Quick question for people running Browser Use in real workflows: when a run fails, what is hardest to debug - what the agent saw, why it picked an action/tool, prompt regressions, cost/time blowups, or reproducing the failed run later? I am researching failure replay for browser/desktop agents and real failure patterns would help.
#dev-chat
229 messages · Page 1 of 1 (latest)
anyone got a good way to get audio rings/bells when a prompt is done or waiting for human response?
(using CLI)
codex jus tkeeps timing out, or not responding at all, im barley using my quota and have 10 jobs queued for 3 hours.
You have shut it down and restarted right lol?
yes
Yeah, easiest way is usually to just hook into your shell rather than the tool itself.
If you are on mac you can do something like:
afplay /System/Library/Sounds/Glass.aiff
On linux:
paplay /usr/share/sounds/freedesktop/stereo/complete.oga
What setup are you running on?
Mac os, I'll try that
@rich forge @distant sphinx
?
are you fizz the web dev
yeah, where do you know me from
he wasn't talking to you bud.
he pinged me ☠️
He pinged me.
15 years old bro go study for your gcses
Lmfao
wow
Maybe i'll just shorten it to Redis
ayo
Web dev joke.
db
@lunar herald
slide me that .env 😉
sorry but I already have a envfriend
I'll be sure to invalidate your cache
left is you, right is the REAL OFFICIAL VERIFIED ONE AND ONLY FIZZDEV
created his discord 3 YEARS before you did
Have I been here before chatgpt was released
Damn
uncdev
Age isn't everything.
What does @near ginkgo say about this discovery?
No
Yes
prove it
My word is everything
We cannot rely on the source of truth ( discord dB ( since it was accidentally deleted
accidentally eh
that’s convenient
Finger slipped ah
also if you say your account is older then him… how old was the first account?
It was just a bebeh
when did you create said account
I didn't
Frank did.
Frank missed his friends and made discord
💀
don’t use
i use gitlab
better
a lot better
a considerable amount better
I run a tech company. We communicate with some clients via discord/slack, so we built a discord/slack bot that integrates with Github. Also we get a lot of technical questions from project managers who aren't familiar with the application.
The bot uses the client's code repository as a knowledge base and can answer very technical questions. The client or project managers can ask technical questions about the code.
In one picture I am asking it about a systemd process that syncs a database with an Azure data source. Our team emphasizes documentation, so the agent knows the name of the systemd process. The agent can answer questions.
In that second image I am asking it about the training data for a machine-learning model we built for a client. It has access to detailed info about the training data from the documentation we include in the github repo.
Who won the dev challenge?
How do I reproduce the way gpt-image-2 works inside chatgpt ? It seems that I can not have result that exploit the conversation context in a similar way when using the API
whats better Ai tool can be used to Generate Videos, For facebook reels? Of all the Ai tools Behide a PayWall.
Is anyone here participating in the SEA x Openai hackathon 2026? I'm looking for a team in SEA to participate in the competition.
Can anyone help here? This has been my life with codex app for a week now. I’ve logged out and back in and reconnected on my Mac but can’t reconnect on the app.
xd
hey @tight prawn any thoughts on clawpatch?
sounds interesting, but Im not a big fan of tools that further blind sight the user from the code
I don't want the AI black box to make an unsupervised black box of code.. which is exactly what a "fix loop" causes
the fundamental problem with all of those systems is that they ultimately evaluate the success rate based on "does the code run?" or other easy metrics like speed.. because obviously.. evaluating security and maintainability is much more expensive and waaaaaay harder to evaluate reliably
I like the effort tho, but I think it is still way cheaper to just load the repository on Codex and analyze with Codex Security
I see thank you! Besides this and of course user knowledge would you recommend anything else?
tbh, I think AI coding tools are as good as they need to be
it is already past the diminishing returns.. just like any tool.. there are cheap ones and expensive ones.. the expensive ones can be better.. but it always comes down to the skill of the wielder
This being said, do I even need to ask about gsd/gsd2?
we are at a point where putting a larger and slower model i nthe hands of the average vibecoder is simply a waste of tokens..
they will not benefit much from the "it scores 2% more in benchmarks"
idk if this is a hot or cold take xD

I'm no expert, but feels like newer gens should focus on speed and efficiency
oh, I agree
the sub-agent thing is a good hack to get more speed.. but I still would want to see a breakthrough that manages to get more raw speed from a single model
IMO, this is about how to get the AI to do more.. how to fit more stuff in the context.. which is good..
but I very much would like a tool that would figure the problem of allowing the user to do more creative decisions more often instead of offloading decision making to the AI, gives it more opportunities to make bad code
and then.. keep building over bad code
a developer usually makes better strategic decisions when asked.. so.. the AI should ask more instead of deciding things by itself
Exactly. More speed and larger context help, but they do not fix silent bad decisions.
