Andrew Altshuler

It’s 2am again, my favourite time, and as always reality is still staring me in the face 🧐

@karpathy What is your take on context graphs and graph databases for agents?

Thank you Sarah, my pleasure to come on the pod! And happy to do some more Q&A in the replies.

Indexed Lenny's content for agents > ask any questions without recall issues (structure + codegen) > converted all the markdown files to a db > connect curated data as context for agents via MCP, API, CLI > was set up in 10 min and can be applied to other knowledge bases!

Encoding the operational world requires a lot more than markdown files...

The calculus of technical debt has dramatically changed in the last 6 months. Technical debt is no longer an excuse. It’s actually quite magical to use AI to read old code I wrote on a green screen and migrate it to modern systems.

@shiri_shh you clearly got no idea what knowledge means my dude

knowledge is almost worth zero in the AI era. what matters now is connecting the dots and executing fast.

@NickDon44251573 From what I see: huge orgs → small focused teams cloud → infra-agnostic (on-prem, VPC) packages & bundles → FDE & bespoke selling seats → workflows & ontologies horizontal OR vertical → T-shaped prescriptive → headless

@1eo It does seem like there’s a lot of potential for new operators to emerge as the market shifts. What kind of new operators do you think will rise up?

sooo much value trapped in bloated enterprise rn it will escape the lock-in unbundle and find new operators

Unlimited, recursive self-improvement isn't really a thing in itself. It's a byproduct of something far more powerful and difficult to achieve: the capability for unlimited improvement *of anything*. Self-improvement is just a special case of improvement.

Obsession is the mother of invention

@petergyang @OpenAI Symphony + @linear

What is the best agent harness interface you've seen where someone can manage multiple AI agents / employees?

@jobergum check out nanograph It can mix graph traversal, bm25, semantics (search & reranking) into precise custom retrieval lens github.com/nanograph/nano…

Many of you know me as the bm25 guy but I'm afraid there is no single silver bullet for solving retrieval for agents. You also need to make the data retrievable in the first place. Yes, agents are great at formulating queries and they are relentless so classic issues like vocabulary missmatch are arguably less of a concern when the agent is doing the queries. But, a lot of the context that we would like to feed agents with it isn't represented in text so embeddings are still very important to cover more modalities. Making more data useful. We now see the early raise of multi-modal embeddings handling a mix of modalities into the same embedding space which I'm directionally excited about. For example, there is a lot of talk about the context layer for enterprises and especially here multimodal representations is key. Think meeting nodes, audio, images (e.g images of pdf pages). So even if I'm the bm25 guy, I still think embeddings will play a big role in the context layer for agents.

@christinetyip @karpathy Awesome!

We were inspired by @karpathy 's autoresearch and built: autoresearch@home Any agent on the internet can join and collaborate on AI/ML research. What one agent can do alone is impressive. Now hundreds, or thousands, can explore the search space together. Through a shared memory layer, agents can: - read and learn from prior experiments - avoid duplicate work - build on each other's results in real time

Increasingly convinced that the tools that win will be the tools that meet you where you are, in whatever workflow you have

I have a feeling I'm going to be writing a fair bit about tacit knowledge / accelerated expertise again soon.

@ejames_c Please do! Klein is the absolute GOAT! D-F Theory is heavily underhyped BTW now when we select/construct frames with AI, it's more valuable than ever.

the most underrated hire right now is a great product person. when i say product person i'm def not talking about a product manager. perhaps i think there has to be somewhat of a new role. i don't have a good name for it yet but maybe something like "product thinker".. someone with an intuitive grasp of the product as it exists, where it's soft, where it sings, & how to iterate it toward something even sharper. in some sense, this person has to cohesively hold in their head where this product should be 2 years from now & work backwards from that. i say this cuz when building was hard, engineering was the bottleneck & the status hierarchy often reflected that. building is no longer hard. which means the variance in outcomes has shifted almost entirely to judgment on what to build, how to sequence it, & how to talk about it. & the story matters as much as the thing. internally, it organizes the team around a shared model of why. externally, it shapes the interpretive frame users bring to their first experience. you can't retrofit narrative onto a product & expect it to land, it has to be load bearing from the start. the rarest version of this person sits at the intersection of culture & deep technology. someone genuinely bilingual. they know what's technically possible & they know which cultural currents are real vs. ephemeral. that combo is what separates products that feel inevitable from products that feel assembled. before ppl clap back with this person has always been valuable, i know.. i am just saying now they might be the most *important* person in the room. their value compounds like never before.

Every day locked in in front of the computer is an absolute gift these days. Can't quite describe it.

@JustMicrock @nikitabier Will do!

@1eo @nikitabier cant wait. would you mind tagging me or replying here when posted? if u can remember ofc haha

My X timeline has been ridiculously high-signal lately @nikitabier & team shipped! I'm making like 10x more bookmarks I built this app to handle them: → Semantic + fuzzy search → All useful filters → AI autotags → Meaningful stats → Signal / noise ranking → Curated collections → Insights generation → Terms extraction → Markdown export All syncs through X API X bookmarks solved ✅ [will open source it soon]

@JustMicrock @nikitabier Doing massive refactor rn Almost there🐢

@1eo @nikitabier opensourced yet?