Writing
Archive · Tags: AI , Carpentries , Claude Code , DataSquad , DevOps , GIS , IMLS , LLM
-
GIS Librarians at US R1 Universities
A survey of dedicated GIS librarian positions at all 187 Carnegie R1 universities, and a look at UCLA, the only UC campus to cut the role. Originally published as a standalone report.
-
UCLA Research Data Infrastructure: an evidence base for One IT
The public evidence package I built for UCLA's 2026 One IT Research Working Group, carrying the Research Data Working Group's 2021-2024 findings into the campus review. Originally published as a standalone site.
-
Teaching bibliometrics with OpenAlex, and finding out my OpenAlex record isn't mine
Building an optional Library Carpentry episode on the OpenAlex API turned into a real lesson in author disambiguation when my own ORCID pointed at someone else's decades of research.
-
Fall 2026 UC Carpentries: organizing, teaching, promoting
A two-week online Carpentries series across UC campuses this September: Software Carpentry, Library Carpentry, and a Sharing Research Software session. I built the site, ran instructor sign-up and outreach, and I'm teaching.
-
What I don't let an AI agent decide
A tool for building Carpentries workshop websites with any AI coding agent. The interesting design choice wasn't the automation. It was where the agent's judgment stops and a hard rule takes over.
-
Library Leadership Is Stewarding the Wrong Thing
Ask what a library leader is supposed to protect and the answers you get are a building, a budget, and a collection. Those are means, not the mission. The job is stewarding the conditions under which knowledge gets made.
-
Rebuilding a Carpentries lesson on software citation taught me that the reframe that mattered wasn't a new standard. It was retargeting the whole lesson from researchers publishing their own code to librarians advising other people's.
-
The prompt that makes a model disagree with me
A standing prompt template whose only job is to stop a second model from agreeing with my own diagnosis. Tested it on my own project-management setup and it found the weak joint in an argument I was ready to act on.
-
The single underscore that broke my task system
A CLI task manager that worked fine in my terminal silently failed inside Claude Code sessions. The cause wasn't a bug in the tool, it was a shell-snapshot mechanism that quietly drops single-underscore function names before the agent ever sees them.
-
How I run three LLMs without getting burned
Claude, ChatGPT, and NotebookLM each do different work in my writing. The thing that keeps them useful is not the model. It's a verification habit borrowed from the reference desk: every claim gets ground-truthed before it ships.
-
Giving our lesson catalog real metadata
A couple weeks ago I noticed our OSPO lesson catalog leaned hard toward one pathway. This week I went in and fixed the metadata underneath it. The cleanup taught me more than the result did.
-
A working list of learn-by-doing open source resources
Gathering the interactive, guided, and gamified resources for learning open source and DevOps skills — the kind that teach by doing, like swirl did for R. A starting list to bring to the UC OSPO education group.
-
Three pieces I bookmarked this year that keep pointing at the same thing: a lot of the software running the world rests on one or two people, and almost no one funds them.
-
What would it take for a university OSPO to support its solo maintainers?
A working note: if critical open source rests on a few unpaid people, and some of them are inside the UC system, what could a university open source program office actually do about it?
-
First pilots have taught — here's what came back
First results from the external community pilots of our open science curriculum for librarians: lessons run long, funder policies move fast, and line-level feedback is gold.
-
The DataSquad model — what makes it work
Link to a DataSquad post on the undergraduate research consultant model, with commentary.
-
Open Science Curriculum for Librarians — IMLS project update
Summary of progress on the IMLS-funded open science curriculum, originally posted on the project site.
-
What research data infrastructure actually requires
A short note on the difference between data storage and real data infrastructure.
-
Building Programs That Outlast Their Founders
Most library programs die when the person who built them leaves. The ones that survive are not better — they are differently structured. Here is what that structure looks like.
-
What Research Data Infrastructure Requires from Library Leadership
Research data infrastructure is not an IT problem. It is a scholarly stewardship problem — and libraries are uniquely positioned to lead it, but only if we're willing to be more than service providers.
-
From Teaching Workshops to Building Research Infrastructure
How eight years of work at UCLA evolved from teaching data skills to building the systems that make open, sustainable research possible.