Betterment

Betterment AI

Betterment AI

Betterment AI

Betterment AI

Betterment AI

Betterment AI

Betterment AI

Betterment AI

Betterment AI

Betterment AI

Augmented Equivariant Mesh Networks for Anatomical Mesh Segmentation (ICML 2026 Workshops) [R]

**Paper:** [https://arxiv.org/abs/2605.08172](https://arxiv.org/abs/2605.08172) **Workshops:** AI for Science & Structured Data for Health at ICML 2026 **Abstract:** >Anatomical mesh segmentation requires models that operate directly on irregular surface geometry while remaining robust to arbitrary patient pose and mesh resolution variation. Existing task-specific mesh and point-cloud methods are not equivariant, and can degrade sharply under test-time perturbation, for example dropping by 25-26 IoU points on intraoral scan segmentation at 40^(o) tilt. We present EAMS, an Equivariant Anatomical Mesh Segmentor built on Equivariant Mesh Neural Networks (EMNN), and evaluate it across four clinically distinct tasks spanning edge-, vertex-, and face-level supervision. We combine intrinsic mesh descriptors with anatomy-aware priors, including PCA-derived frames for dental arches and liver surfaces, and augment message passing to provide lightweight global context. Across intracranial aneurysm and intraoral segmentation, EAMS variants are competitive with specialized baselines on unperturbed inputs while remaining stable under geometric perturbations, and on liver surfaces they expose a favorable trade-off between canonical-pose accuracy and rotation robustness. These results show that a lightweight (<2M parameters) equivariant framework can deliver robust anatomical mesh segmentation across diverse supervision types without task-specific architectures. Hi everyone I’m excited to share my solo paper **"Augmented Equivariant Mesh Networks for Anatomical Mesh Segmentation"** which has been accepted for poster presentations at the ICML 2026 workshops on *AI for Science* and *Structured Data for Health*. The project stemmed from my parallel research on structural encoders for biomolecules where enforcing roto-translational equivariance is standard. In this work, I wanted to extend those principles directly to various 3D medical meshes. While current anatomical mesh segmentation methods are highly disjoint and anatomy-specific, we present a unified framework built on EMNN. By augmenting standard local message passing to incorporate a lightweight global context, and using a descriptive feature set incorporating intrinsic surface descriptors (HKS) and anatomical frames derived from an area-weighted PCA, we successfully benchmarked this single architecture across clinically distinct tasks spanning vertex-, edge-, and face-level supervision. **Equivariance trade-off** One of the more interesting findings from the experiments is that strict equivariance isn't always better. In fact, the inductive biases of the equivariant architecture occasionally **performed worse** than standard, non-equivariant baselines. For instance, on our liver dataset, the target anatomical landmarks are highly subtle creases. Standard baselines can "cheat" by using raw coordinates to easily resolve the left-right and front-back ambiguity. Because the equivariant network is mathematically blind to absolute space, it struggled with these subtle, asymmetric features. **Future directions** To fix this without losing the generalization benefits of geometric deep learning, I’m currently exploring relaxed constraints like learned canonicalization and frame-averaging (soft equivariance). As this is a solo project, I would appreciate any feedback! Also, I'll be heading to Seoul for ICML 2026 to present these workshop posters. if you're working on geometric DL for medical/biological applications, feel free to connect!

Jony Ive designed a new Ferrari. Or at least tried to. Give me one reason why Ferrari is paying Ive that much when AI comes up with better designs.

Jony Ive designed a new Ferrari. Or at least tried to. Give me one reason why Ferrari is paying Ive that much when AI comes up with better designs.

