Claude Code Changed How I Work (Part 1)

  • Author/Source: Scott Cunningham (Baylor), via Substack ("Causal Inference")

  • Original: https://causalinf.substack.com/p/claude-code-changed-how-i-work-part

  • Key Ideas

  • Claude Code is categorically different from "vibe coding" (copy-paste chatbot workflows) — it lives inside your local directory, reads all files, and executes code autonomously in an iterative loop.
  • The core tools (Read, Write, Edit, Bash, Grep, Glob, WebFetch, WebSearch, Task) let Claude Code scan a project, run scripts, see errors, fix them, and re-run without human copy-paste.
  • The CLAUDE.md file provides persistent project context across sessions, solving the "every session starts from zero" problem of chatbot-based coding.
  • Vibe coding creates an "attention problem" — compressing time inputs reduces learning and sustained attention even while tasks get completed. This is especially acute for ADHD.

  • Cunningham found the tool so transformative that he upgraded to $200/month within days and considers it irreversible ("I cannot go back").

  • Summary

Cunningham, an applied microeconomist who has coded in Stata for 20 years, describes his trajectory from traditional coding to vibe coding (2023–2025) to Claude Code (November 2025). He explains that vibe coding — prompting a chatbot, pasting code, running it, pasting errors back — was an improvement but suffered from session discontinuity, attention erosion, and the inability to operate on local files. Claude Code changed this by becoming an agentic system that inhabits the project directory, reads all files, executes code, and iterates autonomously.

The post frames the shift through the lens of ADHD and the economics of attention. Cunningham argues that generative AI compresses time inputs for cognitive tasks, which can undermine both learning and sustained focus — a tradeoff that agentic coding mitigates by maintaining project context and automating the mechanical loop. This is Part 1 of a planned multi-part series; subsequent posts cover collaboration workflows, ML/NLP projects, web scraping, and presentation building.

  • Relevance to Economics Research

Directly relevant as a first-person account from an economics professor adopting AI-agentic coding. Frames the transition in terms economists understand (stated vs. revealed preferences, human capital, isoquants). Highlights the practical workflow shift from chatbot-assisted coding to autonomous agentic coding, and the attention/learning tradeoffs that matter for research productivity.