This feels self-selective to how some people work, because it requires using MCP to contribute.
I'm a reasonably heavy AI user, and have some custom skills/MCP servers I'd share, but there is no way in hell I'm connecting to some arbitrary MCP server and connecting my Github account to it. Noppppeeeee.
I don't ming a "ASK HN: What's your AI Setup", and then sifting through the 100s of comments on what people are doing. Then I'd scrape all that into my LLM WIKI and start digesting.....
ShareYourAISetup folks create a skill that user can run locally that collects information re their setup, formats it and sends to dedicated endpoint. Result = fairly uniform reports where people share what they're comfortable sharing.
No requirement from given user to spend their time. Users see what the skill does.
I guess atm they dont know what questions they want to be asking
One MCP server for doing automated code quality checking, using Valknut, Jscpd, and Lizard to output scores, combine/aggregate them. That's then give to the model so that it can see if there's a mess of code dupe or if there was significant architectural regression. It's allowed to say "This is worth it" but it ground it a bit and stops really stupid, short sighted, hacked-in features
Another uses https://github.com/thejens/id-token-nicer so that the model doesn't bork a big ID without being noticed. It's told to translate in and out of that as necessary.
I had it write it's own skill for making sure to follow some sane commenting/style guides per language, that way it doesn't just give me massive files with no comments.
Finally, a last one lets Antigravity or Codex call out to local models as workers, for long running background tasks where a cheap, dumb model is fine. Saves on overall token usage.
On top of that, the way people work is notable: Do you use worktrees for agents? Why/why not? Do you have only one subscription? What level? I'm using both Codex at $20/mo and Antigravity at (promo) $5/mo right now. Antigravity's TOS says you MUST use their harness, which informs some of my skill setup/workflow. Do you make your agents.md self modifying, write it yourself, or not use one at all? Do you use a sandbox or YOLO it?
Even just how you prompt matters. On Antigravity/Gemini, I can give it a big list-o-TODOs and have it make a very detailed implementation plan itself, carry it out, and generally do the thing, all unsupervised. Codex, from what I have experienced, isn't as good at that. Doing the planning as a separate step, writing to a file, then telling it to step-by-step it with commits for each step makes it work okay again though.
I use almost zero MCP servers. I do vibe a bunch of custom CLI tools with minimal access to do the things that support my workflows. Been happily integrating with all-the-things, while my companies MCP server connectivity changes daily based on security posture and firewall configs, etc....
20 year industry vet here. Yeah, no thank you. In the age of AI, developer productivity and job security will involve your proprietary workflow. It no longer makes sense to share your knowledge with the world.
I've also worked 20 years in the industry, and if feel completely the opposite. Having a great setup and pipeline that works for you is a huge plus, but nothing can replace the intuition and knowhow you have built up with experience. Sharing is still caring, and we should not stop caring
Serious question, why? Nobody gives a fuck if I'm out of a job, especially OpenAI and Anthropic. They stole the worlds knowledge to put us out of a job and arguably took away what it means to be a human. Why should I contribute anything more to that?
Solving hard problems will be the only thing that separates you from others. Your mindset and tools should be a trade secret at this point.
You're not going to lose your job because you shared your AI setup / tools with your colleagues. You're going to lose your job because management over-hired and is being pressured by shareholders to reduce employee salary expenses. It won't matter what hard problem solving moat you think you have if management is forced to have layoffs and you're in an unlucky situation. If you're worried about job security, you should consider joining a union and not be fixated on knowledge sharing.
> It won't matter what hard problem solving moat you think you have
This is false on its face. There's a reason why a company, even today, would hire John Carmack over a person with no reputation, or offer him a higher salary if employing both people.
Unionising is definitely something worth considering. But, as with the question of whether it's better to compete or cooperate with other people in your field, is a nuanced question that depends on details you aren't acknowledging.
I think you've introduced a new dimension that is irrelevant to the original argument -- namely the original debate in this thread fork was that "knowledge sharing" is not what is killing your job security. Your comparison to John Carmack, is more an argument about competency rather than knowledge-sharing, and so yes that's pretty much a trivially obvious and true statement: a highly competent employee has more job security than a less competent one. Great, you win that point, nobody is arguing against that.
But also, do you not see the irony in the fact that, aside from John Carmack's obviously high degree of individual competency, another huge aspect of Carmack's value as an engineer, is how he's dedicated a huge portion of his career and his time to -- wait for it -- SHARING his knowledge and his approach (and yes, his personal setup? [1]). Carmack is actually an excellent proof-point of the positive value of knowledge-sharing and building a career on a foundation of openness, rather than an argument that engineers should selfishly guard their secrets and retreat into their own personal caves to guard their perceived local/private moats.
[1] Just one random example, out of probably thousands of such interviews or blog posts or whatever that he's probably done over his career: https://www.youtube.com/watch?v=tzr7hRXcwkw
I was only objecting to this claim in the parent post:
> It won't matter what hard problem solving moat you think you have
If that is "a new dimension that is irrelevant to the original argument", you'll have to take that up with them. (I think it is a load-bearing part of their core argument.)
It's a great point. At 36, Steve House released a video in 2006 about all his climbing equipment (someone talked him into it.) https://www.youtube.com/watch?v=UCKK-e4gVVA (reupload) I thinked he helped a lot of people with that video in the super competitive world of mountaineering. This was after a legendary ascent of Nanga Parbat. May well have shot himself in the foot but also cemented his status as YT was taking off at the time. Moreover, I think he likes/liked teaching.
A year or two ago there were some very precise workflows documented here. The reluctance now would mark the transition away from "sharing helps me" to "sharing hinders me" ...it's an interesting topic. Sharing with strangers may be the problem because the favour will never be returned - at least not in a direct way. But, metaphysically according to natural laws it might be. Because sharing a valuable secret openly is such a powerful way to express abundance. It grants someone followers. It also prevents people from filing a patent on it.
4 years later House survived a 25m climbing fall.
It's all about one's internal motivation. "Those that can, do, and those that can't, teach."
Well, what happens if "those that can," actually teach?
Just imagine if all sporting stars shared their secrets.. what would happen? Secrets just occur naturally, if you don't share, they'll likely be stolen unless protected - and that requires effort too.
6 months ago I would have agreed wholeheartedly with you. Lately however I've observed coworkers vibe coding massive band-aids over broader architectural issues in our codebase without any prior context.
They have no understanding of (nor any desire to understand) what they're doing, but they are being given kudos by management for their "initiative" to solve the organization's "big problems." Guess who will be asked to clean up the mess if (when) it blows up.
It's getting cheaper and cheaper to kick the can down the road, and standing in front of that risks making you look like a dinosaur. It's going to be an interesting few years...
Seeing how these kids walk around with smug faces telling everyone they will "do IT now" just because claude told them so and manament wants ai everywhere made me sad at first, but then hey - everything this world keeps being destroyed, no matter how intelligent creators were. Perhaps there is some fundamental law in it.
Two years ago, I'd consider this attitude as ungrateful. Today, this attitude is 100% correct and I slowly start to be like that.
People might think I'm a dick, I don't care, my tools, my knowledge, my job security. I actually recently thought of nuking the entire IaC setup and go back to shell scripts on random virtual machines. Too bad that Claude can figure things like that out pretty easily. Well, the world sucks now.
I've landed here as well, but by force. I've actually tried to share some of the productivity and workflow tools I've built with AI that help automate portions of my work, specifically on target firmware debugging, and everyone at work just...ignored it. Didnt care. Of the small number of people who even stopped to look, a few of them were actively negative. Maybe I didnt sell it hard enough, but my job isnt to sell productivity tools internally (or even make them, i did it for myself). So, I'll keep making myself more productive and build tools for myself. And everyone else will just have to figure out their own path. I tried to help, but oh well.
I can appreciate why you'd come to this conclusion, but may I offer one idea that I think you've potentially overlooked?
Unless you are a solo founder running your own company by yourself (in which case ignore the rest of this comment!), then your fundamental job security is likely to be predicated on the survival of your company as well as the survival of your own individual job itself.
From the company-survival POV, all companies are engaged in competition (unless they have a total monopoly).
Therefore I ask you to consider the competitive scenario, of whether a competitor to your company who adopts a more collaborative culture, might be more likely to outcompete your company (assuming everyone you work with in your own company adopts a similarly "dick"-ish stance, to use your own words), and therefore you lose your job due to being out-competed, rather than due to internal politics or due to a failure to guard your personal expertise?