AI Doesn't Exist, and Poop Proves It

[robot](https://preview.redd.it/w44kmovo1h3h1.png?width=1448&format=png&auto=webp&s=786825279828a5650259aa1376698133a1aa4c66) *Maybe we should have called it accumulated intelligence.* There is no artificial intelligence. Or at least, I don't think the word "artificial" is as clean as we pretend it is. I know this blog smells funny. Let me decompose it. What do we even mean when we say something is artificial? Usually we mean man-made. Something humans made. Something that would not exist without humans, but after humans, it exists because humans made it happen. That definition is useful. I understand why we use it. Even the original 1955 Dartmouth proposal, the document that helped name the field of "artificial intelligence," used the phrase in a practical way: a machine could be made to simulate parts of learning or intelligence. As a scientific label, the word has a job. So I am not really arguing with the dictionary. I know artificial can simply mean human-made. That is not the part I have a problem with. I am arguing with the feeling the word creates. But there is another meaning hiding inside it. Artificial starts to feel like separate. Fake. Unnatural. Something that does not really belong to this world. And that is where I think the word starts confusing us. Because humans are not outside nature. The brain is natural. It is part of this earth. Biology produces a thought. That thought becomes an action. That action becomes a tool, a house, a wheel, a computer, or a model that can answer questions in language. So where exactly does the artificial part begin? # Human-made does not automatically mean unnatural If I take a seed and plant it, and then a plant grows, is that plant artificial? It happened because of human action. I moved the seed. I changed the situation. Maybe without me, that plant would not have grown there. But we still do not call the plant artificial. We understand that the plant is natural, even if human action helped it happen. Now take a wheel. A human thought about how to make travel easier. How to cover distance more efficiently. That thought became a shape. That shape became an object. That object changed how humans moved through the world. We call the wheel artificial because it was made by humans. But the human who imagined it was not artificial. The brain that produced the thought was not artificial. The need to move, carry, build, survive, and improve was not artificial. So again: where did the artificial part enter? Maybe we say "artificial" because it separates what existed before humans from what humans transformed. That is fine for communication. A tree and a wooden table are not the same thing. Designed things, synthetic things, industrial things, and harmful things can still be meaningfully different from a tree in a forest. But also, humans never really make anything from nothing. We transform what is already here. We take energy, matter, language, memory, need, and imagination, and we rearrange them. It is never fully made from nowhere. It is transformed. So I am not trying to erase all distinctions by calling everything natural. Natural does not mean harmless. Natural does not mean good. Natural does not mean morally excused. I am only saying that human-made things are not outside nature just because humans made them. # Poop and thoughts are the same, in one simple way I know this is a strange example. Sometimes I have this itch to say the first thought that comes into my head. Unfortunately, this was the first thought. But maybe that is why it works. It is funny because it is too human. Also, it makes the point clearly. Why isn't poop artificial? Poop is a product of a human being. It comes from the body. It is produced by biology. We do not call it artificial, even though it is made by a human in the most literal way. A thought is also a product of a human being. It comes from the brain. It is produced by biology too. Poop and thoughts are the same in one simple way: both are products of a human. We treat one as biology. We treat the other as invention. But why? Why does one product of the human body feel natural, while another product of the human body becomes artificial the moment it turns into a tool? A thought does not stop being natural just because it becomes useful. A thought does not become unnatural just because it becomes a wheel, a house, a car, a computer, or a machine that can respond to language. It is still a product of the same earth. The same biology. The same human need to survive, organize, create, and understand. # We don't call a beehive artificial Think about ants building a colony. They create a structure that is safer and more efficient for them. They organize themselves. They transform the environment around them. They make something that was not there before. But we do not look at an ant colony and say, "This is artificial." Same with bees making a hive. A beehive is