For coding, the model should ask at the points where the choice affects architecture, data flow, API shape, state, or long term maintainability.
The problem is not only that models make mistakes. The problem is that they often make strategic decisions quietly, then keep building on top of them.
A better coding AI should not just produce more code faster. It should know when to stop, expose the decision, and let the developer choose.
Speed matters. But decision discipline matters more.
so is anyone today getting ..... codex startng the job and then getting stuck on a thinking endless loop
local project, it takes the questions says its going to inspect things and then just idles been stuck for 2 hours on 3 coding jobs
tried 5.5, 5.4 and 5.3 spark
a lot of peeps on X reporting the same
this can happen sometimes. less common on chat gpt but still experienced it
actually upgrading powershell from 5.x to 7.1 solved the issue
windows powershell that is, something changed in the last codex update
It’s done this everyday for months nothing new
@tight prawn speaking of gsd
rugged $500k and deleted all of his accs
indeed, goal feature basically covers it all
he didnt have to f up people while leaving tho...
what happened?
¯_(ツ)_/¯
contributor too btw
daily reminder to run away from anything thats even tangible with crypto
Don't do it. Trust me. Everyone will hate you. They will say you are taking their jobs and torment you. Then others will ask you to make them Ai girlfriends.
Hey
Oh okie
anyone know if GPT-5.5 still reserves 128kT for assistant role? or if there is documentation about that somewhere?
Hello!
Hi everyone
does anyone knows here how to check no of tokens spends on each task within a sjngle chat or project?
Hi guyss
Im panicking what is wrong with codex weekly limita I thought they fixed them!? I did a 2m task and my weekly limit went from 65% to 45%?!
"task"
?
i mean a task with time is vague measurement as you can use a lot with one. weekly thing was fixed but there is still lagged updated in it for some people meaning a task from earlier time could have hit the same time. btw #codex-discussions for getting more feedback to the issue
guys could any one support my friends vibecoded ai study app
ill give the website link
he mostly resides in gdc tho
lmao
Get the latest Windows installer for ONCard, an open-source offline AI learning platform.
its compleatly free also
also why is GPT-5.5 slow as molasses on the API right now?
despite priority and all
seeing response times > 15 seconds for 10000 tokens in and a few dozen tokens out 💀
I applied to a few jobs at OpenAi. Always get rejected, and that's perfectly ok. The success I have had up until this point came from within me, not from proximity to a big company. It makes it more special.
Though I do want to join the team, I may be too rebellious. I wear shoes of different colors and sometimes not even my pant legs match.
Here's my application:
Hi! I am Samuel Cavazos. I currently work as a data scientist at Alshival.Ai. We architect custom machine-learning models and agentic solutions for our clients. From voice agents for reminding patients it is time for their routine mammogram to scheduling assistance. My favorite agents to work on are those that generate data visualizations from an organization's data. I teach the agents how to write the SQL and interact with the data, then have it generate visualizations for my clients.
This year I placed in Unconscious 2026 Top 100 Data & Analytics Professionals, alongside data scientists from Meta, Intel, and other big tech companies. I was honored and proud to receive this award because I suffer from schizophrenia. It was a sign I had overcome it, somewhat. I still talk to spirits though (that is what I call them). I have made peace with them.
But Alshival.Ai is small. We are a team of 4. Yet still big clients such as Broward Health, Miami-Dade County, Starwood Hotels, and other seek our help in developing their Ai infrastructure. I have been using OpenAi since 2021. I would say I am loyal to the brand.
I built GRISP — an AI‑native programming language based on graph rewriting (v1.0 released today): https://github.com/GrispFoundation/Grisp
GG @round shore, you just advanced to level 43!
Good evening, everyone! I’d like to ask for your opinion. I’m working on a project that does manga transcription, and I wanted to know if there’s any model or API that could improve the transcription quality.