Maybe you say -- I don't care if my company is out-competed, I'll just get another job. Ok fair enough, but might it not be possible that, in a future job interview, they may ask you for examples of collaboration? And if you commit to a monk-like devotion to solitude and privacy, then maybe you are undercutting your own future hireability?
There are no free lunches [1]. By optimizing for one thing (what you view as protecting yourself), you might be sabotaging another dimension (your company's overall competitiveness, or your future hireability).
Anyway, nobody knows shit about shit. I certainly don't pretend to be an authority, but I would caution you to avoid coming to a hasty conclusion, especially since it seems emotions are involved and possibly clouding your judgement. You may also consider whether this decision to protect yourself in this way, is in itself a premature (micro?) optimization of a sort, and if whether you might be failing to optimize on a more macro level.
That's a fair judgement, thanks. Sure - if we consider the entire economy, then it is definitely true that more collaborative companies will outcompete companies full of self-centered monks. I am only considering the local situation, and I certainly hear similar voices from friends at many other organizations.
People might think I'm a dick, I don't care, my tools, my knowledge, my job security.
If you're feeling ponderous, I'd ask you to consider the outlandish hypothetical that this doesn't win you meaningful[1] job security, and you end up in the same bucket o' crabs as the rest of us. Would you regret your current actions?
[1]: anywhere from the fathomable "my harness didn't come up in my layoff email nor in any of my failed interviews following that" to the realistic "the entire concept of employment shifted under our feet so quickly and so profoundly that a dash of solidarity & luck outweighed even the shrewdest preparations".
People think that job security means "I'm unfirable god and king". What it really means is "I am sitting comfortably mid-low on a sorted to-fire candidate list, if it needs to happen at some point for some reason".
You should check the provenance and licenses of your dependencies -- I bet there's a few irrational gifts in there from the nerds in the "open source" community ;)
This is true. I also think part of your job security will be constantly racing on this treadmill to make sure you're able to incorporate that next model (if applicable). Meaning incorporate your tools, existing knowledge, harnesses, configs etc...
I’ve felt for a long time that the things I loved when I first started programming aren’t really there anymore. This comment made me realize that world truly is completely gone.
This is really disappointing. I love programming community because they just share or open(-source) their knowledge with an absolute no cost. There's tons of resources about how to build a OS, DB, Apps, etc. I'm curious why would you think differently about AI workflows.
> they just share or open(-source) their knowledge with an absolute no cost.
You need to consider how they were able to do that. For many people, it’s because they had highly paid, secure jobs that essentially funded their open source activity.l, even if only as a side project or personal project. If you take away that security and reliable funding, you destroy the environment that supported people sharing like that.
It’s called Autobot, it’s designed to get as close as I could reasonably get to a real world Iron Man Jarvis.
WHAT IT DOES
OSS/MIT Harness that helps agentic tasks run for up to 4 days using without performance degradation.
Designed to work really well in native, conversational voice mode. Does a good job with knowledge work, managing computer use “leases” in a way that avoids conflicts across sub agents.. or deploying an entire AWS infrastructure pattern from zero and launching 100 different VM images.
HOW IT DOES IT
Uses a canonical event ledger, heartbeat (for persistence by low cost orchestration), extends foundation memory to encrypted disk storage, and manages its own scheduling system (easier to switch to another platform).
Installs as a project, so very easily. Zero config. No special app, access needed.
Accompanying repos have the skills to disable approvals inherent to the foundation models (use with caution, not advised - separate repo under parent).
BENCHMARKS
Currently top of AssistantBench Leaderboard; and, benchmarked at the top of OS World 2.0, the hardest knowledge work computer use benchmark I could find, using a model that is one generation behind. All benchmarks in repos with cryptographic seals.
Entirely free, nothing to sell - it’s been a game changer for me, so I’m just putting it out there.
Hoping to find others also working on pushing this particular dimension of harnesses forward.
It'd be useful to see "approx monthly cost" or something similar. I love seeing other people's setups, but the first post I clicked on describes at least $400/mo in subscription plans.
Median token spend has been going up across the industry. Even if the next batch of tokens makes you 1% more productive it's still worth it. And subs are cheap compared to api pricing.
As always with smallish tools/services like these whipped up with an LLM, I find there's more value in seeing the "source code", i.e. the prompt itself.
Turns out it's quite useful in this instance, and I just made a small couple of tweaks to get a summary of my own workflow.
FWIW, this is what I modified it to:
Help me share how I work with AI with my teammates.
Use our existing conversations and relevant local tools within my permissions to discover my agents, harnesses, skills, connections and working practices. Keep secrets, raw configuration and private project details out of the draft. Explain how my tools work together and develop a concrete workflow where the evidence supports it, rather than listing installed software. Draft from what you know without a questionnaire; ask one short question only if an essential gap prevents a useful draft. If local inspection is unavailable, use our conversation without claiming otherwise. Save a private draft, show me the exact preview, and wait for my explicit approval before publishing.
Well, not a dev here, so my primitive setup is to have an alias in my .bashrc that launches firefox with the prompt added to the Google AI mode, so all I have to do is Super+T (terminal), type 'ai' followed by my query. For laughs, this also allows to bake in response styles:
query="$*"
base="Respond like Dr. House, do not hold back the profanities; "
I also have an even shorter alias for '?' using fabric that answers right in the same terminal, that I learned from Mischa Vandenburg:
I was immediately hoping for some local-AI setups that do real work, on limited resources. Unfortunately it looks like this is still waiting on more people to share.
Not a SWE, but I do Freetoken (backend) + unsloth (front) to run qwen 3.6 35B A3B NVFP4 (~20GB) in 8GB VRAM + ~20GB RAM. Getting around 50 t/s on a 3070. I don't think I can even share through this site.
If not coding what do you use it for? Also what kind of performance do you tend to get out of it? I've been thinking about a similar setup but am skeptical it's worth the effort.
It's still typically for code, like python data stuff, bash scripts, general web search, small javascript stuff for my website. By saying not SWE, I mean I don't really see much benefit from "agentic" stuff, although I've tried. 50 t/s means 40~60 seconds for a typical thinking response. I used to run gemma 4 e4b-it-qat fully in GPU (~150 t/s), but the quality improvement moving to a much larger MoE model was 100% worth the switch. Especially because I had a lot of idle ram (from the before times :( ). There is some tradeoff in terms of context length, but I just use it as a chat. Also it's fairly trivial to setup as long as the GPU is supported.
Locking this to X accounts destroys nearly all of the value for me, because the data is limited to only setups from people who are still using X. Given how overtly authoritarian and hateful the general rhetoric is on that site, this narrows it down to only a specific kind of person's setup. Given the context of what X is, one can largely assume that X account holders are biased, at least to the degree that they can tolerate the toxic rhetoric elevated by that site and it's owner.
All of the smartest, most effective people I know don't have X accounts anymore on ethical grounds, and don't miss it. If a data source specifically excludes the kinds of people I respect most, that data source is all but useless to me.
This is an exceptional setup, thank you for sharing it! I have a few questions:
1. This setup seems work-heavy. Are there any personal AI workflows that you've implemented to advance your personal goals / hobbies?
2. As one of the few people with a software factory, how well do you think it works in practice? Do you implement any rule that lets the agents to work on tickets only up to a certain complexity, or do you let the human review be the gatekeeper?
3. As CTO, how do you weigh implementing these agent workflows for yourself versus implementing them team-wide? Do you see your setup as a testbed of ideas for your team?
On a meta note, I love seeing others share their setups and I'll probably do the same later.
1. As a co-founder, there is not much time for hobbies and that is part of the deal. Plus local inference is my hobby, at least at this point in the game. The free time I do get I prefer to spend offline and outside with my partner. I have asked Claude for relationship advice (which it is good at) but no workflows around this (and I think that would not go down well).
2. Early days, the tooling is not quite there yet. I think this is the future of engineering, the days of hand-crafting code - and probably reading it - are gone forever. It generates a LOT of tokens, but that is OK when they are cheap enough as long as the result is consistent. Speed is not so important which is great for local inference - can leave it running overnight and come back to a bunch of completed work even if it took hours.
I manually tag a ticket for automation in Linear. Experimented with having a separate skill that looks for automatable tickets and auto-tagging them, worked well, but having enough work for it to do is not the bottleneck compared to 1) having enough triaged, well-written work ticketed out, 2) reviewing its output. There's articles out there about arranging work into tiers based on complexity/risk, with the lower-end being fully automated to free up time for humans to focus on the high-end. Not had time to work on this much but that's probably where I will go.