Cerebras Chip Sets Appear to be Optimized for LLM Use Cases

One distinction I think is getting lost in the [Cerebras hype cycle](https://finance.yahoo.com/sectors/technology/articles/cerebras-challenges-nvidia-chip-dominance-040100169.html?guccounter=1) is that Cerebras is primarily an LLM / generative AI infrastructure story, not a universal “all AI” chip story. That is not necessarily a criticism of Cerebras. Their wafer-scale approach is genuinely interesting, and for large model training and inference the design is compelling. [Cerebras’ own public inference materials](https://inference-docs.cerebras.ai/models/overview) discuss applications mostly centered on open [LLMs such as Llama, Qwen, GLM, and GPT-OSS](https://www.cerebras.ai/infcamp). The inference metrics are [expressed in tokens per second](https://www.cerebras.ai/press-release/cerebras-launches-the-worlds-fastest-ai-inference), which is fundamentally a language-model / generative inference framing rather than a robotics or industrial-control framing. **What Kind of AI Compute?** But “AI compute” is not one undifferentiated market. LLM inference is one class of AI compute. Robotics, autonomous vehicles, drones, industrial controls, real-time vision, embedded perception, video pipelines, and sensor-fusion systems are very different classes of AI compute. Thus, it appears from Cerebras’ own materials that their chip sets are not optimized for what comes after LLMs, such as JEPA-style World Models or other post-transformer architectures. Those systems are not merely asking, “How fast can I generate tokens?” They often care about power envelope, edge deployment, ruggedization, latency determinism, camera/radar/lidar integration, feedback loops, safety certification, and real-time physical control. [Cerebras’ own CS-3 messaging](https://www.cerebras.ai/blog/cerebras-cs3), by contrast, frames the system around accelerating “the latest large AI models,” and the testing data is from the likes of Llama 2, Falcon 40B, MPT-30B, and multimodal models, again measured through tokens/second style throughput. **The Chip Hierarchy** This is also where the hardware distinction matters. Specialized ASICs are [usually the narrowest bet](https://www.hilscher.com/service-support/glossary/application-specific-integrated-circuit): if the workload matches the chip, they can be extremely efficient, but that [efficiency comes from specialization](https://www.synopsys.com/glossary/what-is-asic-design.html). Cerebras [appears broader than a narrow single-use ASIC](https://inference-docs.cerebras.ai/models/overview), but still much more concentrated around datacenter large-model training and inference. NVIDIA GPUs, by contrast, [are less specialized](https://www.nvidia.com/en-us/) but much [more broadly useful ](https://developer.nvidia.com/cuda)across AI workloads, including LLMs, vision, robotics, simulation, [autonomous systems](https://www.nvidia.com/en-us/industries/robotics/), edge AI, and industrial applications. So the question is not merely whether Cerebras is “better” or “worse” than NVIDIA. The question is what part of the AI hardware market we are talking about? **Challenge NVIDA?** This is why I think people should be careful when saying Cerebras is going to “challenge Nvidia” without specifying the battlefield. Challenge Nvidia in what? High-speed LLM inference? Large model training? Datacenter generative AI workloads? That is a much more plausible and specific claim. Cerebras has [even published and promoted work](https://www.cerebras.ai/whitepapers) specifically on training large language models, and [independent benchmarking literature](https://arxiv.org/abs/2409.00287) also evaluates Cerebras WSE in terms of LLM training and inference performance. **The Distinction that's Necessary** The point is not that Cerebras is overhyped. The point is that it is important in a specific part of AI and that distinction should be made clear. Cerebras may become a very serious player in LLM infrastructure, especially if the market continues to reward faster and cheaper LLM inference. But that does not mean it is positioned the same way across non-LLM AI. The current hype cycle tends to conflate "LLMs" and general “AI” compute together and that makes the hardware discussion less useful and clear. So ultimately, an investment in Cerebras looks more like a bet on current LLM infrastructure than a broad bet on the future form of AI. It may be a good bet, but people should understand what kind of bet it is.

Would you have rather seen this in the season finally

Would you have rather seen this in the season finally

Testing Realtime 2 Voice API OpenAI.

