Talks, long reads and apps, alongside the notes.
Data job title family tree
83 data job titles, from the 1950s keypunch operator to the AI engineer: an interactive family tree with the history and sources of each title.
AI Will Make Your Organization Obsolete. If You Have One.
The same agentic strategy in two organizations, two opposite outcomes. Agents do not just automate tasks, they dissolve positions, and only what someone answers for remains.
Agentic Analytics Benchmark: What Does a Good Answer Cost?
Haiku, Sonnet, Opus and Fable, at every effort level, on a simple analytics question and then a vague one. The simple one separates no one, the vague one separates everyone.
The Bottleneck Will Always Be Human
In three years, the Jolimoi Data Team grew its pull requests per person sixteenfold. The bottleneck moved four times and always came back to the same place, human time.
Graphe × Ontologie
A two-level 3D view: the ontology (skills, rules, agents) set above the graph of gold tables. The what below, the why and the how above.
The story of a data team that outlived its own job
The follow-up to Once upon a time, Agentic Analytics, on video: 28 minutes on a data team that outlived its own job.
Once upon a time, Agentic Analytics
The story of a data team and of the job that outlives it: a lean bet, automation, then agents, until the Data Analyst role itself gets rethought.
Build, buy… or AI ?
Rebuild everything in-house, thanks to AI and because of it? We play along, with a few exceptions. What we buy is no longer what we think we buy.
MySQL to S3 with AWS DMS: What actually works in production
MySQL to S3 with AWS DMS: seven production lessons, from partitioning to binlogs, to ingest hundreds of TB with almost no maintenance.
The Hidden Traps of Scaling Data Operations, and How to Beat Them
Scaling data operations: five rules and the traps they avoid, drawn from conversations with Heads of Data, CTOs, CEOs and CPOs.
Multi-label Classification Problems with Large Language Models
Multi-label text classification with an LLM, without a training set: few-shot prompting, token costs, metrics and production. With Martin Labenne, based on our article on the Jolimoi tech blog.
Databricks Cost Management
Databricks bills you twice, for the cloud and for DBUs. Reading the bill, tagging, querying system tables, choosing compute and doing the Photon math.