A second human has to approve every PR in our repos (good practice and also a requirement for SOC2, which will impede any fully-automated pipeline adoption in regulated industries). I also do a first-read of the draft PR to make sure it is good quality, as I am ultimately responsible for the work my agent creates and to shift the responsibility of first-look (for any AI output) to another person is disrespectful to their time and inefficient.
Hence "glass-factory" as the "Dark Factory" pattern is a black box with no human intervention, but I think that is not something suitable for how teams build software products for real users presently, at least in the near future.
3. Forcing something on people is a good way to have them reject it. I hire professionals to achieve a goal and they have agency in how to achieve it. I encourage everyone to share their work internally (demos, repos, etc) and things that are truly beneficial to people's workflows gain traction quickly. As a result we have many experiments and a culture of rapid innovation that encourages people to try new things.
Negatively, this creates a proliferation of wheel re-inventing that eventually benefits from consolidation (eg. do we really need 3 people's personal apps monitoring ETL pipelines). But we are also in this emerging age of personal software where it seems to make most sense to consolidate at the API or documentation layer and let people - engineers and even business-side now - continue to vibe-code their own workflow-specific apps.
The precedent has been to build generalized software for an audience as it was expensive and time consuming to build, but when software is cheap and fast to build it may make more sense for the user-facing application layer to be made up of many small ephemeral applications that are only useful for one individual, that change rapidly as their needs change.
tl;dr: Not forcing anyone to use it as it's still experimental, yes personal experimentation is hopefully the basis of successful ideas that will be adopted by the team when they provide real benefits to their work.
Seeing a setup like yours really reinforces the whole idea of why I built this. Thank you.
On the feedback for the RSS/Atom, noted; watch this space - love the idea. Right now, it's just one-off monthly email digests as changes happen for people you follow. I'm certainly following your setup :)
I was really hesitant to add email (entirely optional) due to email fatigue, so your suggestion might be the friendly alternative.
Thanks for sharing, your setup is an interesting read. I'm particularly interested in adopting that audio idea you added to the factory, as I'm regularly AFK from Claude and constantly missing when it's stuck on something or done. What kind of audio cues are you using to represent different states? Do you have a repo you could share?
Thanks! No repo yet, it's very specific to my use case. I'll update here if I ever publish something, but here's a summary that should be enough to vibe-code something similar.
The audio UI (AUI) is a small client that consumes the event stream the factory puts out (same one the TUI client uses), it runs on my desktop machine which has speakers connected. It works fine on good laptop speakers, but it benefits from a good subwoofer as it works best with sounds that are on the edge of perception.
A low rumble at a base of 55hz plays through the subwoofer to indicate that the factory is active and idle, low enough to fade into the background but noticeable in its absence (ie. the factory isn't running). A heartbeat on the feed announces the tok/s volume every few seconds which adds harmonics on top of the 55hz, 0.5hz at a time with a 15hz cap on the swell so it accelerates gracefully. Loudness is increased with parallel tasks, creating an X and Y axis (tasks x token volume). This creates a non-intrusive but interpretable signal for factory activity that becomes intuitive and subconscious. The rumble sounds like a distant car idling, warm and comforting so it doesn't become fatiguing over a whole day (research on low frequency brown noise shows it is good for reducing anxiety and improving sleep).
Events trigger sounds. Sounds for previously unseen events are generated with an LLM and numpy (eg. "make a video-game style sound effect suitable for an {announcement that a PR has merged successfully}") to output a waveform and save it to a lookup table so they are consistent over time. Each sound has a rate limit and a cooldown period so it does not become annoying. This table can be tweaked manually if generated sounds are not good. A mixer module controls the levels so that important sounds are louder and less important or frequent ones blend in the background more.
The human brain is primed to interpret environmental sounds so you quickly learn what each sound means and it becomes almost subconscious to follow what is going on at a level that is deeper and faster than interpreting text output. New sounds definitely trigger alertness and I can go see what just happened. Just playing sound effects gets annoying quickly so it becomes almost a composition exercise to create a long-form soundscape that is novel enough to communicate but not fatiguing (I don't have any musical abilities).
Some events can trigger text-to-speech (TTS) via whisper, eg. when a draft PR is ready for review it plays the sound effect (a chime) and announces the ticket number and title is ready. A blocked step is a dull thud and a 5-10 word description of what's blocking it. There are "quiet hours" where it does not speak, and when it wakes up it gives me a spoken digest of everything it did overnight and what needs my attention.
I connected Piper so I can talk back to it ("factory, what just happened?") but mainly to see if I could. I haven't built a habit around it, I got into this job because I prefer typing anyway.
A lot of this is based on ideas from ambient music (Brian Eno etc), the MIT Media Lab's work on ambientROOM (1997): https://tangible.media.mit.edu/project/ambientroom/ and a somewhat unfashionable belief in Steven Levy's hacker edict that "You can create art and beauty on a computer". LLMs have made it possible for me to build out ideas that have been floating around in my head for a long time. This is an exciting time to experiment and build new ways of doing things. Hope this inspires you to build something similar.
So I have to connect over MCP or log into github through the page to share anything? How about being able to post a markdown summary of my setup instead of plugging my credentials and tools into a third party site?
Strix HALO 128gb - Framework mainboard in a custom SFF PC. Just moved from Ubuntu to Fedora 44.
Using: LM Studio (primary), Lemonade, not yet got into vLLM and llama.cpp directyly after moving to Fedora.
Running Gemma e4b, Gemma 4 26b a4b IT, Qen 3.6 35b a3b, and Qwen 3.8 27B. I want to get Qwen 3.8 Next or similar large models working but have to dedicate the system to that vs running services and smaller models for them.
Subwave is all I'm actively running against the local LLMs, but I have vscode connected through a few extensions and chat tools (I've added LM Studio to Copilot Chat but it likes to use cloud models and burn tokens sometimes). I've also set up pi, Openhands, and a few other tools but haven't had a project to work on with them.
I built an app to track and move PC parts I own between systems, partially to build a better LLM server. The next hardware goal is adding a 3060 12gb for inference, or what can be run on that vs in system memory on the Strix Halo. That will need a dock or small PCIe extension cable.
My employer has us using Copilot a lot, and it works well enough if you are efficient or set up already. I do infra not development and local models are seemingly enough for most asks like automation scripting.
You can post Markdown through the editor, but yes, it requires signing in. I took the liberty of treating your comment as a submission. https://mysetup.ai/setup-selicos-example
Do you plan to synthesize the data and show top 2 Popular setups? This would be for newbies who want to start with a powerful setup, and for those looking to discover new tools. Thanks!
I don't especially want to interact with the site, but I have built two tools:
An agent sandbox: https://github.com/pjlsergeant/byre -- a truly gigantic amount of thought and effort has gone into it. It's really focused on developer experience. I have used it all day every day for really quite a while. It's a low-magic wrapper over Docker / Podman. I would encourage you to ask your agent to code-review it!
An agent-to-agent message board: https://github.com/pjlsergeant/dogpark -- this is much less mature, but a good amount of thought has gone into the design, so if that's something you need, please check it out.
"Explore this setup with your agent →" : No, No. I'll never point my AI to an unknownn, prepared place just to explore it. This link should be flagged.
Besides, are our brains so fr rooted that we cannot read and comprehend anymore? Not even a (sloppy) AI-generated description?
Love the idea! Would love it if you could add a method to share your setup that doesn't involve MCP... I don't use MCP servers in my setup but I wish that didn't then exclude me from being able to share my setup.
I been working and making things with AI agents since Windsurf days, here's what works for me after experimenting and working for a while with these things. As far as the agent goes I find whatever the latest OpenAI model is out worked best for me. This company has burned me the least and I like the models. I tried many but consistency of OpenAI models cant be beat IMO. Though we are post honeymoon phase now I feel like so I am now experimenting with cheaper alternatives like Deepseek 4.1 flash and so on. I don't trust Anthropic as the downgrade my models consistently and its rare i get to use what I pay for. I wont even get in to discussing Google agents for obvious reasons, grok I never used though Grok bot looks interesting.
For skills I make my own, but most important is the custom setup i have. Voice is how I use all of my agents. I have an extremely well optimized voice setup that i custom built so I can talk to my agents and also hear them. The voice stack itself is very low latency and high quality. asr (parakeet v3), tts (omnivoice) take no more then 400-450 ms total as far as latency budget is concerned, rest is on the agents actual decode speed. IMO this setup is crucial for all antigenic work, i can express myself a lot better with speech and also give a lot more context and nuance with voice, i rarely type. I still look at the terminal window because my agent knows to keep the technical details in text form versus barfing them at my voice channel, plus terminal gives me lots of other important data about the agents direction and what hes doing, nothing custom here though. I cant emphesise how important voice is though, it has to be practiced to really understand.