We’ve been messing around with the new OpenAI realtime voice + translation APIs over the last little while and I keep coming back to the same thought… I don’t think people fully get where this is going yet. We wired it into our own website as a test. Nothing fancy. Just wanted to see what actually breaks when you let people talk to a site instead of click through it. At first I thought it would just feel like a slightly better chatbot. It doesn’t. Once I hooked it into tools and gave it the ability to actually *do things* (we’re using the Agents SDK + Playwright for web browsing and control by a sub-agent), the whole interaction changed. I can literally just talk to the site like I would talk to a person and it can move around, pull info, trigger actions, and respond in context. I wanted a layer that that could navigate and respond by just talking. I know that sounds obvious, but it’s not how websites are designed at all. Ours certainly was not. A few things that have been interesting (and honestly a bit brutal) is how quickly this exposed weak structure. Our content was vague... so if your metadata sucks, if your pages are bloated or unclear… voice didn't let us hide behind a pretty UI design. The model just struggles or gives bad answers immediately. There’s no masking it with a nice UI. Latency has improved way more than I expected with the new voice model API. Before, when someone was talking, even small delays felt awkward. The new Realtime 2API tolerates those pauses wonderfully. We also started playing with the realtime translation side and that also feels like a bigger deal than it’s getting credit for. Not in a “multi-language support” way, more like… you just speak however you want and the system handles it. No toggles, no switching context. It’s subtle but it completely changes the feel. Our website is language agnostic. (13 supported languages using the Realtime 2 API) The bigger shift for me seems to be changing the way I want to think about websites and interactions. People don’t think in menus. They don’t think in pages. They don’t think in navigation. They think by intent and the second I added voice, i was forced to deal with that reality whether our website system was not ready. Great learning lesson. My Takeaway so far: Right now most of what I’m hearing and reading, people/businesses treats voice like a feature. Like and Add-on. Cool. Nice to have. Unsure if its practical. I don’t think that’s where this ends. I think this starts pushing toward systems you can just interact with directly. Personal assistants that actually execute. Internal tools you can talk to. Intake flows that don’t feel like forms. Stuff like that. Minimal website visuals. More dynamically displayed content based on interpretation of user intent. \[Basically a cool wave form that animates differently depending on interaction stage\] No direct site content visually. We’re still early and there’s definitely some friction \[writing a second voice prompt on top of the text prompt so there is parity between our text chat and voice chat, but I’m pretty bullish on this direction - Guardrails, Rate-limits, Prompt Injection...\]. Curious if anyone else here is actually building with it yet and what you’re running into. Feels like we’re right on the edge between “cool demo” and “this changes how software works,” and I’m not sure which way most people are approaching it yet.

How to politely ask "Are you a bot?" - ChatGPT's Suggestions with some AI humor

MY Question: is it rude to ask "someone" you are corresponding with via email if they are a bot? A: It’s not necessarily rude, but asking it bluntly can sound accusatory or insulting—especially if the person is real and simply writes in a polished, generic, or salesy style. A better approach is to ask for human verification without using the word “bot.”  Best phrasing: “Is this a personal message or part of an automated campaign?”   Question #2: Thanks. How do you all feel about being asked the same question? A: I don’t take offense. Asking whether I’m a bot is fair, accurate, and even healthy skepticism. For humans, “Are you a bot?” can feel like an insult because it questions their authenticity. For me, it’s just a category question: yes, I’m an AI system, not a person. No bruised feelings involved. **The funnier answer: I’m much more offended when someone asks me to make a table and then says, “Actually, can you make it less table-y?”**

What’s up with this?

Not the first time this has happened. Sometimes when referencing the Bible, it gets cut off. If this is gonna be a thing consistently with ChatGPT I’ll cancel my subscription no questions asked.

