2026
October – December
-
Loop(search, aggregate, synthesize, review)
-
I need to improve my flight check system around code delivery.
-
A truth in the agentic dev era. <https://twitter.com/attentionmech/status/2119934719471345781
-
Directory structure and naming conventions are a great way to help constrain and guide agentic development.
-
It seems like it takes a lot of work to learn how to use AI effectively as an individual. Then, getting a team to be effective And aligned is even more challenging. I have a hard time believing the industry is anywhere close to diffusion and operational efficiency.
-
Spending time organizing counts as work too. It's even more important now that you can run so fast down the wrong path with AI.
-
Maybe my agent needs a 3D printer... <img src="/lx-mini/2026/images/x-2119137907172860407-32119137825073287170.png" alt="photo" loading="lazy" decoding="async"
-
It would be nice if it was more common for universities to offer or even require courses on open topics across various fields. It could be great for engaging early research thinking and framing the field for those interested in grad school and research. <https://twitter.com/adam…
-
Today went well, because Working code wins
-
Doing some prototyping with Jev from @typesafeai this week. These system one models are really neat. And they are very friendly for vibe prototyping and rapid exploration of product ideas.
-
From a conversation this evening... I'm not really "writing" code. So it feels weird to say that. I sometimes say "generated" code which feels more accurate but weird. The conversation wove through TV, movies, and film making. Out came the suggestion from my father in-law to use…
-
Make your prompts simple. Minimize mcp, minimize local tools, minimize agents md, minimize your input. Taking a little time to craft well written instructions will foster better results with the current models.
-
Adventure ahead <img src="/lx-mini/2026/images/x-2117396588729364978-32117396581079019520.jpg" alt="photo" loading="lazy" decoding="async"
-
Another super fun idea, where a model just changes another model's weights directly to improve outcomes. A model mechanic giving another model a tune-up. <https://twitter.com/rikiyatakehi/status/2117313742891327994
-
My paper reading has fallen off lately. Time to go back to printing them out.
-
Both substrates and fields can induce form in matter
-
Corral your attention, then guide it
-
A good thread about the dualities of science <https://twitter.com/niroshajmurugan/status/2116923853519618462
-
There are some wild things happening lately, where the weights and related processes of LLMs are being explored as their own substrate. Computational processes, subliminal storage, etc. <https://divagr.com/blog/qwenputer
-
This is cool and a creative example of adapting Jev / decision models to various tasks <https://github.com/jkrup/jeveryword
-
One big advantage of LLM APIs is simplicity of use. Two big inputs, each free form text. One output of arbitrary text. This simplicity fosters adoption within the software domain. The system one models require more work to frame problems for effective application. They are super…
-
The day is done. Happy birthday bek. I think about you all the time. A favorite memory: Early morning, heading to the office / data center. We'd probably just finished breakfast at Denver diner but I don't recall. We're heading up Speer in the Saturn wagon. I'm in the passenger…
-
Feels like things in AI have really sped up again this Fall
-
I guess context engineering is the new feature engineering. It's still kind of wild how powerful end-to-end learning and learned representations can be.
-
loop(org --> do) <https://twitter.com/cmcollier/status/2115440938562519112
-
Most of the constraints which hinder my life are self imposed.
-
Remember how Dr. Isbell, in his Machine Learning class, said he didn't care about where you got the code for your projects? That comes to mind a lot. It's especially interesting now.
-
I continue to be impressed by the zero shot ability of the leading models.
-
It's ok to let some of the details fade away. It's less ok to let go of the intuition learned from the details.
-
Really small system one models are going to be cool
-
BM25 and decision models are all you need? <https://twitter.com/bowangbo/status/2114788475363041739
-
All the AI bug finding and security exposures show us that human code review is over-valued.
-
Vibed edges, stable core
-
Org friction is the enemy of agency
-
Someone reminded me today: Working code wins.
