David Gasquez
LightNews LightNews
banner
davidgasquez.com
David Gasquez
@davidgasquez.com
5.1K followers 1.2K following 970 posts
Data @ Protocol Labs.

Open Data, Open Source, Open Protocols.

Walks taker. Progressive Metal enjoyer.

davidgasquez.com
Posts Replies Media Videos
Reposted by David Gasquez
simonwillison.net
At the risk of starting the flame war to end all flame wars...

Modern LLMs (GPT-5.1, Claude 4.5, Gemini 3) produce excellent code and can be a significant productivity boost to software engineers who take the time to learn how to effectively apply them - especially if used with coding agent tools
November 27, 2025 at 7:55 PM
davidgasquez.com
Been using Codex CLI for some tasks like tagging and entity extraction.

Turns out tasks get way easier when you force more small, local decisions without global context and then a few "aggregations" on top.

Accidentally rediscovered MapReduce again! 😅
davidgasquez.com David Gasquez @davidgasquez.com · Oct 22
I've been using this pattern to "specialize" Codex for vaguely defined tasks like classification, filtering, soft sorting, ...

davidgasquez.com/specializing...

Made more than 10,000 invocations so far (reusing my ChatGPT subsciption) and am really happy with the pattern!
’ve been using Claude Code to take care of scrappy data cleaning tasks for a while. These days though, I’m using Codex as my coding agent. Similar to what I did with Claude, I’ve been “fine-tuning” Codex CLI to work on a few different vaguely defined tasks like classification, voting, filtering, or ranking. The pattern in this post works surprisingly well when you have the following conditions: Loosely defined open-ended tasks. e.g., tagging tweets with a set of predefined labels, extracting structured information from a GitHub issue, … Powerful agentic capabilities. Doing the task requires something more than a simple llm call or PydanticAI script. e.g., using gh api CLI to get the number of stars of a repository. Structured outputs. You need a response in a certain shape! This is something codex exec can do that claude couldn’t and is really powerful. e.g., return exactly True or False and nothing else. Save money! Unlike llm or other tools/libraries that require an OPENAI_API_KEY, Codex can use your ChatGPT subscription, making things “free”.
November 22, 2025 at 10:31 AM
Reposted by David Gasquez
gordon.bsky.social
the two types of organizations: spreadsheets and cults
November 17, 2025 at 5:52 AM
davidgasquez.com
Awesome job with this new series @jed.co! Love the topics and conversations.
source.coop Source Cooperative @source.coop · Sep 22
🎙️Get ready for our inaugural podcast! Join us on Sept 30 at 11am PT with host @jed.co and guest @dbreunig.bsky.social for an in-depth conversation about the data products that fueled #AI’s rapid advancement and why they’ve reached their limits #LLM

Tune in on YouTube Live: greatdataproducts.com
November 11, 2025 at 3:03 PM
davidgasquez.com
I've been using this pattern to "specialize" Codex for vaguely defined tasks like classification, filtering, soft sorting, ...

davidgasquez.com/specializing...

Made more than 10,000 invocations so far (reusing my ChatGPT subsciption) and am really happy with the pattern!
’ve been using Claude Code to take care of scrappy data cleaning tasks for a while. These days though, I’m using Codex as my coding agent. Similar to what I did with Claude, I’ve been “fine-tuning” Codex CLI to work on a few different vaguely defined tasks like classification, voting, filtering, or ranking. The pattern in this post works surprisingly well when you have the following conditions: Loosely defined open-ended tasks. e.g., tagging tweets with a set of predefined labels, extracting structured information from a GitHub issue, … Powerful agentic capabilities. Doing the task requires something more than a simple llm call or PydanticAI script. e.g., using gh api CLI to get the number of stars of a repository. Structured outputs. You need a response in a certain shape! This is something codex exec can do that claude couldn’t and is really powerful. e.g., return exactly True or False and nothing else. Save money! Unlike llm or other tools/libraries that require an OPENAI_API_KEY, Codex can use your ChatGPT subscription, making things “free”.
October 22, 2025 at 6:38 PM
davidgasquez.com
I've been curating some "Agentic Coding" tips and tricks for a while. Check them out if you're into that!

davidgasquez.com/handbook/art...
David Gasquez
David Gasquez personal website
davidgasquez.com
October 16, 2025 at 6:12 PM
Reposted by David Gasquez
simonwillison.net
I've become one of those people who run multiple coding agents at once - here are my notes on embracing the parallel coding agent lifestyle simonwillison.net/2025/Oct/5/p...
Embracing the parallel coding agent lifestyle
For a while now I’ve been hearing from engineers who run multiple coding agents at once—firing up several Claude Code or Codex CLI instances at the same time, sometimes in …
simonwillison.net
October 5, 2025 at 1:56 PM
Reposted by David Gasquez
gergely.pragmaticengineer.com
When you try to predict about the impact of AI tools on software engineering - a great analogy I heard from @simonwillison.net is to look at the impact of smartphones (+ the amazing cameras they have) on filmmaking.