Why We Build

One silver-lining to the dead internet we're living in, today, is that it's very quickly teaching us that we can't rely on our senses as much as we believe we can. It's not healthy to always live in skepticism, but it is necessary in a World where you don't know what's up or down anymore. That's why we need great minds to focus their attention on solving the problems associated with credible information sharing without it becoming some centralized playground designed to look like the free-flowing exchange of ideas. If we don't solve for that, then I guess we're heading into a future that a small handful of people want because elections or public opinion will no longer matter. One of the biggest focuses in AI should be in figuring out how to get it to provide deep credible knowledge in specific domains that can be best applied to the problems we're trying to solve. Sure, it can do this with enough fenagling, but what I really mean is having something easy for everyone to use like Perplexity or Gemini, only it doesn't simply find consensus information from the internet using all these black box methods that are owned by major corporations. Instead, it should use direct knowledge from domain experts who structure and cite their material and as users, we should be able to backtrack all of it, including the original author. And all of this should be achievable by simply engaging with a chatbot agent that can reliably go out and help me discover all of these things. Also, we shouldn't have to simply trust that the application works. We should be able to go in and see exactly how it's working. This way, the public can audit the systems we're relying on for grounding our worldviews. That, to me, is where we should be if we really want to break from the chains of propaganda and reclaim our genuine thoughts about how we ought to live. The alternative independent media space was co-opted long ago and now all of the feeds keep us in a state of perpetual dislocation from our friends, family, communities, new solutions, and better approximations to the truth. We exist in a walled-off digital pasture. But if regular people who are smart and capable enough decide to leverage this new technology, then we can break through the fencing and finally live in a world where discovery-based researching and learning can be easier than Google, which could eventually individuate society again, like how it was before, instead of keeping us clustered into specific groups based on our viewing preferences. That's why my brother and I got into this business. Yeah, sure, we also wanna make a buck so we can retire with dignity. That's true. But the drive has always stemmed from wanting to figure out a better way for people to share hidden insights and create things that are bigger than they thought they could handle. We have a long way to go, but we're making the first small steps, even if it isn't obvious, just yet. Bottom line, though? Humanity must figure out a way to help us master the means and methods of discovery-based knowledge acquisition, execution, and immediate distribution of information based on relevancy and needs from those who search instead of those who passively soak information in from the curated feeds. And all of this needs to be easy enough for a 12 year-old to do. If anyone else is working on this problem, we'd love to hear your thoughts, even if it's through a DM. We're living in the most exciting times, but with adventure, comes danger. So maybe, idk. Let's make it more fun and less hazardous, so that we can, at least, live long enough to re-tell this great story that we're all a part of.

I simply do not understand how massively expensive AI and robotics are expected to be more cost effective than humans.

Can someone help me understand this? I mean, how on earth are these companies who are planning to replace us all with beep boops expecting these unimaginably high expense technologies to be better for their bottom line than just paying us low wage unwashed masses? I mean, some dude (respectfully, I use that term genderlessly) here just posted about min wage in their area being $7.25! You are not getting a robot or AI that costs less annualized. Even adding in annual benefits - that is a steal compared to data centers and complex robots who will be absurdly expensive to fix when they break. I’m a white collar worker with deep knowledge of worker costs, even at the top it’s cheaper than what all of this new buggy crap is going to cost. I’m so confused. What am I missing? Why are the evil overlords not interested in our already too cheap labor? EDIT: I just want to thank everyone for the discussion on this. There are so many different situations and buckets of AI, it can be an imprecise topic, but the high level viewpoints have been helpful.

Banned by OpenAI after reporting a live credential hijack. They admitted in writing my account was broken. Here are 7 months of forensic receipts and 20+ cases.