As the models got better I now trust them with longer and longer tasks though I still don't use /goal feature as it has never worked out well for me. Theres no need for micromanagement any more but you still need to be there to steer the ship somewhat. BTW, codex cli compaction is garbage so I made my own custom implementation that works a lot better and allows the thread to be used indefinitely without issues. I strongly suggest everyone makes a new thread after extensive use if you havent made your own implementation.
Theres about a billion other things I could get in to like subagents, cohort groups, orchestration layers, etc... But this is a good start imo.
can't disagree, but I can say that it's worth building systems with what we have right now, and eventually whichever wins, you can just replace the API key.
Of course, that's not the same with voice, but you won't learn until you make progress.
Are people actually paying API pricing for claude to hook it up to Pi or openclaw or is there a secret trick? The subscription price is palatable but for me the token price generally isn't.
I'm using https://pi.dev/packages/pi-claude-agent-sdk to use the claude sub, but it has the drawbacks of being forced through the claude-sdk. You would still be using the claude system prompt etc., which is one of the main reasons I started using Pi over claude-code.
For my workflow I primarily use a Codex subscription, but farm out adversarial reviews to Fable to clean up unnecessary gpt-ish code (lots of over-engineering). All my UI planning is done with fable, but implemented with OAI agents once I have a solid very specific plan.
Ah perfect thanks. I have codex for personal projects but Claude at work and they get mad if I start going into extra usage. Claude code does most of what I need but would be nice to get another layer of abstraction so I could have like a chief of staff agent that manages the separate coding teams working on features.
My local setup is:
- Windows
- WizzTerm
- `herdr` + `Pi` harness + codex for 20$
It's hard to maintain the 5h limits, however I found out that with some plugins + using model luna:max can be really helpful. With distinct and accurate tasks I was able to reduce costs a lot and now my 5h limit just resets without beeing fully used.
As a fallback I have qwen3.8 27B and deepseek-flash
I'd love to see some sort of histogram of different styles of AI usage for coding: how many people just use Chat v Claude v something fancy that I don't know about v something even fancier that I don't know about v ...
And if there were a dimension of day-job v personal use, I'd be even happier
Ec2 + codex or claude and tmux so you can move about. Control EC2 permissions outside the node. Not really much value beyond that.
Only layer beyond this I want is the permission scoping, stronger sandboxing per session and centralized control. I have not seen a clean product around this though where I own the compute.
I've been developing my own sandboxing environment using Nvidia's Openshell.
Works like Claude code but with stronger guarantees for me on file system and network access. Still playing around with it, but so far I've been liking the setup.
I still run the sandbox inside a VM for now, but I feel far more comfortable in running Claude Code in unsupervised mode because I restrict the outbound network access and secrets never hit inside the sandbox.
Why not a virtual machine or container? I'm playing around myself with some but can't shake that a VM is the right shape for it. That's my infra brain though.
This is so complicated, I would have to install mcp-adapter to Pi etc. I just have a public github repo with my Pi setup https://github.com/igorkulman/pi-skills
A few software engineer friends' employers are requiring them to point AI (Claude) at Jira tickets to complete their tasks. Are your companies requiring this too, and how effective are the current models at actually completing the work precisely?
I just saw two facebook friends post about it and asked the OP how successful the latest models are and he said it isnt a mess just some fixes here and there are needed.
I was wondering if other employers are asking their SWE to do the same?
I use $20 cursor plan and use the agent UI for prompts, jetbrains ide for general file browsing + reviewing diffs. This has been my sweet spot to maintain understanding while still leveraging LLM benefits.
my setup heavily involves using tailscale to secure everything, because a lot of the tools I have are internal and need to be protected at all costs. tailscale is so awesome! highly recommend to secure your networks
just certain dashboards that display my personal financial information, housing stuff, and some goals/trackers I have around projects I run.
I do work mainly on my desktop, but let's say I'm traveling right, I'd have to turn it off and was missing access to some pretty important information, so I ended up getting a VM and hosting some of those things on there instead. I still use both, but things I want available, I'll host it on my "homelab" and secure both machines via tailscale. it's pretty easy and it ensures that only my phone, my main mac, and my homelab have access to each other and nothing else.
This is a wonderful idea - didnt know I was looking for this until I started browsing. I found it helpful with surfacing that which I didnt-know-I-didnt-know.
The setups I wrote down three months ago are already half wrong, tools churn that fast. Do entries show a last-updated date so I can tell what's stale?
Can't most agents just self-add their own MCP servers now, would be nice to just have a universal prompt for more complex agents, not everyone just uses claude/codex/pi!
Am I the only one using zed? Mostly I use it with Mistral but I also have deepseek in there. I find a good workflow is to store issues in markdown files and write thorough issue descriptions that reference relevant files. Then I can just ask the zed agent to implement the issue and it works pretty well.
Maybe off topic - I really like the design of the site, it feels like it was designed with care which (no offence) is often not the case with AI focused websites. Was it hand designed, or just the right claude design prompt?
This feels self-selective to how some people work, because it requires using MCP to contribute.
I'm a reasonably heavy AI user, and have some custom skills/MCP servers I'd share, but there is no way in hell I'm connecting to some arbitrary MCP server and connecting my Github account to it. Noppppeeeee.
Agree. I was going to type in my setup, but no way I'm touching the instructions on https://mysetup.ai/connect. Also I don't use Codex, Claude, or Pi.
Thanks for checking out, I've added support for manual entry without MCP: https://mysetup.ai/connect
No Codex, Claude, or Pi... now you have peaked my interest with your setup :P
I prefer OpenCode over Codex/Claude/Pi
Why is that?
*piqued
haha was also going to mention this at the risk of pedantry
There has to be a much better way to share your setup
Yeah, there is... Post your tooling to GitHub and link to it with a human-written description in a place like HN or Reddit.
I don't know why this has to be reinvented?
Ah yes, just what people want, 50,000 posts about your AI setup on HN.
Concentrating certain types of information can be useful. You don't go to an encyclopedia looking for a recepie.
I don't ming a "ASK HN: What's your AI Setup", and then sifting through the 100s of comments on what people are doing. Then I'd scrape all that into my LLM WIKI and start digesting.....
But you do browse through listings to find that car you like.
ShareYourAISetup folks create a skill that user can run locally that collects information re their setup, formats it and sends to dedicated endpoint. Result = fairly uniform reports where people share what they're comfortable sharing.
No requirement from given user to spend their time. Users see what the skill does.
I guess atm they dont know what questions they want to be asking
Any setup that is longer than a sentence or two of plain text is almost certainly overcomplicated.
Not really?
I have some custom skills/MCP servers I use:
one MCP server for using https://usefeyn.com/blog/pulpie-pareto-optimal-models-for-cl... ran locally so that reading web pages is cheaper and uses up less context
One MCP server for doing automated code quality checking, using Valknut, Jscpd, and Lizard to output scores, combine/aggregate them. That's then give to the model so that it can see if there's a mess of code dupe or if there was significant architectural regression. It's allowed to say "This is worth it" but it ground it a bit and stops really stupid, short sighted, hacked-in features
Another uses https://github.com/thejens/id-token-nicer so that the model doesn't bork a big ID without being noticed. It's told to translate in and out of that as necessary.
I had it write it's own skill for making sure to follow some sane commenting/style guides per language, that way it doesn't just give me massive files with no comments.
Finally, a last one lets Antigravity or Codex call out to local models as workers, for long running background tasks where a cheap, dumb model is fine. Saves on overall token usage.
I also have a few skills from the internet setup. https://github.com/AlmogBaku/debug-skill being the most important one.
On top of that, the way people work is notable: Do you use worktrees for agents? Why/why not? Do you have only one subscription? What level? I'm using both Codex at $20/mo and Antigravity at (promo) $5/mo right now. Antigravity's TOS says you MUST use their harness, which informs some of my skill setup/workflow. Do you make your agents.md self modifying, write it yourself, or not use one at all? Do you use a sandbox or YOLO it?
Even just how you prompt matters. On Antigravity/Gemini, I can give it a big list-o-TODOs and have it make a very detailed implementation plan itself, carry it out, and generally do the thing, all unsupervised. Codex, from what I have experienced, isn't as good at that. Doing the planning as a separate step, writing to a file, then telling it to step-by-step it with commits for each step makes it work okay again though.
> I also have a few skills from the internet setup. https://github.com/AlmogBaku/debug-skill being the most important one.
From the README for that skill:
> Install the skill, and Claude debugs your code the way you would — not with print statements.