Noticed that those able to shoot pro films using a smartphone are... pros?
October 2, 2025 at 8:26 AM
Reposted by David Gasquez
pfrazee.com
I personally see collapse as an inevitability for every individual institution. Nothing lasts forever. The smart way to build systems is to enable reforms so that they can adjust to reality without destroying what works in the process.

This is pretty baked into the philosophy of Bluesky
April 24, 2024 at 10:02 PM
Reposted by David Gasquez
gordon.bsky.social
if only there were a decentralized, immutable, and versioned way to fix this!
conradhackett.bsky.social Conrad Hackett @conradhackett.bsky.social · Sep 14
LINK ROT: 38% webpages that existed in 2013 were no longer available 10 years later.

Even among pages that existed in 2021, 22% no longer accessible just two years later. This is often because individual page was deleted or removed on otherwise functional website.

Many implications for knowledge 🧪
A line chart showing that 38% of webpages from 2013 were no longer accessible one decade later.
September 15, 2025 at 6:57 AM
Reposted by David Gasquez
narphorium.com
I wrote up some thoughts about why we’re seeing a gap in AI tool use among different types of developers
narphorium.com/blog/top-dow...
1/4
Top-Down vs. Bottom-Up Development | Shawn Simister
Why do AI tools work for some developers but not others?
narphorium.com
August 6, 2025 at 4:59 PM
davidgasquez.com
Is there a name for the pattern where you try building the same thing (feature, project, ...) a bunch of times with different agents to get a sense of the implementation design space?

I usually run things at least 2 or 3 times until I "feel" what is the way I want it to be built.
September 4, 2025 at 9:22 AM
davidgasquez.com
Both Evidence and Observable Framework seem to be slowing down in development.

- github.com/observablehq...
- github.com/evidence-dev...

Not sure why. Anyone has any context or ideas?
August 27, 2025 at 10:07 AM
davidgasquez.com
My dotfiles repository is now mirrored on @tangled.sh too!

tangled.sh/@davidgasque...
August 12, 2025 at 2:49 PM
davidgasquez.com
Spent some time refining the approach and benchmarking it. Just published a blog post about the core idea.

Can a bunch of LLM Agents be used to rank an arbitrary set of items in a consistent way? 🤖

davidgasquez.com/ranking-with...
August 8, 2025 at 1:38 PM
davidgasquez.com
Wrote a post about a small pattern I've been using for a while. Replacing chunks of bash scripts with `llm` or inline Claude Code calls.

For example, for data cleaning/structuring.

davidgasquez.com/scrappy-data...
While working on a small analysis over highly unstructed data at work, I came up with a very hacky but effective and cheap way to clean and process messy data. There is almost no code involved. The main downside is that requires a Claude Code subscription for it to be cost effective!
davidgasquez.com
August 6, 2025 at 2:47 PM
Reposted by David Gasquez
gordon.bsky.social
usually the baseline is “nothing” so the right heuristic is usually not “is it ideal” but rather “is it better than nothing?”
July 29, 2025 at 3:52 PM
davidgasquez.com
Are there any benchmarks/evals that track model performance over time?

I've heard things like Claude/Cursor getting dummier over time (haven't noticed much myself) and am wondering if there is some data to back this.
July 25, 2025 at 6:46 AM
davidgasquez.com
New data oriented file format just dropped.

FastLanes, "like Parquet, but with 40% better compression and 40× faster decoding". 👀

Seems it can exploit correlations between columns and have fully SIMD friendly encodings to help with vectorization.

github.com/cwida/FastLa...
GitHub - cwida/FastLanes: Next-Gen Big Data File Format
Next-Gen Big Data File Format. Contribute to cwida/FastLanes development by creating an account on GitHub.
github.com
July 24, 2025 at 3:12 PM
davidgasquez.com
Asked Claude to move my handbook notes from Obsdian to my Astro website. Very happy with the results!

davidgasquez.com/handbook/dat...

I'll keep the Obsidian Publish version around while I work on making it look as good.
Data Culture The data team needs to be focus on delivering insights and supporting decisions. The outcome of the data team are decisions and a shared context across the organization that makes coordination easier. Your goal as a data professional is to facilitate decision making and help surface/investigate the performance of a business (e.g. operational). Learning to drive decisions quickly, a bias to action, is a critical competency for an analyst. Every skill you learn – communication, writing, experimentation, metric design – supports this. If analysis is not actionable, it does not really matter. Analysis must drive to action. Clear results won’t spur action themselves. The organization needs to be ready to pivot when something isn’t working. Data doesn’t make decisions, people do. Data’s impact is tough to measure — it doesn’t always translate to value The value of “insights” is often unknown. The Data Team should be building and iterating the Data Product. Notebooks are a workshop. Production systems are the factory. Not everything needs to be put into production. Not everything should be a notebook. You need both. Lean in to the strength of each. Data is fundamentally a collaborative design process rather than a tool, an analysis, or even a product. Data works best when the entire feedback loop from idea to production is an iterative process. To get buy in, explain how the business could benefit from better data (e.g: more and better insights). Start small and show value. Run Purpose Meetings or Business Metrics Review. Purpose Meetings are 30 min meetings in which stakeholders, engineers and data align on the goal of a release and what is the best way to evaluate the impact and understand its success. Align on the goal, commit on metrics and design the data. Data Culture The data team needs to be focus on delivering insights and supporting decisions. The outcome of the data team are decisions and a shared context across the organization that makes coordination easier. Your goal as a data professional is to facilitate decision making and help surface/investigate the performance of a business (e.g. operational). Learning to drive decisions quickly, a bias to action, is a critical competency for an analyst. Every skill you learn – communication, writing, experimentation, metric design – supports this.
July 23, 2025 at 10:23 AM
davidgasquez.com
Will 2025 finally be the year of the Linux desktop?