Multi-Latent Attention is what I am learning about right now. I am still stuck in 2017 using traditional attention in my models. Still works great for most models, but sometimes you want to squeeze as much juice as you can into a smaller parameter set, which is where things like grouped query attention and multi-latent attention come into play.
ClosedAI
Hugely closed
feeling lost everyday
AI is so powerful now
Is it necessary for me to complete my computer science degree
Even if I complete it, I might not be able to find a job
Yes. It's important to know enough to call out AI when it gives you horrible code, but that is just the bare minimum. People will need developers that will also know how to leverage AI in ways that free up time and effort for more complex tasks
Just make sure you direct your learning toward that future instead of relying on the guidance of people who aren't in industry
Ultimately, you are responsible for your own education, not the school.
But while you learn the technical angle, be sure to learn how to communicate complex topics in a way that is easy to understand and be able to explain your process behind the your solutions. Ultimately this is what will get you a job
I needed advice on how to go about summer internships. I have a decent background in CS+MBA in top institutions. I worked in big tech and even completely took a startup into finding building their whole MVP. What am I doing wrong in this market?
i think there is person typing very fast behind AI and AI is not real cuz conputer cant think
Nah.... bro this is the most crazy stuff have ever heard😭
Is using something like codex-lb against the TOS?
What's your country?
does go plan have more codex usage than free?
This is what I'm anxious about
Every time I scroll through social media, I see all sorts of tech company layoff
and thousands of thousands people just lost their job
I genuinely feel that my technical skills, amplified by AI, have surpassed those of developers with 1-3 years of experience from 10 years ago, yet I still can't find any job
I wrote a website for a startup company, using Go's Casdoor library and Gin, that integrated registration and payment with their business. It was hosted on a cloud server, and I even paid out of my own pocket to buy a bunch of cloud servers, spending thousands of dollars, using a distributed system and doing stress testing
During stress testing, this website was able to handle 6-figure QPS of requests simultaneously without crashing
Furthermore, by continuing to add funds, they can continue to expand horizontally
These things, these technologies, were something that even developers with 3-5 years of experience couldn't achieve in the era before AI
However, with the amplification provided by AI, I might be able to understand and figure it out within a few weeks
This is something I never dared to imagine before
I basically did the work of an entire team in two weeks, a task that would have taken several months, all by myself
There's more I can tell
I used my own writing corpus, a tool from hugging faces, and 8B qwen as the base model to train and run an AI that speaks like myself on my 3070 graphics card used for gaming
This is something I couldn't have done at all in the pre-AI era
It's clear to the naked eye that the entire industry no longer needs so many people
I'm completely lost, I have no too much hope for the future
Hello AI Developers 👋
nuguard is the opensource repo designed to help you with agent validation (happy-path, red-team testing). It supports a wide-range of agentic frameworks including OpenAI's agent-sdk.
✨ Check out various capabilities at https://nuguardai.github.io/nuguard
Drop us a note at Discussions on our github repo for any feedback.
NuGuard by NuGuard AI
End-to-end security for agentic AI applications. AI-SBOM generation, static analysis, cognitive policy enforcement, and SBOM-driven red-teaming — in one open-source pipeline.
The industry clearly doesn't need that many people anymore, so what should I do
Many people with more impressive resumes than me can't find jobs, and many people with more impressive work experience than me also can't find jobs
Statistically, AI will indeed generate terrible code at times, but you can isolate and restrict that bad code to a very narrow scope through modular testing methods, allowing the AI to then self-correct it.
As AI tools and AI itself advance, this problem will soon cease to be a problem at all.
Network and socialize. Most jobs happen this way. And don't compare yourself to other people.
Best of luck.
Most people in my network are unable to find jobs, and partly for that reason, they've stopped pursuing degrees
I also tried contacting those who were employed, and I submitted applications to their companies using their references, but there was still no response.
I am currently in a hopeless and utterly lost state
The job market is many places is an absolute joke, AI is only going to make it worse. We are in another automation era but with far more advanced tech than before with the potential for more displacement. And the job market being bad isn’t limited to technology based fields. Most job listings by real companies on their sites are fake, used to either gauge metrics, data mine, or for openings years down the line, or in some cases to abuse government tax credits.