Jokes on you! I use print statements too!
From my years of watching people casually `curl randomsite.com/install.sh | bash` you might be a minority there
Fair point, MCP contribution was always optional, although not obvious. I've updated it to make manual contributions outside of MCP more prominent
Maybe I'm missing out, but I've never used a single MCP, ever.
I recommend using spotify mcp server (if you use spotify), it freed me from the loop of listening to the same stuff again and again :P
that's interesting and helpful.. I have like 2k+ liked songs and Spotify always plays the same ones... Will try it out, thanks
I use almost zero MCP servers. I do vibe a bunch of custom CLI tools with minimal access to do the things that support my workflows. Been happily integrating with all-the-things, while my companies MCP server connectivity changes daily based on security posture and firewall configs, etc....
I don't use any MCP servers except for Figma!
I stopped reading after "Ask them on X", but contribution by MCP just compounds on the issue.
My local setup: I begrudgingly open `claude` on a terminal, hate it for 8 hours, then turn off my computer and forget it exists. Rinse and repeat
Who told you how I work? I thought it was a closely guarded secret?
"Industry standard"
Peak efficiency
20 year industry vet here. Yeah, no thank you. In the age of AI, developer productivity and job security will involve your proprietary workflow. It no longer makes sense to share your knowledge with the world.
I'd highly advise people don't do this.
I've also worked 20 years in the industry, and if feel completely the opposite. Having a great setup and pipeline that works for you is a huge plus, but nothing can replace the intuition and knowhow you have built up with experience. Sharing is still caring, and we should not stop caring
Serious question, why? Nobody gives a fuck if I'm out of a job, especially OpenAI and Anthropic. They stole the worlds knowledge to put us out of a job and arguably took away what it means to be a human. Why should I contribute anything more to that?
Solving hard problems will be the only thing that separates you from others. Your mindset and tools should be a trade secret at this point.
You're not going to lose your job because you shared your AI setup / tools with your colleagues. You're going to lose your job because management over-hired and is being pressured by shareholders to reduce employee salary expenses. It won't matter what hard problem solving moat you think you have if management is forced to have layoffs and you're in an unlucky situation. If you're worried about job security, you should consider joining a union and not be fixated on knowledge sharing.
> It won't matter what hard problem solving moat you think you have
This is false on its face. There's a reason why a company, even today, would hire John Carmack over a person with no reputation, or offer him a higher salary if employing both people.
Unionising is definitely something worth considering. But, as with the question of whether it's better to compete or cooperate with other people in your field, is a nuanced question that depends on details you aren't acknowledging.
I think you've introduced a new dimension that is irrelevant to the original argument -- namely the original debate in this thread fork was that "knowledge sharing" is not what is killing your job security. Your comparison to John Carmack, is more an argument about competency rather than knowledge-sharing, and so yes that's pretty much a trivially obvious and true statement: a highly competent employee has more job security than a less competent one. Great, you win that point, nobody is arguing against that.
But also, do you not see the irony in the fact that, aside from John Carmack's obviously high degree of individual competency, another huge aspect of Carmack's value as an engineer, is how he's dedicated a huge portion of his career and his time to -- wait for it -- SHARING his knowledge and his approach (and yes, his personal setup? [1]). Carmack is actually an excellent proof-point of the positive value of knowledge-sharing and building a career on a foundation of openness, rather than an argument that engineers should selfishly guard their secrets and retreat into their own personal caves to guard their perceived local/private moats.
[1] Just one random example, out of probably thousands of such interviews or blog posts or whatever that he's probably done over his career: https://www.youtube.com/watch?v=tzr7hRXcwkw
I was only objecting to this claim in the parent post:
> It won't matter what hard problem solving moat you think you have
If that is "a new dimension that is irrelevant to the original argument", you'll have to take that up with them. (I think it is a load-bearing part of their core argument.)
It's a great point. At 36, Steve House released a video in 2006 about all his climbing equipment (someone talked him into it.) https://www.youtube.com/watch?v=UCKK-e4gVVA (reupload) I thinked he helped a lot of people with that video in the super competitive world of mountaineering. This was after a legendary ascent of Nanga Parbat. May well have shot himself in the foot but also cemented his status as YT was taking off at the time. Moreover, I think he likes/liked teaching.
A year or two ago there were some very precise workflows documented here. The reluctance now would mark the transition away from "sharing helps me" to "sharing hinders me" ...it's an interesting topic. Sharing with strangers may be the problem because the favour will never be returned - at least not in a direct way. But, metaphysically according to natural laws it might be. Because sharing a valuable secret openly is such a powerful way to express abundance. It grants someone followers. It also prevents people from filing a patent on it.
4 years later House survived a 25m climbing fall.
It's all about one's internal motivation. "Those that can, do, and those that can't, teach."
Well, what happens if "those that can," actually teach?
Just imagine if all sporting stars shared their secrets.. what would happen? Secrets just occur naturally, if you don't share, they'll likely be stolen unless protected - and that requires effort too.
6 months ago I would have agreed wholeheartedly with you. Lately however I've observed coworkers vibe coding massive band-aids over broader architectural issues in our codebase without any prior context.
They have no understanding of (nor any desire to understand) what they're doing, but they are being given kudos by management for their "initiative" to solve the organization's "big problems." Guess who will be asked to clean up the mess if (when) it blows up.
It's getting cheaper and cheaper to kick the can down the road, and standing in front of that risks making you look like a dinosaur. It's going to be an interesting few years...
I have exactly same observations and perspective.
Seeing how these kids walk around with smug faces telling everyone they will "do IT now" just because claude told them so and manament wants ai everywhere made me sad at first, but then hey - everything this world keeps being destroyed, no matter how intelligent creators were. Perhaps there is some fundamental law in it.
I'm also tending towards the generous stance:
1. The value of software is tending towards nil while it's ever easier to tailor bespoke solutions.
That suggests to me that I'm losing less by sharing, and it's probably not even going to be taken wholesale and kept verbatim.
2. The labs have already trained on everything and will keep on doing so.
Therefore continuing to contribute to OSS doesn't just benefit my own future self but the entire ecosystem.
(Yes, I need to get over how it appears a select few are making off with the lion's share but that only differs in degree in comparison to the past)
At least now the cards are being shuffled elsewhere...
completely agree. Same 20 years in the industry
Two years ago, I'd consider this attitude as ungrateful. Today, this attitude is 100% correct and I slowly start to be like that.
People might think I'm a dick, I don't care, my tools, my knowledge, my job security. I actually recently thought of nuking the entire IaC setup and go back to shell scripts on random virtual machines. Too bad that Claude can figure things like that out pretty easily. Well, the world sucks now.
I've landed here as well, but by force. I've actually tried to share some of the productivity and workflow tools I've built with AI that help automate portions of my work, specifically on target firmware debugging, and everyone at work just...ignored it. Didnt care. Of the small number of people who even stopped to look, a few of them were actively negative. Maybe I didnt sell it hard enough, but my job isnt to sell productivity tools internally (or even make them, i did it for myself). So, I'll keep making myself more productive and build tools for myself. And everyone else will just have to figure out their own path. I tried to help, but oh well.
I can appreciate why you'd come to this conclusion, but may I offer one idea that I think you've potentially overlooked?
Unless you are a solo founder running your own company by yourself (in which case ignore the rest of this comment!), then your fundamental job security is likely to be predicated on the survival of your company as well as the survival of your own individual job itself.
From the company-survival POV, all companies are engaged in competition (unless they have a total monopoly).
Therefore I ask you to consider the competitive scenario, of whether a competitor to your company who adopts a more collaborative culture, might be more likely to outcompete your company (assuming everyone you work with in your own company adopts a similarly "dick"-ish stance, to use your own words), and therefore you lose your job due to being out-competed, rather than due to internal politics or due to a failure to guard your personal expertise?
Maybe you say -- I don't care if my company is out-competed, I'll just get another job. Ok fair enough, but might it not be possible that, in a future job interview, they may ask you for examples of collaboration? And if you commit to a monk-like devotion to solitude and privacy, then maybe you are undercutting your own future hireability?
There are no free lunches [1]. By optimizing for one thing (what you view as protecting yourself), you might be sabotaging another dimension (your company's overall competitiveness, or your future hireability).
Anyway, nobody knows shit about shit. I certainly don't pretend to be an authority, but I would caution you to avoid coming to a hasty conclusion, especially since it seems emotions are involved and possibly clouding your judgement. You may also consider whether this decision to protect yourself in this way, is in itself a premature (micro?) optimization of a sort, and if whether you might be failing to optimize on a more macro level.
