suninsight

Very cool and nicely executed ! Definitely see a lot of value in this.

I was actually building a version of this using NonBioS.ai, but this is already pretty well done, so will just use this instead.

So I can attest to the fact that all of the things proposed in this article actually works. And you can try it out yourself on any arbitrary code base within few minutes.

This is how: I work for a company called NonBioS.ai - we already implement most of what is mentioned in this article. Actually we implemented this about 6 months back and what we have now is an advanced version of the same flow. Every user in NonBioS gets a full linux VM with root access. You can ask nonbios to pull in your source code and ask it to implement any feature. The context is all managed automatically through a process we call "Strategic Forgetting" which is in someways an advanced version of the logic in this article.

Strategic Forgetting handles the context automatically - think of it like automatic compaction. It evaluates information retention based on several key factors:

1. Relevance Scoring: We assess how directly information contributes to the current objective vs. being tangential noise

2. Temporal Decay: Information gets weighted by recency and frequency of use - rarely accessed context naturally fades

3. Retrievability: If data can be easily reconstructed from system state or documentation, it's a candidate for pruning

4. Source Priority: User-provided context gets higher retention weight than inferred or generated content

The algorithm runs continuously during coding sessions, creating a dynamic "working memory" that stays lean and focused. Think of it like how you naturally filter out background conversations to focus on what matters.

And we have tried it out in very complex code bases and it works pretty well. Once you know how well it works, you will not have a hard time believing that the days of using IDE's to edit code is probably numbered.

Also - you can try it out for yourself very quickly at NonBioS.ai. We have a very generous free tier that will be enough for the biggest code base you can throw at nonbios. However, big feature implementations or larger refactorings might take time longer than what is afforded in the free tier.

how about halarax ...halucinate and paralax

So we tried that route - but problem is that these interfaces aren't suited for asynchronous updates. Like if the agent is working for the next hour or so - how do you communicate that in mediums like these. An Agent, unlike a human, is only invoked when you give it a task.

If you use the interface at nonbios.ai - you will quickly realize that it is hard to reproduce on slack/discord. Even though its still technically 'chat'

I think if you use Cursor, using Claude Code is a huge upgrade. The problem is that Cursor was a huge upgrade from the IDE, so we are still getting used to it.

The company I work for builds a similar tool - NonBioS.ai. It is in someways similar to what the author does above - but packaged as a service. So the nonbios agent has a root VM and can write/build all the software you want. You access/control it through a web chat interface - we take care of all the orchestration behind the scene.

Its also in free Beta right now, and signup takes a minute if you want to give it a shot. You can actually find out quickly if the Claude code/nonbios experience is better than Cursor.

1. Multi-Agent is divide a part into tasks and hand off each part to a different Agent. This is different in the sense that a task is not divided into parts aprior. When the agent gets to a roadblock - lets say it is unable to fix a software issue - it rolls up to a deep think model to unblock. But you might be right that the difference is too subtle too notice.

2. "they got me looking at what was built with their product but I can’t see the actual code. Feels scammy" - What do you mean by you can't see the actual code ? You can just signup and use NonBioS to build software. And you can see the code written by NonBioS in multiple ways - ask it give you a downloadable zip, ask it to checkin the code to github, ask it to show you the code on the screen. Infact that the black boxes which scroll up, you can just expand them and see the code it is writing directly.

He is NOT talking about multi-agent systems, which is exactly why he is calling it an Agency. The author goes to great length to explain why this is NOT a multi-agent system because it can be easily misunderstood to be that.

This isn't multi-agents at all. Infact if you read the article in detail, you will realize that the author goes in detail to explain how this system is different from multi-agents. And this is exactly why the author calls it "Agency" because it is fundamentally different from multi-agents.

I agree that multi-agent doesnt work in practice. But this isnt that.

So what we do at NonBioS.ai is to use a cheaper model to do routine tasks, but switch to a higher thinking model seamlessly if the agent get stuck. Its most cost efficient, and we take that switching cost away from the engineer.