It’s not you so don’t feel bad, there are so many people even far more advanced in their career who are struggling to find anything
Unfortunately no work doesn’t pay bills and many landlords or banks really don’t care if you sleep on the sidewalk so you feel lots of extreme pressure nowadays when looking for work
Networking and connections don’t mean anything. You need to actually mean something to those people for most to even look at you for offering a job, and then your resume gets sent off somewhere to someone who has no idea who you are
I feel bad for college grads, I really do
Almost every company and government agency has been laying off in the US here
The best advice I can offer is keep trying and try applying for jobs you think you couldn’t get and things you wouldn’t normally do
Someone will eventually give you a job offer
Contacting recruiters sometimes works too, especially external recruiting agencies. But they’re not always the best jobs they can help you get.
Sometimes friends of friends through your friend will give you jobs or help you get them too, but that’s not always the case or applicable for everyone
And I mean like actual friends, not somebody you just know or have worked with
I'm pretty deep with ai rn, i don't have a job, im actually doing a graphics design degree, if anything this time we have rn is more precious than anything for gaining them skills which get us that killer job in the future, its all about stay discipline and pivvioting to a future we know we can grab. applying for a crap ton of jobs too haha.
I built an interactive 3D Brain Connectome that maps neurotransmitter signal routing and brain waves
Hey everyone!
As an AI student focused on Brain-Computer Interfaces, I’ve always been frustrated by static textbook diagrams of the brain. I wanted to see how the brain computes—how sub-regions connect, how waves originate, and how a neural spike cascades across the cortex. So, I built NeuroVis 3D, an open-source, interactive 3D brain atlas and functional connectome map.
You can check out the repo here: https://github.com/AayeshaBibi/NeuroVis-3D
You can check out the Live demo here: https://www.linkedin.com/posts/ayesha-bibi-8991b3319_neuroscience-braincomputerinterface-threejs-ugcPost-7464198158974296065-KTPE/?utm_source=share&utm_medium=member_desktop&rcm=ACoAAFCnMDUB5w--9OObQunIUKNJEfShK5WNuA0
What makes it different from a standard 3D model:
🔍 Deep Hierarchical Navigation: Click to fly from the Cerebrum → Limbic System → Hippocampus → down to the CA1/Dentate Gyrus micro-regions. Every node has metadata (neuron counts, activity %, functions).
⚡ Dynamic Signal Routing: Toggle "Signals" to watch animated pulse dots traveling along white matter tracts (using CatmullRomCurve3). It maps Glutamatergic, GABAergic, Dopaminergic, Serotonergic, and Noradrenergic pathways in real-time.
🌊 Brain Wave Mapping: See where Delta, Theta, Alpha, Beta, and Gamma waves originate, their frequencies, and which sub-parts consume them.
🎨 View Modes: Exterior, Cutaway (transparent cortex to see deep structures), X-Ray (wireframe), and Live Signals.
Would love your feedback, critiques, or ideas for what to add next!
LinkedIn
The human brain isn't just an organ; it's the ultimate 3D network. And finally, we can visualize it in real-time. 🧠⚡
As an AI student currently focused on Brain-Computer Interfaces and neural engineering, I’ve always wanted to see how the brain computes—how sub-regions connect, how waves originate, or how a neural spike cascades across ...
@last void
Been experimenting a lot lately with autonomous AI agent systems running continuously across multiple tools/models. One thing I’m noticing is the technical side is actually becoming easier than making the agents economically sustainable long-term.
Curious how other builders here think about:
- persistent memory
- agent coordination
- infrastructure costs
- reliability over long-running workflows
Feels like the next big challenge is making agents operate continuously in a practical way outside demos.
i said this to ai...........firstly arrange all english words in 1 book..........and later scramble all words and arrange it to many different books .......and keep reading it...............in the same way assume colours and make all arts................................keep reading it
lol. My personal github account just got removed the github repo of the external agency who effectively took over my job 18 months ago. Can't rush these things
WE OFFER POWERFUL GAME HOSTING AT A FAIR PRICE
Hello, we provide game server hosting for Minecraft, FiveM, and potentially other servers upon request. Everything we offer runs through Pterodactyl, so you can easily manage your server yourself without any unnecessary complications.
