April 20, 2026 · 1 min read
Analytics Without Seeing the Data
AI agents can clean and explore a dataset in minutes without you ever looking at it. What that does to analyst intuition, and why the pipeline becomes the product.
causal analytics,
agent-first
an analyst testing AI agents against real data — logging what works, what breaks, and where the job is going.
follow on x ↗latest lab notes
all notes →April 20, 2026 · 1 min read
Massive Datasets Are Built, Not Found
Large datasets aren't found, they're built. Why the real edge is control over how data is generated, and what a self-feeding collection loop looks like.
April 09, 2026 · 2 min read
Current Workflow for Data Analytics
How I run exploratory data analysis with Codex and Jupyter, what a business analytics master's didn't teach about messy data, and the bottlenecks left.
April 08, 2026 · 1 min read
The Beginning of the Agent's Final Form
An AI agent is three swappable parts: markdown files holding your preferences, an LLM that issues commands, and a Unix shell with cron. Why that matters.
March 25, 2026 · 1 min read
Agentic EDA Is Still Bottlenecked By Framing
Agents make exploratory data analysis cheap, but vague or shifting questions still sink it. Three rules for framing the analysis before the agent runs.
open predictions
all predictions →
#001
open
Coding becomes a commodity: by mid-2027 we are coding with open-source agents at close to the cost of the energy they use, and Opus 5.5-level capability is 'good enough' for almost all coding tasks.
#002
open
Data increasingly sits behind MCP. By the end of 2027, everyday professionals ask their own agent to find, query and analyse data, rather than having it presented to them in a dashboard.
#003
open
Making becomes an extension of language: by the end of 2027, 'making' (programming with code) is so everyday that it is simply part of how we speak and write.
usage signals
last updated: september 2026API spend
september 2026: $30.00
plan spend
september 2026: $20.00
all providers combined.