But broadly agree to the argument of the post - just spending more might still be worth it.

This will not end well.

Most of our stuff is built in house actually, simply because everything else is still kind of catching up. You can find a bunch of information on the blog (https://www.nonbios.ai/blog)

The only software that we use is Langfuse for observability and that too was breaking down for us. But they launched a new version - V3 - which might still work out for us.

I would suggest to just use standard non-AI specific python libraries and build your own systems. If you are migrating from N8N to a self hosted system then you can actually use NonBioS to build it out for you directly. If you join our discord channels, we can get an engineer to help you out also.

It is AI Software Dev called NonBioS.ai

As someone who works for a company having a real Agent in production, (not a workflow), I cannot disagree more than the very first statement here: Use Agent Frameworks like Langraph. We did exactly that, and had to throw everything away just a month down the line. Then we built everything from scratch and now our system scales pretty well.

To be fair, I think there might be a space for using Agent Frameworks, but the Agent space is too early for a good enough framework to emerge. The semi contrarian though, which I hold to a certain extent, is that the Agent space is moving so fast that a good enough framework might NEVER emerge.

Yes, but we dont believe that this is a 'fundamental' problem. We have learnt to guide their actions a lot better and they go down the rabbit a lot less now than when we started out.

That is very accurate with what we have found. <thinking> models do a lot better, but with huge speed drops. For now, we have chosen accuracy over speed. But speed drop is like 3-4x - so we might move to an architecture where we 'think' only sporadically.

Everything happening in the LLM space is so close to how humans think naturally.

Yes it works really well. We do something like that at NonBioS.ai - longer post below. The agent self reflects if it is stuck or confused and calls out the human for help.

So managing context is what takes the maximum effort. We use a bunch of strategies to reduce it, including, but not limited to:

1. Custom MCP server to work on linux command line. This wasn't really a 'MCP' server because we started working on it before MCP was a thing. But thats the easiest way to explain it now. The MCP server is optimised to reduce context.

2. Guardrails to reduce context. Think about it as prompt alterations giving the LLM subtle hints to work with less context. The hints could be at a behavioural level and a task level.

3. Continuously pruning the built up context to make the Agent 'forget'. Forgetting what is not important is what we believe a foundational capability.

This is kind of inspired by the science which says humans use sleep to 'forget' not useful memories and is critical to keeping the brain healthy. This translates directly to LLM's - making them forget is critical to keep them focussed on the larger task and their actions alligned.

It only seems effective, unless you start using it for actual work. The biggest issue - context. All tool use creates context. Large code bases come with large context out of the bat. LLM's seem to work, unless they are hit with a sizeable context. Anything above 10k and the quality seems to deteriorate.

Other issue is that LLM's can go off on a tangent. As context builds up, they forget what their objective was. One wrong turn, and in the rabbit hole they go never to recover.

The reason I know, is because we started solving these problems an year back. And we aren't done yet. But we did cover a lot of distance.

[Plug]: Try it out at https://nonbios.ai:

- Agentic memory → long-horizon coding

- Full Linux box → real runtime, not just toy demos

- Transparent → see & control every command

- Free beta — no invite needed. Works with throwaway email (mailinator etc.)

https://nonbios.ai - [Disclosure: I am working on this.]

- We are in public beta and free for now.

- Fully Agentic. Controllable and Transparent. Agent does all the work, but keeps you in the loop. You can take back control anytime and guide it.

- Not an IDE, so don't compete with VSCode forks. Interface is just a chatbox.

- More like Replit - but full stack focussed. You can build backend services.

- Videos are up at youtube.com/@nonbios

No we dont use QEMU - never heard of them till now. We built our own software from scratch - using Ubuntu - for AI. We are completely on the cloud. Every user gets a full Ubuntu Cloud VM for his NonBioS AI Engineer to work on.

