for $6/day from locations which would normally drive in?

Taking NJT one way is $6.75 for me. I also need to take a subway into the office (@$2.75/ride) to make it in on time so $19/day. One could buy a monthly pass at $184/mo or move to somewhere with PATH but for most people dropping $184/month is a lot of money for train tickets and most people who live in NJ need a car for life in NJ.

I don't think this is a good state of things. It would be great if we had plans to expand PATH or increase service times of NJT or build more housing next to NJT/PATH that normal people could afford but all of these seem very unlikely.

A lot of remote dev environments have limitations when it comes to certain types of development. For example, ios and android app development can be tricky. Or game development where you need to have GPUs and build artifacts may be slow to download to your machine.

Are there any guidances for how to fix this?

(Opinions are my own)

I use (and work on) this: https://www.usenix.org/conference/srecon16europe/program/pre...

Basically docs live next to code or in a team-owned folder if there is no code. Code review happens whenever you change docs.

Have you shared your BUILD files upstream?