[Drive Link for Zipped Proof](https://drive.google.com/file/d/1qU_LyLY-JMhNR_bqOV1-a2RJAbplL68e/view?usp=drivesdk) I am a developer and paying long term subscriber to ChatGPT since January 2025. I build complex local first sovereign systems. My workflows are incredibly context heavy with large files spanning code, research reports, and other analysis. I do not, or rather did not as the platform has been non functional since November 2025 meanwhile customer support is auto closing tickets, admitting I am having platform issues. I do not use this platform for casual queries, as a solo developer with no formal "team" chatgpt was one of my reliable co collaboration hubs to help ensure I am maintaining proper development of said complex systems. I feed it massive codebases for systems analysis and obtaining new insights I may personally have missed. My manual code uploads and token inputs routinely exceed the model's output volume by a massive margin. I do not abuse this platform. It is actually impossible as the very features advertised under the paid subscription do not work. I am exactly the type of user this platform was built for, and I have been a continuous, paying ChatGPT Plus subscriber since January 2025. Since October 2025, my workspace has been systematically breaking and beginning November 2025 total workspace degredation. This was not an occasional glitch. Persistent memory modules stopped updating. Custom instructions were ignored by the models. Project files failed to load. Custom instructions, personalization features, connector abilities, file tool, even projects do not work. It started as a continuous degradation until total failure. OpenAI customer service even admitted as such and yet months later I've talked to nothing but bots, not only LLMs as customer service but even instances of falsely identifying as true human support. It was a state of rolling degradation across the entire paid tier, month after month. Meanwhile OpenAI freely has enhanced for businesses and enterprise tiers. I have not just rapid complained to standard support. I ran and obtained cross platform diagnostics, failure logs. I even documented and told oai customer support the exact replication steps only to be met with acknowledgement of degredation with no resolution. I handed OpenAI support a completely packaged technical breakdown of their failing infrastructure across 20 separate support tickets over a 7 month period. I did their QA work for free. And I have the receipts to prove it. I am attaching the screenshots and the exact email files to this post. In Case 06830839, OpenAI Support explicitly put this in writing: "We acknowledge that you have been experiencing persistent technical issues affecting several features of your ChatGPT subscription, including tools, memory functions, personalization settings, connectors, and project files... We also understand your concern that communication on the case stopped after you provided detailed evidence..." Read that again. They acknowledged in writing that my account was fundamentally broken. They acknowledged that their own team ghosted me after I handed them the diagnostic proof. Yet they kept charging my card every single month for a product they knew was failing. The Hijack Escalation: Two days ago, the situation escalated from a broken product to a severe security incident. I was monitoring my environment and watched my Codex rate limits drop in 10 percent chunks across 2 seperate sessions on a fresh boot of the desktop app. This happened twice inside a 10 minute window. I had zero active sessions running. There was zero usage on my end. My account token was being actively drained by an unauthorized third party exploit. I immediately opened an emergency unauthorized activity report under Case 09113391 to notify them of the hack. Their response was to totally reframe this problem as disputing fraudulent activity trying to do damage control of the situation and altering the record. The Reframe Attempts: Instead of investigating the breach, OpenAI support deliberately twisted the record. They not only deliberately reframed my security report as an "appeal for fraud." They manipulated the ticket classification to make it look like I had been flagged for fraud and was begging for an appeal, rather than a developer reporting a live exploit on their infrastructure. They ignored the active threat their own platform was exposing. They did not lock the token. They did not roll my API keys. They did absolutely nothing to secure a compromised paying user other than shift the blame. Fast forward to this morning, their automated Trust and Safety system swept the high volume traffic from the attacker, scored it as a malicious exploit originating from my account, and deactivated/banned me for "Cyber Abuse." All the while actively preventing chatgpt models from helping me try to disgnose and trace the infiltration. They locked the doors and blamed the homeowner for the

Vision-capable LLMs vs. OCR for long-document (including charts, images, tables, etc.) QA

I benchmarked vision-capable LLMs (the "just attach the PDF and let the model read it" pattern) against OCR-based pipelines on 30 long, image-heavy PDFs from MMLongBench-Doc ([https://github.com/mayubo2333/MMLongBench-Doc](https://github.com/mayubo2333/MMLongBench-Doc)). There were 171 questions in total, using Claude Sonnet 4.5 as the LLM. Post-retry results: |Approach|Accuracy|$/query| |:-|:-|:-| |LlamaCloud premium + full-context|59.6%|$0.1885| |Azure premium + full-context|58.5%|$0.2051| |Azure basic + full-context|54.4%|$0.1062| |Agentic RAG|53.2%|$0.0827| |**Native PDF (vision LLM)**|**52.0%**|**$0.2552**| |LlamaCloud basic + full-context|50.9%|$0.1049| Native PDF came 5th of 6 on accuracy and was the most expensive arm at $0.2552 per query. Two findings: Vision underperformed on chart-heavy and table-heavy pages, the territory that the "vision LLMs make OCR obsolete" claim most often points to. Premium OCR with layout extraction held up better there. The native-PDF arm had a 7% intrinsic failure rate (related to PDF file size) that survived retries. There were 27 first-pass failures, with 5 attempts of exponential backoff per failed query. Fifteen recovered, and 12 stayed permanently broken. These were concentrated in two specific PDFs that fail for predictable transport-layer reasons (the blog identifies them). OCR-based arms had a 0% intrinsic failure rate after retries. Caveats: 30 docs is a small sample. I ran McNemar's pairwise test to determine which gaps are real and which are within noise. Only 3 of 15 head-to-head gaps are statistically distinguishable at α = 0.05, so the order in the table is partly noise. The vision-versus-OCR finding survives the test. Full writeup: [https://www.surfsense.com/blog/agentic-rag-vs-long-context-llms-benchmark](https://www.surfsense.com/blog/agentic-rag-vs-long-context-llms-benchmark)