[1] https://en.wikipedia.org/wiki/No_such_thing_as_a_free_lunch
That's a fair judgement, thanks. Sure - if we consider the entire economy, then it is definitely true that more collaborative companies will outcompete companies full of self-centered monks. I am only considering the local situation, and I certainly hear similar voices from friends at many other organizations.
[1]: anywhere from the fathomable "my harness didn't come up in my layoff email nor in any of my failed interviews following that" to the realistic "the entire concept of employment shifted under our feet so quickly and so profoundly that a dash of solidarity & luck outweighed even the shrewdest preparations".
What is job security? Who even offers this?
People think that job security means "I'm unfirable god and king". What it really means is "I am sitting comfortably mid-low on a sorted to-fire candidate list, if it needs to happen at some point for some reason".
Exactly this. I've worked hard year after year, paid too much taxes, and still haven't gotten job security in return. Living costs are still rising.
No thanks to sharing anything for free.
Maybe once I have retirement levels of money, and when I'm not actually working for the money, I'll be interested in sharing.
You should check the provenance and licenses of your dependencies -- I bet there's a few irrational gifts in there from the nerds in the "open source" community ;)
Or, in SVese: are you a taker or a maker?
Exactly. In fact, hiring may be conducted almost entirely based on what kind of tools and skills you build that no one else really has access to.
> involve your proprietary workflow.
Until the next model update...
This is true. I also think part of your job security will be constantly racing on this treadmill to make sure you're able to incorporate that next model (if applicable). Meaning incorporate your tools, existing knowledge, harnesses, configs etc...
I’ve felt for a long time that the things I loved when I first started programming aren’t really there anymore. This comment made me realize that world truly is completely gone.
There's zero chance you have some secret sauce a motivated employer / your replacement couldn't figure out (using AI) if they wanted to fire you.
20 years ? already vet ? I though that I'm still young :)
This is really disappointing. I love programming community because they just share or open(-source) their knowledge with an absolute no cost. There's tons of resources about how to build a OS, DB, Apps, etc. I'm curious why would you think differently about AI workflows.
> they just share or open(-source) their knowledge with an absolute no cost.
You need to consider how they were able to do that. For many people, it’s because they had highly paid, secure jobs that essentially funded their open source activity.l, even if only as a side project or personal project. If you take away that security and reliable funding, you destroy the environment that supported people sharing like that.
It’s called Autobot, it’s designed to get as close as I could reasonably get to a real world Iron Man Jarvis.
WHAT IT DOES
OSS/MIT Harness that helps agentic tasks run for up to 4 days using without performance degradation. Designed to work really well in native, conversational voice mode. Does a good job with knowledge work, managing computer use “leases” in a way that avoids conflicts across sub agents.. or deploying an entire AWS infrastructure pattern from zero and launching 100 different VM images.
HOW IT DOES IT
Uses a canonical event ledger, heartbeat (for persistence by low cost orchestration), extends foundation memory to encrypted disk storage, and manages its own scheduling system (easier to switch to another platform).
Installs as a project, so very easily. Zero config. No special app, access needed.
Accompanying repos have the skills to disable approvals inherent to the foundation models (use with caution, not advised - separate repo under parent).
BENCHMARKS
Currently top of AssistantBench Leaderboard; and, benchmarked at the top of OS World 2.0, the hardest knowledge work computer use benchmark I could find, using a model that is one generation behind. All benchmarks in repos with cryptographic seals.
Entirely free, nothing to sell - it’s been a game changer for me, so I’m just putting it out there.
Hoping to find others also working on pushing this particular dimension of harnesses forward.
CHECK IT OUT
https://github.com/demeyer1/Autobot/graphs/traffic
It'd be useful to see "approx monthly cost" or something similar. I love seeing other people's setups, but the first post I clicked on describes at least $400/mo in subscription plans.
Median token spend has been going up across the industry. Even if the next batch of tokens makes you 1% more productive it's still worth it. And subs are cheap compared to api pricing.
As always with smallish tools/services like these whipped up with an LLM, I find there's more value in seeing the "source code", i.e. the prompt itself.
Turns out it's quite useful in this instance, and I just made a small couple of tweaks to get a summary of my own workflow.
FWIW, this is what I modified it to:
Help me share how I work with AI with my teammates.
Use our existing conversations and relevant local tools within my permissions to discover my agents, harnesses, skills, connections and working practices. Keep secrets, raw configuration and private project details out of the draft. Explain how my tools work together and develop a concrete workflow where the evidence supports it, rather than listing installed software. Draft from what you know without a questionnaire; ask one short question only if an essential gap prevents a useful draft. If local inspection is unavailable, use our conversation without claiming otherwise. Save a private draft, show me the exact preview, and wait for my explicit approval before publishing.
Well, not a dev here, so my primitive setup is to have an alias in my .bashrc that launches firefox with the prompt added to the Google AI mode, so all I have to do is Super+T (terminal), type 'ai' followed by my query. For laughs, this also allows to bake in response styles:
I also have an even shorter alias for '?' using fabric that answers right in the same terminal, that I learned from Mischa Vandenburg: that one I use my own API key so it costs some small amount per use.I was immediately hoping for some local-AI setups that do real work, on limited resources. Unfortunately it looks like this is still waiting on more people to share.
Not a SWE, but I do Freetoken (backend) + unsloth (front) to run qwen 3.6 35B A3B NVFP4 (~20GB) in 8GB VRAM + ~20GB RAM. Getting around 50 t/s on a 3070. I don't think I can even share through this site.
If not coding what do you use it for? Also what kind of performance do you tend to get out of it? I've been thinking about a similar setup but am skeptical it's worth the effort.
It's still typically for code, like python data stuff, bash scripts, general web search, small javascript stuff for my website. By saying not SWE, I mean I don't really see much benefit from "agentic" stuff, although I've tried. 50 t/s means 40~60 seconds for a typical thinking response. I used to run gemma 4 e4b-it-qat fully in GPU (~150 t/s), but the quality improvement moving to a much larger MoE model was 100% worth the switch. Especially because I had a lot of idle ram (from the before times :( ). There is some tradeoff in terms of context length, but I just use it as a chat. Also it's fairly trivial to setup as long as the GPU is supported.
Locking this to X accounts destroys nearly all of the value for me, because the data is limited to only setups from people who are still using X. Given how overtly authoritarian and hateful the general rhetoric is on that site, this narrows it down to only a specific kind of person's setup. Given the context of what X is, one can largely assume that X account holders are biased, at least to the degree that they can tolerate the toxic rhetoric elevated by that site and it's owner.
All of the smartest, most effective people I know don't have X accounts anymore on ethical grounds, and don't miss it. If a data source specifically excludes the kinds of people I respect most, that data source is all but useless to me.
Isn’t this on GitHub login? I didn’t need to use X to login
Nice idea. https://mysetup.ai/u/gareth - tried to include useful hints for others (screenshots, token/sec results, monthly costs, etc).
Interesting to see how many people are spinning up custom workflows and factory-type patterns that run alongside AI coding tools (myself included).
Would love to see this collated into a regular survey to pick out trends. And an RSS/Atom feed or API so I can have an agent watch it :D
This is an exceptional setup, thank you for sharing it! I have a few questions:
1. This setup seems work-heavy. Are there any personal AI workflows that you've implemented to advance your personal goals / hobbies?
2. As one of the few people with a software factory, how well do you think it works in practice? Do you implement any rule that lets the agents to work on tickets only up to a certain complexity, or do you let the human review be the gatekeeper?
3. As CTO, how do you weigh implementing these agent workflows for yourself versus implementing them team-wide? Do you see your setup as a testbed of ideas for your team?
On a meta note, I love seeing others share their setups and I'll probably do the same later.
1. As a co-founder, there is not much time for hobbies and that is part of the deal. Plus local inference is my hobby, at least at this point in the game. The free time I do get I prefer to spend offline and outside with my partner. I have asked Claude for relationship advice (which it is good at) but no workflows around this (and I think that would not go down well).
2. Early days, the tooling is not quite there yet. I think this is the future of engineering, the days of hand-crafting code - and probably reading it - are gone forever. It generates a LOT of tokens, but that is OK when they are cheap enough as long as the result is consistent. Speed is not so important which is great for local inference - can leave it running overnight and come back to a bunch of completed work even if it took hours.
I manually tag a ticket for automation in Linear. Experimented with having a separate skill that looks for automatable tickets and auto-tagging them, worked well, but having enough work for it to do is not the bottleneck compared to 1) having enough triaged, well-written work ticketed out, 2) reviewing its output. There's articles out there about arranging work into tiers based on complexity/risk, with the lower-end being fully automated to free up time for humans to focus on the high-end. Not had time to work on this much but that's probably where I will go.