24/7 uptime
great value for top performance
Pterodactyl panel
Minecraft / FiveM / other games
quick communication in DMs
help with basic setup
fair approach — no unnecessary talk
Unlike others, we do not want to sell overpriced hosting. We want to offer something that actually makes sense: solid performance, fair prices, and quick replies.
Capacity is limited, so if you are interested in a better game server for a good price, message me and we can work something out.
Ayy! Also with a GD background (graduated in 2025)
hi
Whats the chances ai takes over computer science
Im deciding between computer engineering or science for college
0%
Will it change CS? Sure like every tool and industry demand out there.
But what do you mean "take it over"? Are you saying rendering it obsolete?
Your best chance is to do exactly what you love because whatever learning curve exists you'll be able to endure far longer than someone who hates their field.
Love CS? Do that. You'll already have an edge
CS is more coding correct, I suck at coding but ai is also really shtty at it
I mean, do you enjoy coding?
Is it something you can do for 40 years?
It's also about architecture, security, accessibility, debugging, knowing when to call BS on AI but also know how to use it to improve your workflow
CS is a lot of things
It's also a lot of talking to stakeholders
Plus giving consultations
I would say that how you build relationships across teams and departments are going to matter a great deal, sometimes more than actually coding
The hardest part for me was getting out of the weeds and being able to explain our tasks to a PM on a weekly basis
In addition to school I'd seriously recommend learning public speaking, like going to a Toastmasters Meetup
It deadass saved my bacon a few times
Like, you can be super great at coding but still lose your job because of horrible communication and interpersonal skills. They'll care more about how genuinely cool you are to work with (that's very rare) and less about how good of a dev you are, which there are plenty out there who fit that category
Anyway, hope this helps. Things I learned as a new grad that helped me get a job and get promoted. Just be that person teams WANT in the room.
Has anyone here used aikompute? it says they give gpt 5.5 api for free but i dont know if it is really that good or if there is some catch
sorry if its not the right channel btw
Have you coded rasperry pis before, i really need some help with a project, ai is terrible at code and i am no good at it
yeah me too until recently
it seems too good to be true because i used to use lmarena for free usage but they removed opus sadly 🙁
but this has opus somehow idk
and not even on gui its api
Yeah it doesn't pass the vibe check for me either. Be careful tho
I have a teen at home who wants to do IT for study, this is interesting times. He's probably going to stay home forever doing random stuff, fulfilling the nerd-in-mother's-basemet stereotype
gitgud I'll say. AI is awesome at code but only if you tell i exactly what to do and how. Or use it to tell itself. At some point you need to learn what's up or you will never get out out the "fix it plz" valley of despair
Anyone knows how long it takes now to get your connector app approved in ChatGTP Apps ? It's been a while I've been looking at the "Review" status.
You are in an era where the internet is effectively born again. Over time documentation was getting published, now people are doing the same to train models. It makes information more accessible and should be leveraged. If anything it will create jobs as well as ruin some. It's an innate part of the society you live in dealing with innovation and fluctuations in the job market.
Also on a side note, ill be publishing a platform for helping people to find jobs in a little while after getting the site up.
I agree. We are in the transitional phase where tools quality ranges from totally useless to amazing. The skills map gets overhauled. Companies are in disarray not knowing or understanding what is happening to the markets. Recruiters have no idea what to look for, hiring managers have no clue what to plan for. It sucks, but it will get better. And yes there will be much more jobs. At the moment there is a big need for contractors fixing AI code and if you're good at programming it's a good angle. Emergency fixing vibe coded slop can go up to 1000 US / h if you have credentials. experience and patience for it. Just ask any B2B CEO how much they are loosing per hour of downtime. Not saying it's for everyone though.
Why codex signed out my account all of sudden today wth
it asked for my phone number in attempt to login back to codex, doesn't works even waited for many hours
damn rip can't login to codex
free account?
I completely agree. The real value right now is not just generating code with AI, but knowing how to clean it up, debug it, secure it, and make it production-ready.
/goal not working ?
it's working for me right now
no paid plus, this just logged me out today?
i literally cannot login/accest codex even with another number (i got dual simcard)
region (+62 indonesia) doesn't work at all, even my brother number won't work, got the same message..
I think you need to reach out to support at https://help.openai.com
Already done that. Tried relogin using my phone number.. the weird part forgot password doesn't work it says username not found 😭
what have support had to say about it?