Really cool to see folks like @mitsuhiko.at or @ssp.sh trying out Arch, Hyprland, and other modern Linux stuff thanks to Omarchy!

More people, more polished experience, more dotfiles to learn from!
July 22, 2025 at 4:55 PM
davidgasquez.com
The end dataset also serves as a much smoother API!

- No key required
- Batch export friendly
- Explorable in the browser
❯ curl -s "https://huggingface.co/datasets/datania/aemet/raw/main/valores-climatologicos/2025/07/16.json" | jq '.[] | select(.nombre == "GRANADA AEROPUERTO")' { "fecha": "2025-07-16", "indicativo": "5530E", "nombre": "GRANADA AEROPUERTO", "provincia": "GRANADA", "altitud": "560", "tmed": "30,3", "tmin": "18,8", "horatmin": "05:08", "tmax": "41,8", "horatmax": "16:09", "dir": "11", "velmedia": "2,2", "racha": "10,3", "horaracha": "16:30", "sol": "13,6", "presMax": "955,3", "horaPresMax": "Varias", "presMin": "951,0", "horaPresMin": "16", "hrMedia": "29", "hrMax": "60", "horaHrMax": "Varias", "hrMin": "13", "horaHrMin": "16:02" } ~
July 21, 2025 at 1:58 PM
davidgasquez.com
Spent some time this weekend playing with AEMET's APIs and wrote a small post sharing what I learned and how I approached the project.

davidgasquez.com/exporting-ae...

I'm really enjoying working on these small, one-off projects as a way to experiment with different techniques. 🧪
This is a small post to share how that went and some learnings around AEMET APIs and how I developed the project (LLMs, uv, …). Code and data are open and available on GitHub and HuggingFace.
davidgasquez.com
July 21, 2025 at 10:56 AM
davidgasquez.com
Updated linkweaver (custom notes & research helper) with a couple of new tricks.

1. Saves the URL content locally
2. Accepts any command to preprocess the generated Markdown. E.g: `llm "Clean the following Markdown"`.

github.com/davidgasquez...

Want to try it? `uvx linkweaver` and enjoy! 😀
Screenshot of linkweaver running. It is downloading a few urls ❯ uvx linkweaver --help usage: linkweaver [-h] [--list-links] [--xml-cat] [-x COMMAND] [-v] [--retries N] [-q] [--dry-run] [--no-color] [-f] [input_files ...] Extract URLs from markdown files and save each link as individual markdown files in resource folders. By default, existing files are skipped to save time and bandwidth. positional arguments: input_files One or more markdown files to process options: -h, --help show this help message and exit --list-links, -l List all unique URLs found in the files (no downloading) --xml-cat Concatenate files with their resource folders in XML structure -x, --exec COMMAND Execute a shell command on the markdown output before saving (e.g., 'llm -t mdclean') -v, --verbose Enable verbose output with detailed progress information --retries N Number of retry attempts for failed URL fetches (default: 3) -q, --quiet Minimize output (opposite of verbose, but still show errors/warnings) --dry-run Show what would be done without actually doing it --no-color Disable colored output -f, --force Force redownload of files even if they already exist Examples: linkweaver notes.md # Process file, skip existing resources linkweaver --force notes.md # Force redownload all resources linkweaver --list-links notes.md # List all URLs found linkweaver --xml-cat notes.md # Concatenate with resources linkweaver -x 'llm -t clean' notes.md # Process with custom command linkweaver --retries 5 notes.md # Use 5 retry attempts for failed URLs linkweaver --quiet --dry-run notes.md # Preview actions quietly linkweaver -q -v notes.md # Quiet mode overrides verbose linkweaver --no-color notes.md # Disable colored output Common workflows: linkweaver --list-links *.md | head -10 # Preview first 10 URLs
July 16, 2025 at 3:18 PM
davidgasquez.com
One of the best things I'm using LLMs for is to make lots of small UNIX style CLIs! 🤩

I'm talking about simple/silly things like listing all the unique links in a bunch of markdown notes (`xl *.md`).

What a time!
July 16, 2025 at 9:32 AM