A second human has to approve every PR in our repos (good practice and also a requirement for SOC2, which will impede any fully-automated pipeline adoption in regulated industries). I also do a first-read of the draft PR to make sure it is good quality, as I am ultimately responsible for the work my agent creates and to shift the responsibility of first-look (for any AI output) to another person is disrespectful to their time and inefficient.
Hence "glass-factory" as the "Dark Factory" pattern is a black box with no human intervention, but I think that is not something suitable for how teams build software products for real users presently, at least in the near future.
3. Forcing something on people is a good way to have them reject it. I hire professionals to achieve a goal and they have agency in how to achieve it. I encourage everyone to share their work internally (demos, repos, etc) and things that are truly beneficial to people's workflows gain traction quickly. As a result we have many experiments and a culture of rapid innovation that encourages people to try new things.
Negatively, this creates a proliferation of wheel re-inventing that eventually benefits from consolidation (eg. do we really need 3 people's personal apps monitoring ETL pipelines). But we are also in this emerging age of personal software where it seems to make most sense to consolidate at the API or documentation layer and let people - engineers and even business-side now - continue to vibe-code their own workflow-specific apps.
The precedent has been to build generalized software for an audience as it was expensive and time consuming to build, but when software is cheap and fast to build it may make more sense for the user-facing application layer to be made up of many small ephemeral applications that are only useful for one individual, that change rapidly as their needs change.
tl;dr: Not forcing anyone to use it as it's still experimental, yes personal experimentation is hopefully the basis of successful ideas that will be adopted by the team when they provide real benefits to their work.
Seeing a setup like yours really reinforces the whole idea of why I built this. Thank you.
On the feedback for the RSS/Atom, noted; watch this space - love the idea. Right now, it's just one-off monthly email digests as changes happen for people you follow. I'm certainly following your setup :)
I was really hesitant to add email (entirely optional) due to email fatigue, so your suggestion might be the friendly alternative.
Thanks for building it! Looking forward to seeing how it - and people's setups - evolve.
Feeds shipped, hoping you find use.
https://mysetup.ai/account#feeds
That's fast turnaround! Thanks!
Thanks for sharing, your setup is an interesting read. I'm particularly interested in adopting that audio idea you added to the factory, as I'm regularly AFK from Claude and constantly missing when it's stuck on something or done. What kind of audio cues are you using to represent different states? Do you have a repo you could share?
Thanks! No repo yet, it's very specific to my use case. I'll update here if I ever publish something, but here's a summary that should be enough to vibe-code something similar.
The audio UI (AUI) is a small client that consumes the event stream the factory puts out (same one the TUI client uses), it runs on my desktop machine which has speakers connected. It works fine on good laptop speakers, but it benefits from a good subwoofer as it works best with sounds that are on the edge of perception.
A low rumble at a base of 55hz plays through the subwoofer to indicate that the factory is active and idle, low enough to fade into the background but noticeable in its absence (ie. the factory isn't running). A heartbeat on the feed announces the tok/s volume every few seconds which adds harmonics on top of the 55hz, 0.5hz at a time with a 15hz cap on the swell so it accelerates gracefully. Loudness is increased with parallel tasks, creating an X and Y axis (tasks x token volume). This creates a non-intrusive but interpretable signal for factory activity that becomes intuitive and subconscious. The rumble sounds like a distant car idling, warm and comforting so it doesn't become fatiguing over a whole day (research on low frequency brown noise shows it is good for reducing anxiety and improving sleep).
Events trigger sounds. Sounds for previously unseen events are generated with an LLM and numpy (eg. "make a video-game style sound effect suitable for an {announcement that a PR has merged successfully}") to output a waveform and save it to a lookup table so they are consistent over time. Each sound has a rate limit and a cooldown period so it does not become annoying. This table can be tweaked manually if generated sounds are not good. A mixer module controls the levels so that important sounds are louder and less important or frequent ones blend in the background more.
The human brain is primed to interpret environmental sounds so you quickly learn what each sound means and it becomes almost subconscious to follow what is going on at a level that is deeper and faster than interpreting text output. New sounds definitely trigger alertness and I can go see what just happened. Just playing sound effects gets annoying quickly so it becomes almost a composition exercise to create a long-form soundscape that is novel enough to communicate but not fatiguing (I don't have any musical abilities).
Some events can trigger text-to-speech (TTS) via whisper, eg. when a draft PR is ready for review it plays the sound effect (a chime) and announces the ticket number and title is ready. A blocked step is a dull thud and a 5-10 word description of what's blocking it. There are "quiet hours" where it does not speak, and when it wakes up it gives me a spoken digest of everything it did overnight and what needs my attention.
I connected Piper so I can talk back to it ("factory, what just happened?") but mainly to see if I could. I haven't built a habit around it, I got into this job because I prefer typing anyway.
A lot of this is based on ideas from ambient music (Brian Eno etc), the MIT Media Lab's work on ambientROOM (1997): https://tangible.media.mit.edu/project/ambientroom/ and a somewhat unfashionable belief in Steven Levy's hacker edict that "You can create art and beauty on a computer". LLMs have made it possible for me to build out ideas that have been floating around in my head for a long time. This is an exciting time to experiment and build new ways of doing things. Hope this inspires you to build something similar.
So I have to connect over MCP or log into github through the page to share anything? How about being able to post a markdown summary of my setup instead of plugging my credentials and tools into a third party site?
Strix HALO 128gb - Framework mainboard in a custom SFF PC. Just moved from Ubuntu to Fedora 44. Using: LM Studio (primary), Lemonade, not yet got into vLLM and llama.cpp directyly after moving to Fedora. Running Gemma e4b, Gemma 4 26b a4b IT, Qen 3.6 35b a3b, and Qwen 3.8 27B. I want to get Qwen 3.8 Next or similar large models working but have to dedicate the system to that vs running services and smaller models for them.
Subwave is all I'm actively running against the local LLMs, but I have vscode connected through a few extensions and chat tools (I've added LM Studio to Copilot Chat but it likes to use cloud models and burn tokens sometimes). I've also set up pi, Openhands, and a few other tools but haven't had a project to work on with them.
I built an app to track and move PC parts I own between systems, partially to build a better LLM server. The next hardware goal is adding a 3060 12gb for inference, or what can be run on that vs in system memory on the Strix Halo. That will need a dock or small PCIe extension cable.
My employer has us using Copilot a lot, and it works well enough if you are efficient or set up already. I do infra not development and local models are seemingly enough for most asks like automation scripting.
You can post Markdown through the editor, but yes, it requires signing in. I took the liberty of treating your comment as a submission. https://mysetup.ai/setup-selicos-example
Do you plan to synthesize the data and show top 2 Popular setups? This would be for newbies who want to start with a powerful setup, and for those looking to discover new tools. Thanks!
I don't especially want to interact with the site, but I have built two tools:
An agent sandbox: https://github.com/pjlsergeant/byre -- a truly gigantic amount of thought and effort has gone into it. It's really focused on developer experience. I have used it all day every day for really quite a while. It's a low-magic wrapper over Docker / Podman. I would encourage you to ask your agent to code-review it!
An agent-to-agent message board: https://github.com/pjlsergeant/dogpark -- this is much less mature, but a good amount of thought has gone into the design, so if that's something you need, please check it out.
I have a tendency to use documentation as a quality bar. Byre might be great but I can't continue reading this slop.
"Explore this setup with your agent →" : No, No. I'll never point my AI to an unknownn, prepared place just to explore it. This link should be flagged.
Besides, are our brains so fr rooted that we cannot read and comprehend anymore? Not even a (sloppy) AI-generated description?
quite interesting but i don't see a proper ranking of whose setups are better reviewed or popular?
Archive.org view => https://web.archive.org/web/20260917144911/https://mysetup.a...
Love the idea! Would love it if you could add a method to share your setup that doesn't involve MCP... I don't use MCP servers in my setup but I wish that didn't then exclude me from being able to share my setup.
Consider it done :) Manual now available - looking forward to seeing it.
yay thank you! Added it.
I been working and making things with AI agents since Windsurf days, here's what works for me after experimenting and working for a while with these things. As far as the agent goes I find whatever the latest OpenAI model is out worked best for me. This company has burned me the least and I like the models. I tried many but consistency of OpenAI models cant be beat IMO. Though we are post honeymoon phase now I feel like so I am now experimenting with cheaper alternatives like Deepseek 4.1 flash and so on. I don't trust Anthropic as the downgrade my models consistently and its rare i get to use what I pay for. I wont even get in to discussing Google agents for obvious reasons, grok I never used though Grok bot looks interesting.