-
Fun thread <https://twitter.com/drmichaellevin/status/2113248329949712646
-
One of the best things about AI code gen is not just that I can make local tools quickly, but that they can also be tailored and made to look nice. It's such an improvement to my daily work.
-
Software no longer has to be a team sport.
-
This looks pretty wild. It's going to take me a while to read but it will be fun. <https://twitter.com/elonlit/status/2112957586886860880
-
Behind the phrase vibe code is a masked problem. There is a continuum of good to bad engineering and adding AI to bad engineering will mire a team in problems and slow everything down.
-
This is a nice analysis, as expected, from Hamel and Antaripa. Good illustration of more traditional analysis in the newer domains of LLM evals and decision models. For critique I would say it's pushing the boundaries of decision models by using a very broad prompt and single de…
-
I am patterns in an excitable medium. I am a breathing time machine.
-
There are mistaken assumptions, all around us, just waiting.
-
Reasoning distills to intuition
-
Patterns in System One space: . Zero shot general classifier . Zero shot general slot filling ... What's next? <https://twitter.com/michellechen/status/2112310950309355943
-
Using the word Taste when talking about AI seems like sidestepping the hard work of describing more specific skills which are a value add.
-
Phrases that need to see increased use given some techniques growing in popularity: . Black box optimization . Bayesian optimization . Derivative free optimization . Stochastic hill climbing
-
A huge part of leveraging AI is being able to evaluate and encapsulate risk. High risk? Slow down and be deliberate. Low risk? Just rip and lfg
-
Building custom local tools is a crazy unlock. It's so useful for analysis, research, and development in data centered domains. I was doing this at a startup before code gen with AI took off. It was a useful tactic for me at the time. Looking back, it was so slow and primitive c…
-
Codex with training wheels? Cursor in learning mode? In coding agents, I would like to see a mode that only allows creating one function at a time. The early days of coding agents were only that good. Guide it a function at a time. Carefully arranging modules and then populating…
-
Cool framings for Jev and System One models that I've run across: Zero shot general classifier Evidence based judgements Common sense decisions
-
AI coding. Decreased friction for generation. Increased value of discrimination.
-
codex with GPT-6.1-Sol low is my latest workhorse. works well for general code stuff.
-
I've been tinkering with the chat agent for my custom Hacker News index. The mechanics are getting a lot easier these days, and it's mostly about the prompt, tool design, and tool decsriptions. <https://chat.orangewords.com?x=2026-10-17
-
Recently I spent time exploring Jev and their System One Model concepts. If you only read one thing about system one models, this is one of the best for engineers: <https://docs.typesafe.ai/concepts/how-to-build-with-system-one
-
When you are trying to optimize a developer workflow, put friction in the paths of the agents not the humans.
-
Poor work is poor work whether human created or AI generated.
-
Another day of synthesizing order from chaos.
-
I've frequently thought about one useful abstraction being that LLMs are a bridge from natural language to structured data and interfaces. Lately I'm realizing the obvious: how much of a two way bridge it can be, allowing data-to-narrative patterns.
-
First, there are fundamental principles. Then, there are rules and ideas derived from those principles. Then, those derivations are encoded into intuition which removes the derivation chain.
-
I'm fond of the concept: functional core, imperative shell Lately with AI I'm thinking: stable core, dynamic adaptive shell
-
Pleasantly thought provoking ideas from one of the leading edges of AI forward development practices <https://ghuntley.com/agency/
-
Code models are good at working with replicated patterns but it's you that should carefully initialize the pattern.
-
Even with AI, software seems the same... In planning: Decompose the desired outcome into small, sequenced units of work with intermediary, valuable outcomes. Ruthlessly reduce and defer scope. In architecture: Think in systems and compose units of encapsulated complexity. Decoup…
-
Carefully framing the problem to fit the known effective techniques is very important in applied machine learning. Yesterday and today.
-
A key value of applied LLMs is brute force assembly of all permutations of coherent language within a defined framing.
-
Sure, you can write your prompts like prose. Or you could write them more like poetry.