We covered this a fair bit on our blogs: - https://www.nonbios.ai/post/why-nonbios-chose-cloud-vms-for-... - https://www.nonbios.ai/post/private-linux-vms-for-every-nonb...

I also did not get it, but now I get it a bit, I think.

Look at it this way. You have to get some work done - maybe book a flight ticket. So you go to two sites - first you go to flight fare comparison, then you book the ticket on the airline website. And you have to do it in code.

There are two ways you can do it.

First Way 1. Understand the API of the flight comparison portal. 2. Understand the API for the airline website. 3. Write code which combines both these API and does the task.

Second Way 1. Message a coder friend who knows the API of the flight comparison portal and ask him to write code to get the cheapest flight. 2. Message another coder friend who knows the API of the airline portal and ask him to book a flight.

Both ways are possible, but which one do you think is Less Work ? Which one is 'cognitively' easier ? Which one can you do while driving a car with one hand ?

It should be clear that the second way is easier. Not only is the second way easier, but if the task requires multiple providers and a lot of context, it might be the only way possible.

The first way is analogous to LLM's doing API calls. The second way is analogous to LLM's doing MCP Servers. MCP servers reduce the cognitive cost to do a task to the LLM - which dramatically increases their power.

Very cool product !

We, at NonBioS.ai [AI Software Dev], built something like this from scratch for Linux VM's, and it was a heavy lift. Could have used you guys if had known about it. But can see this being immediately useful at a ton of places.

Key questions:

1. The key data point seems to be Figure 6a. Where it compares performance on BABILong and claims Titans performance is at ~62%, as compared to GPT-4o-mini at ~42% for 100k sequence length.

However, GPT-4o and Claude are missing in this comparison - maybe because they perform better ?

2. There is no example provided of the Neural Memory Module in action. This is the first question I would ask of this paper.

> Doctors will often recommend exercise, but I find that these days even moderately strenuous exercise like riding a bicycle destroys my sleep quality for several days. There's something about it that appears to be too physiologically stressing, even though ten years ago I was a happy as a regular gymgoer.

I had something similar like this. I think I was able to fix it. The theory is that your sleep is still poor, even though you sleep through the night. This is causing high cortisol levels during day time and higher resting heart rate. This is elevated further after doing moderate exrercise and takes a long time to get back to normal as your sleep isnt adequate. If your heart rate doesnt go down enough, then your sleep quality gets destroyed.

The solution, for me and I am guessing for you, is this: Stop the cycleing. First fix sleep. Track it using Wellue O2 Ring. If the scores are not good, the reconfigure CPAP - use sleepapnea reddit for inputs. Once sleep is sorted as per O2 Ring, then it might take a few months for you to recover. After that you can restart moderate exercise and things should be fine.

So we are still on V2.7 - works pretty good for us. Havent tried V3 yet, and not looking to upgrade. I think the next big feature set we are looking for is a prompt evaluation system.

But we are coming around to the view that it is a big enough problem to have dedicated saas, rather than piggy back on observability saas. At NonBioS, we have very complex requirements - so we might just end up building it up from the ground up.

Thanks for the pointer !

We are actually toying with building out a prompt evaluation platform and were considering extending langfuse. Maybe just use this instead.

Bunch of them : Langsmith, Lunary, Phoenix Arize, Portkey, Datadog and Helicone.

We also picked Langfuse - more details here: https://www.nonbios.ai/post/the-nonbios-llm-observability-pi...

I did a similar test and tried to pull up certain categories of individuals I am interested in, with their names and linkedin profile links. ChatGPT hallucinated the names and the links.

I simply cannot move to a search, where there is random hallucination because having to check for each and every result for hallucination defeats the purpose of search itself.

You dont have to deal with it. Only a question of time before better AI comes out and starts refactoring this to better looking code. These are just growing up pangs and one of those problems which should go away on their own.

The bigger problem is what to do of the humans who are fixing this sloppy code right now.

Biggest issue i have with mosh is that it you cant scroll up the history. It is kind of a deal breaker for me.