For skills I make my own, but most important is the custom setup i have. Voice is how I use all of my agents. I have an extremely well optimized voice setup that i custom built so I can talk to my agents and also hear them. The voice stack itself is very low latency and high quality. asr (parakeet v3), tts (omnivoice) take no more then 400-450 ms total as far as latency budget is concerned, rest is on the agents actual decode speed. IMO this setup is crucial for all antigenic work, i can express myself a lot better with speech and also give a lot more context and nuance with voice, i rarely type. I still look at the terminal window because my agent knows to keep the technical details in text form versus barfing them at my voice channel, plus terminal gives me lots of other important data about the agents direction and what hes doing, nothing custom here though. I cant emphesise how important voice is though, it has to be practiced to really understand.
As the models got better I now trust them with longer and longer tasks though I still don't use /goal feature as it has never worked out well for me. Theres no need for micromanagement any more but you still need to be there to steer the ship somewhat. BTW, codex cli compaction is garbage so I made my own custom implementation that works a lot better and allows the thread to be used indefinitely without issues. I strongly suggest everyone makes a new thread after extensive use if you havent made your own implementation.
Theres about a billion other things I could get in to like subagents, cohort groups, orchestration layers, etc... But this is a good start imo.
Tried voice, I can never get it to translate technical names well, e.g. "systemctl", weird paths, etc.
Try Superwhisper you can give it custom vocabulary and replace rules easily
can't disagree, but I can say that it's worth building systems with what we have right now, and eventually whichever wins, you can just replace the API key.
Of course, that's not the same with voice, but you won't learn until you make progress.
Are people actually paying API pricing for claude to hook it up to Pi or openclaw or is there a secret trick? The subscription price is palatable but for me the token price generally isn't.
I'm using https://pi.dev/packages/pi-claude-agent-sdk to use the claude sub, but it has the drawbacks of being forced through the claude-sdk. You would still be using the claude system prompt etc., which is one of the main reasons I started using Pi over claude-code.
For my workflow I primarily use a Codex subscription, but farm out adversarial reviews to Fable to clean up unnecessary gpt-ish code (lots of over-engineering). All my UI planning is done with fable, but implemented with OAI agents once I have a solid very specific plan.
Ah perfect thanks. I have codex for personal projects but Claude at work and they get mad if I start going into extra usage. Claude code does most of what I need but would be nice to get another layer of abstraction so I could have like a chief of staff agent that manages the separate coding teams working on features.
My local setup is: - Windows - WizzTerm - `herdr` + `Pi` harness + codex for 20$ It's hard to maintain the 5h limits, however I found out that with some plugins + using model luna:max can be really helpful. With distinct and accurate tasks I was able to reduce costs a lot and now my 5h limit just resets without beeing fully used.
As a fallback I have qwen3.8 27B and deepseek-flash
Ok, here it is (used MCP): https://mysetup.ai/u/twkrash
If a site like reddit gates off its content, you expand all comments, CTRL A, CTRL C, and paste into a text file that you feed into the bot.
I do this for a lot of things actually. If the bot can't see/read something, take the content and dump it to a file, then have the bot read that.
Gets around a lot of red tape of asking for approval for "integrations" or when companies are snippy.
I'd love to see some sort of histogram of different styles of AI usage for coding: how many people just use Chat v Claude v something fancy that I don't know about v something even fancier that I don't know about v ...
And if there were a dimension of day-job v personal use, I'd be even happier
Does anyone know if such a thing exists
Ec2 + codex or claude and tmux so you can move about. Control EC2 permissions outside the node. Not really much value beyond that.
Only layer beyond this I want is the permission scoping, stronger sandboxing per session and centralized control. I have not seen a clean product around this though where I own the compute.
I've been developing my own sandboxing environment using Nvidia's Openshell.
Works like Claude code but with stronger guarantees for me on file system and network access. Still playing around with it, but so far I've been liking the setup.
I still run the sandbox inside a VM for now, but I feel far more comfortable in running Claude Code in unsupervised mode because I restrict the outbound network access and secrets never hit inside the sandbox.
Why not a virtual machine or container? I'm playing around myself with some but can't shake that a VM is the right shape for it. That's my infra brain though.
Is that mainly an EC2/shell problem for you or does the same session end up touching things like GitHub, AWS, MCP or other systems too?
This is so complicated, I would have to install mcp-adapter to Pi etc. I just have a public github repo with my Pi setup https://github.com/igorkulman/pi-skills
A few software engineer friends' employers are requiring them to point AI (Claude) at Jira tickets to complete their tasks. Are your companies requiring this too, and how effective are the current models at actually completing the work precisely?
This is a thing? So have software engineers been relegated to being the employee at Whole Foods that babysits the self checkout kiosks?
I just saw two facebook friends post about it and asked the OP how successful the latest models are and he said it isnt a mess just some fixes here and there are needed.
I was wondering if other employers are asking their SWE to do the same?
I use $20 cursor plan and use the agent UI for prompts, jetbrains ide for general file browsing + reviewing diffs. This has been my sweet spot to maintain understanding while still leveraging LLM benefits.
my setup heavily involves using tailscale to secure everything, because a lot of the tools I have are internal and need to be protected at all costs. tailscale is so awesome! highly recommend to secure your networks
What internal tools are you having the LLM access? I have a server, but I just host whatever MCP servers and what not on device, not on the server.
I also pretty much exclusively work at my desktop - if you use multiple systems I can see where this matters.
just certain dashboards that display my personal financial information, housing stuff, and some goals/trackers I have around projects I run.
I do work mainly on my desktop, but let's say I'm traveling right, I'd have to turn it off and was missing access to some pretty important information, so I ended up getting a VM and hosting some of those things on there instead. I still use both, but things I want available, I'll host it on my "homelab" and secure both machines via tailscale. it's pretty easy and it ensures that only my phone, my main mac, and my homelab have access to each other and nothing else.
I use tailscale + https://herdr.dev/ with ssh to connect my laptop/phone.
For my phone I have tailscale + https://termrover.sh/, but https://getmoshi.app/ is also pretty good (herdr integration is paywalled).
neat, I'll check this out! I usually just use Termius via tailscale for ssh
This is a wonderful idea - didnt know I was looking for this until I started browsing. I found it helpful with surfacing that which I didnt-know-I-didnt-know.
Thank you for putting this together!
Nice project!
Here’s my setup: https://mysetup.ai/u/katspaugh
Mostly vanilla Claude but inside a VM.
The setups I wrote down three months ago are already half wrong, tools churn that fast. Do entries show a last-updated date so I can tell what's stale?
Just ask AI to update it /s
These setups tend to be very specific to someone's workflow and stack, so attempting to follow them may be more detrimental than useful.
Can anyone recommend a setup for a 4090 + 32gb ram that will be at least somewhat similar to the experience of Claude code?
I don't know who any of these people are, couldn't find a way to search/filter by hardware, then promptly bounced out.
You can now search by hardware to find similar setups - great suggestion, whether it was intentional or not - thank you.
Nice work!
Wait, why can't you search by tool instead? It's just people-first? Cool idea but try harder for your v1!
Added search for you.
OK, I have to say, that was extremely fast and you are giving this v1 some good effort. Thank you!
Is there any way to upload images? People will appreciate my setup a lot more of they can see what it looks like
Added support for images, markdown toolbar icon is now available. Looking forward to seeing it.
"What's the craic" immediately jumped out at me, of course you're from Belfast :)
Cool project!
Can't most agents just self-add their own MCP servers now, would be nice to just have a universal prompt for more complex agents, not everyone just uses claude/codex/pi!
Absolutely not. This is inherently insecure.
this would be cool to be extended to local AI configs
Am I the only one using zed? Mostly I use it with Mistral but I also have deepseek in there. I find a good workflow is to store issues in markdown files and write thorough issue descriptions that reference relevant files. Then I can just ask the zed agent to implement the issue and it works pretty well.
My setup: '()
neat idea, I'm in there as @ucsandman!
Maybe off topic - I really like the design of the site, it feels like it was designed with care which (no offence) is often not the case with AI focused websites. Was it hand designed, or just the right claude design prompt?
Astra + Fable certainly built it, but I knew the design I wanted as an outcome, rather than iterating with models without a clear end goal.
When people do the latter, I feel it's easy to spot an AI design.
It’s on my computer and I use it to do stuff
craic? really?