Severyn Bell

Building AI systems that actually work. Helping your Business and life becoming AI native
Researcher, Builder and Engineer

[Severyn Bell @SeverynBell]

How I feel about recent news:
> AGI is here, our IPO depends on you believing it

[Severyn Bell @SeverynBell]

@waterloo_intern crazy work. Mad respect! Reading it...

[Severyn Bell @SeverynBell]

The smarter someone is, the worse their advice can be. They forget what it’s like not to know and assume you’ll just “get it”

[Severyn Bell @SeverynBell]

AI rewards depth: infra veterans moved into model serving, while 20-year researchers focus on specific LLM problems.
But the application layer feels like chaos: people jump between unproven ideas, declaring each one obsolete before you can even test it.

[Severyn Bell @SeverynBell]

@chaumian Is this a good read for an RL beginner?

[Severyn Bell @SeverynBell]

@zodchiii This set up keeps shipping garbage while you sleep 😄

[Severyn Bell @SeverynBell]

@thekenndubisi What's your issue with building things exactly? You can't distribute something until it's built, which takes time.

[Severyn Bell @SeverynBell]

@dhruvtwt_ Extremely real

[Severyn Bell @SeverynBell]

@jhoangeth It literally happened to me rn. Used Sonnet, asked 3 questions, and asked to plan. This is insane

[Severyn Bell @SeverynBell]

@zuess05 These "comment bait" posts are so funny man.

[Severyn Bell @SeverynBell]

@0xCygaar You guys all run the same "write me a viral post" agent.

[Severyn Bell @SeverynBell]

@svpino You guys all run the same "write me a viral post" agent.

[Severyn Bell @SeverynBell]

@jordymaui You forget to mention that it costs from 5-20 USD 😄

[Severyn Bell @SeverynBell]

@haider1 L framing

[Severyn Bell @SeverynBell]

@prathamgrv does this work only for ML paper or could be applied to other more complex topics?

[Severyn Bell @SeverynBell]

@gregisenberg It's not how his auto research works. It adjusts a small number of parameters in a well-defined narrowed problem. What you're describing is a marketing fluff.

[Severyn Bell @SeverynBell]

@shinfxh @grok explain 😂

[Martin Bauer @martinmbauer]

Detects 29% more cancer cases while being faster and more efficient

All the hype and doom about AI overshadows what a step-change it already is

[Rohan Paul @rohanpaul_ai]

The paper says the best way to manage AI context is to treat everything like a file system.

Today, a model's knowledge sits in separate prompts, databases, tools, and logs, so context engineering pulls this into a coherent system.

The paper proposes an agentic file system where every memory, tool, external source, and human note appears as a file in a shared space.

A persistent context repository separates raw history, long term memory, and short lived scratchpads, so the model's prompt holds only the slice needed right now.

Every access and transformation is logged with timestamps and provenance, giving a trail for how information, tools, and human feedback shaped an answer.

Because large language models see only limited context each call and forget past ones, the architecture adds a constructor to shrink context, an updater to swap pieces, and an evaluator to check answers and update memory.

All of this is implemented in the AIGNE framework, where agents remember past conversations and call services like GitHub through the same file style interface, turning scattered prompts into a reusable context layer.

----

Paper Link – arxiv. org/abs/2512.05470

Paper Title: "Everything is Context: Agentic File System Abstraction for Context Engineering"

[Martin Bauer @martinmbauer]

Correct! Just as a reminder: this is what a Transformer found after looking at 10M solar systems