ej88

5 ms·

61.
by ej88 8mo ago
Really interesting updates to their 2025 experiment. Repeat devs from the original experiment went from 0-40% slowdown to now -10-40% speedup - and METR estimates this as a 'lower-bound' more devs saying they dont even want to do
62.
We are changing our developer productivity experiment design(metr.org)
88 points by ej88 8mo ago|61 comments
63.
by ej88 8mo ago
surprised nobody responded with the most straightforward, occams razor explanation they think what they're doing is actually good for society not everyone is in the hackerspace libertarian / socialist sphere i used to work for a p
64.
by ej88 8mo ago
Unfortunately agreed, i see swe (already) going the way of finance most jobs are okay comp 9-6s with a small group of elite talent making extremely outsized comp more emphasis on pedigree, exclusive hiring pipelines from top schools
65.
by ej88 8mo ago
Even the source article in the first link, https://www.nber.org/papers/w34836 the same firms "predict sizable impacts" over the next three years late 2025 was an inflection point for a lot of companies
66.
by ej88 8mo ago
claude code 1B+ arr ant 10xing ARR, oai harvey legora sierra decagon 11labs glean(ish) base10(infra) modal(infra) gamma mercor(ish) parloa cognition regulated industries giving these companies 7/8-fig contracts less than 2 years from i
67.
by ej88 8mo ago
the top % of talent is still extremely hard to get, perhaps moreso saw an article recently where every sector is seeing a reduction in IT/devs except for tech and ai companies if your company is in a sector where eng is a cost-center a
68.
by ej88 8mo ago
We're already seeing eye-watering, blistering growth from the new hot applied AI startups and labs Imo the wave of top down 'AI mandates' from incumbent companies is a direct result of the competitive pressure, although it pr
69.
by ej88 8mo ago
what doesn't make sense? "writing" the code is implementation you still need good swes to distinguish if the generated code is good or bad and adjust the agent and plan the system ime opus is smart enough to oneshot medium to
70.
by ej88 8mo ago
I would posit another large factor is "owning" the software comes with the long tail of edge cases, bugs, support, on-call, regulations, etc... that an established SaaS has learned and iterated on for many years with many customer
71.
by ej88 8mo ago
the perimeter of the tasks the LLMs can handle continuously expands at a pretty steady pace a year ago they could easily one shot full stack features in my hobby next.js apps but imploded in my work codebase as of opus 4.6 they can now one
72.
by ej88 8mo ago
"not enough people are emotionally prepared for if it’s not a bubble"
73.
by ej88 8mo ago
i am somewhat worried in the short term about ai job displacement for a subsection of the population for me the 2 main factors are: 1. whether your company's priority is growing or saving - growing companies especially in steep competi
74.
by ej88 8mo ago
ai-first via mandate: throes of a company trying to appear cutting edge ai-first via the actual technology being built: talent magnet despite the HN bubble appearing otherwise
75.
by ej88 8mo ago
the new meta is purchasing rl environments where models can be self-corrected (e.g. a compiler will error) after sft + rlhf ran into diminishing returns. although theres still lots of demand for "real world" data for actually econ
76.
by ej88 8mo ago
i mean llms happy to be corrected but im not aware of any direct improvements llms bring to ultra low latency market making, time to first token is just too high (not including coding agents) from talking to some friends in the space theres
77.
by ej88 8mo ago
oh, i meant copilot as in microsoft copilot in vscode. i havent used zed so can't speak to it but if it works for you it works!
78.
by ej88 8mo ago
copilot is a much worse harness, although recently improvements in base model intelligence have helped it a bit if you can, use cc or codex through your ide instead, oai and anthropic train on their own harnesses, you get better performance
79.
by ej88 8mo ago
1. execs likely have spend commits and pressure from the board about their 'ai strategy', what better way to show we're making progress than stamping on some kpis like # of agents created? 2. most ai adoption is personal. peo
80.
by ej88 8mo ago
seems to be a pattern: [Company that's getting disrupted by AI: Fiverr, Duolingo]: rush to adopt internal AI to cut costs before they get undercut by competition [Company that's orthogonal: Box, Ramp, HFT]: build internal tools to
81.
by ej88 8mo ago
It's not like you need to take a course. The frontier models are the best, just using them and their harnesses and figuring out what works for your use case is the 'investing in learning'.
82.
by ej88 8mo ago
i would disagree with your dismissal of it as theatre. it's not forced, and it would probably be worse if I skipped those meetings :). meeting customers is not the same mental load as focused coding but it's still work that needs
83.
by ej88 8mo ago
fwiw i think there's a balance here at my current company i happen to work 70hrs/week but it doesn't feel like a ton of work, i'm having fun and let's be honest a chunk of the "work" is meetings & hang
84.
by ej88 9mo ago
I would add on that the most of the premium of a modern SWE has always been on understanding problems and systems thinking. LLMs raise the floor and the ceiling, to where the vast majority of it will now be on systems and relationships
85.
by ej88 9mo ago
the specific tasks (i.e. writing code) might disappear but the actual work of constructing reliable systems from vague user requirements with an essentially unbounded resource (software) will exist
86.
by ej88 9mo ago
I'm quite convinced that software (and, more broadly, implementing the systems and abstractions) seems to have virtually unlimited demand. AI raises the ceiling and broadens software's reach even further as problems that previousl
87.
by ej88 9mo ago
I would rather learn how to use these tools effectively now and ship more and of higher quality now. If these tools improve to the point where anyone can pick it up - that's great! I enjoyed my head start while it lasted. If these tool
88.
by ej88 9mo ago
Most of the gains come from post-training RL, not pre-training (OpenAI's GPT 5.2 is using the same base model as 4o). Also the article seems to be somewhat outdated. 'Model collapse' is not a real issue faced by frontier labs
89.
by ej88 9mo ago
I'm sorry your teammates have skill issues when it comes to using these tools.
90.
by ej88 9mo ago
I primarily find them useful in augmenting my thinking. Grokking new parts of a codebase, discussing tradeoffs back and forth, self-critiques, catching issues with my plan, etc.

More ›