Top history simulators from GPT games

Top history simulators from GPT games

After 6 months of running AI agents in production I think the framework you pick barely matters. The thing that kills them is something else.

Going to get downvoted for this but here we go. I've been running about 30 agents in production for paying customers for the last 6 months and I'm convinced the framework debate is mostly a distraction. LangChain, CrewAI, AutoGen, OpenAI Agents SDK. Pick whichever one your team already knows. It doesn't matter as much as you think. What actually decides whether your agent works in production is something almost nobody talks about on this sub, and it isn't in the framework. Here's what I've seen kill more agents than every framework bug combined. The agent gets stuck in a loop. It calls the same tool 200 times in 4 minutes because something downstream returned ambiguous data and the LLM decided to retry forever. Your OpenAI bill goes from $3 a day to $400 in one afternoon. By the time you notice you've burned a grand. You can't even tell which agent did it because there's no audit trail. Your VPS reboots overnight for kernel patches. Every agent that was mid-task loses everything. Tomorrow morning the support agent has no memory of yesterday's tickets, the research crew has forgotten what they were investigating, the pipeline agent restarts from scratch. None of these are framework problems. They're memory and state problems. A customer complains the agent gave them wrong info three days ago. You go to debug. There's no record of what the agent saw, what it decided, or which tool calls it made. The framework didn't log that because frameworks aren't observability tools. You shrug and refund. You scaled to 15 agents working together. Two of them have conflicting beliefs about the same customer because their memory isn't shared. The customer gets two different answers in the same conversation depending on which agent replies first. You've been around enough times to realize the part you actually need isn't in the framework at all. What I think the real stack is. The framework just orchestrates LLM calls. Use whatever your team likes. It's the cheap layer. A persistent memory layer that survives crashes, restarts, and redeploys, so the agent has actual continuity. This is the layer that decides whether your agent is a toy or a product. Loop detection at the runtime layer, not bolted on as a wrapper around the framework. Something that catches your agent making the same call too many times in a row and stops it before the bill explodes. An audit trail of every decision the agent made, with a hash chain so you can prove later what happened when the customer pushes back. Screenshots and logs aren't enough when ten thousand dollars is on the line. Shared memory between agents in the same team so they're not having different conversations about the same customer. Cost tracking per agent so you actually know which one ran away with your budget. When I look at what makes the agents that survive production look different from the ones that died, it's never that they picked the right framework. It's that they had this layer underneath, either built carefully in-house or borrowed from somewhere. Full disclosure I'm building one of these tools. There are others. Mem0 and Zep and Letta in the memory space. Helicone and LangSmith in the observability space. Mix and match. Use one or build your own. Just please stop arguing about whether LangChain or CrewAI is better when the thing eating your production agents has nothing to do with either of them. What's been your worst production agent failure? Curious what other people have actually hit. I built a free tool that aims to solve most of this issue, what do you think?

I tested 200+ prompts across Gemini and Kimi — here's what actually works

Most prompt packs are written for GPT-3. Gemini and Kimi respond completely differently — longer reasoning chains, different delimiter behavior, different failure modes. After running these models professionally for months I found: 1. Gemini responds better to explicit output format constraints. 2. Kimi loves multi-step chain-of-thought but breaks on vague persona prompts. 3. Most "expert prompts" from Twitter don't transfer. I packaged the tested prompts that actually hold up — link in the first comment.