2025
April – June
-
If you're into search operations and architecture, we're entering into one of the most interesting periods in a long time. And it's not because of embeddings, vector DBs, hybrid search, or RAG based chat. It's because retrieval has a whole new customer base in the form of agents…
-
If you're a dev and you can explain how you write and assemble code in clear, effective English then don't worry. YGMI.
-
There is so much arbitrage opportunity sprinkled throughout daily life right now, just with knowing how to use the consumer AI services much less the APIs and more complex machine learning.
-
It's cool how @cursorai seems to be adding a sequence of my viewed code into a context buffer. I was looking at several relevant code chunks, and that seemingly informed the tab completion when I placed the cursor where I wanted to create new functions in another file.
-
tab completion in @cursorai is starting to feel like it should be called thought completion
-
It's generally true I can generate and iterate to useful code more quickly with AI. However, a more interesting part of the experience is that it's a different level of energy required. Also, if I'm generating code to help with analyzing some data, I don't feel like I'm switchin…
-
. @ProgramWithAi shared a talk about how MCPs are boring and it can be more powerful to just provide the LLM with a tool like eval(). It reminded me once again, how much we need to relinquish certain kinds of historic controls. These tools need different guards.
-
How Anthropic defines agents and how PocketFlow abstracts them https://codycollier.com/lx/2025/anthropic-definitions-and-pocketflow.html?x-2025-06-07x
-
The cost and time to produce code is trending to zero. This isn't just a disruption to the cost of doing business. It's also opening up new green fields for implementation patterns and options. Just in time code generation is one example. At the same time, we are upending user e…
-
Ok. It's time to learn about the internal mechanisms of agents. https://codycollier.com/lx/2025/journey-to-agent-understanding.html?x=2025-06-05
-
This post about AI augmented programming is getting lots of attention. It's interesting and has plenty of pointed remarks. The responses though, across communities, read like a Rorschach test. https://fly.io/blog/youre-all-nuts/
-
I recently had the opportunity to drop in and refactor an 8k line code base that was completely generated through iterative vibe-ish coding. After many hours of refactoring I ..felt.. this metaphor of sculpting down a nice block of marble to something more refined. The meta… htt…
-
This is a cool addition. It's more than just chunking too. Generally, lexical search algorithms like BM25 and TF-IDF are tailored for a world of whole documents. Then, lots of modern embeddings and semantic retrieval benefit from smaller text chunks (and maybe prefix/suffix… htt…
-
Working with LLMs feels a lot like the early experience of playing Adventure.
-
This paper is interesting: https://openreview.net/forum?id=wYGBWOjq1Q At a high level, this is the second time I've run across someone showing a certain kind of pre-training (vs random initialization) doesn't help, on a certain kind of downstream training and task. The other was…
-
A nice description of the unusual implications of adding well integrated AI assistance to programming https://registerspill.thorstenball.com/p/theres-beauty-in-ai
-
A pleasant telnet server exit banner:
-
Let's check in on cursor usage... (500-286) - ((286/18) 12) = 23.3 whew. 23 to the good.
-
In evals, LLM as Judge is just Looks Good To Model
-
Have some fun. Click through to the original post and respond to try this... https://x.com/ProgramWithAi/status/1914069084100059597
-
It's great to see @vespaengine continue to gain ground via a significant product like @perplexityai https://x.com/vespaengine/status/1912130460496375946
-
I'm finding these social and psychological aspects of building things with AI to be equally interesting if not more than the tech aspects. https://x.com/ProgramWithAi/status/1910367734421930451
-
A snapshot of a pdp11 programming card. Tech back then had nice aesthetics all around. Partly for @ProgramWithAi who has been posting pdp stuff. Also, let's give a nod to @usgraphics for keeping some of this spirit going.
-
Great thread on what async + agents may imply for search system architecture https://x.com/jobergum/status/1909687138532176195
-
What AI is doing is separating people who prioritize outcomes from people who prioritize internal craft.
-
A coworker needed some fun images for a presentation on our recent AI work. I grabbed a screenshot of htop, while the modeling program was running, not too different from the one here. Apparently htop was first released in 2004, which is a while back, though still new relative t…