AI at Work: Two-Guide Content Production Session

**Date:** 2026-04-08

**Duration:** ~25 minutes

**Repos touched:** assortedLLMTasks

Context & Motivation

User wanted to create shareable guides for non-technical people on how to use AI effectively. The work evolved across three iterations:

1. Initial request: “Share a walkthrough of how to set up this environment for a non-technical person who hasn’t done anything like this before, with prompting and context best practices.”
2. Pivot: User realized a corporate audience wouldn’t have the same freedom to install CLI tools and set up their own environment. Requested a second guide focused on applying AI in a work context — web interfaces, corporate tooling, building custom tools without coding.
3. Enrichment: User asked “Are there any more specifics from my implementations that would be useful to mention/include?” — triggering a deep exploration of the memory system, feedback rules, guidance files, and project implementations to extract transferable lessons.

Decisions Made

1. Two Separate Guides Instead of One

• Decision: Keep the CLI setup walkthrough as-is and create a separate work-focused guide rather than merging them
• Alternatives considered: Rewriting the first guide to be more generic; adding a “corporate” section to the original
• Rationale: The audiences are fundamentally different. Someone setting up their own Claude Code environment has different needs than someone using claude.ai through a company plan. Merging would dilute both.
• Trade-offs: Some prompting content overlaps between the two guides

2. Anonymized Real Lessons vs. Generic Advice

• Decision: Drew from actual implementation mistakes and learnings (verify-before-asserting, dual-persist, graduated autonomy, pipefail bug, bakeoff results) but anonymized them for the corporate audience
• Alternatives considered: Keeping it fully generic; linking directly to the specific implementations
• Rationale: Generic advice (“be specific in your prompts”) is everywhere. Real stories with real consequences are what make a guide stick. But the audience doesn’t need to know about Discord webhooks or autonomous dev agents.
• Trade-offs: Lost some specificity in anonymizing, but gained relatability

3. CRAFT Framework for Prompting

• Decision: Introduced CRAFT (Context, Role, Action, Format, Tone) as the prompting mental model
• Alternatives considered: Using existing frameworks (RISEN, CO-STAR, etc.)
• Rationale: CRAFT is simple enough to remember and covers the five things that matter most. The acronym is intuitive and the letters map to natural questions you’d ask yourself.
• Trade-offs: It’s not the only valid framework — but having any framework beats having none

4. “Building Your Own Tools” Section Focused on Zero-Install Options

• Decision: Emphasized Google Apps Script, browser bookmarklets, HTML files, Sheets formulas, Slack Workflow Builder — all things that require zero IT approval to create
• Alternatives considered: Including Python scripts, npm tools, browser extensions
• Rationale: The audience is non-technical and in a corporate environment. Anything requiring npm install or admin access is a dead end. The listed tools genuinely require nothing but a browser.
• Trade-offs: More powerful tooling (actual scripts, APIs, automation platforms) is out of scope

What Was Built / Changed

Guide 1: Claude Code Setup Walkthrough

**File:** assortedLLMTasks/tasks/2026-04-08-claude-code-setup-walkthrough.md

**Commit:** 66efefd

8-part guide covering:

• What Claude Code is and why terminal-based
• Full setup: terminal install (WSL for Windows), Node.js, Claude Code, auth
• Basic usage and permission system
• The CLAUDE.md instruction system (project + personal)
• Prompting best practices (7 techniques with before/after examples)
• Context management (window limits, session hygiene, institutional knowledge)
• 4 practical workflow examples
• Common pitfalls

Guide 2: AI at Work Productivity Guide

**File:** assortedLLMTasks/tasks/2026-04-08-ai-at-work-productivity-guide.md

**Commits:** 5ad7914, 4fa4b0e

9-part guide covering:

• Mindset shift (search engine → junior colleague)
• Corporate AI landscape (what you can/can’t do, AI usage policies)
• CRAFT prompting framework with worked examples
• Few-shot prompting, chaining, constraints, “Act As” technique
• 5 real work patterns (first draft accelerator, data translator, review partner, learning accelerator, process builder)
• Building custom tools without coding (Google Apps Script, bookmarklets, HTML calculators, Slack automations)
• 6 lessons from real AI systems (mistakes→rules, verify don’t assume, dual-persist, context crash recovery, A/B test approaches, graduated autonomy)
• Common pitfalls (copy-paste trap, confidentiality, automation without understanding, overreliance)
• Measuring progress
• Quick reference prompt starters table

Learnings Captured

Drawn from existing memory/feedback (not newly created):

• feedback_verify_before_asserting.md → Lesson 2 (verify don’t assume)
• feedback_test_before_asking.md → Lesson 2 (“done” means verified)
• feedback_learnings_dual_persist.md → Lesson 3 (save in two places)
• feedback_pipefail_grep.md → Referenced in mistakes-become-rules framing
• project_claude_bakeoff.md → Lesson 5 (A/B test approaches, adversarial buying research)
• project_autonomous_dev.md → Lesson 6 (graduated autonomy, production deploy incident)

No new memory or guidance files created

This session was content production, not system/workflow changes. The learnings were already captured — this session repackaged them for an external audience.

Open Items & Follow-ups

• Consider posting to WordPress as standalone articles — these guides are long enough and useful enough to be blog posts on pezant.ca
• The two guides could link to each other — the CLI guide could link to the work guide for people who can’t install things, and vice versa
• Team sharing — the work guide is specifically designed to be shareable with colleagues. Consider distributing via the interview practice tool or buying assistant channels where non-technical people interact
• The prompting section overlaps — if a third guide is ever written, extract the prompting principles into a standalone reference that both guides link to

Key Files

• assortedLLMTasks/tasks/2026-04-08-claude-code-setup-walkthrough.md — CLI setup guide
• assortedLLMTasks/tasks/2026-04-08-ai-at-work-productivity-guide.md — Corporate AI guide
• privateContext/deliverables/closeouts/2026-04-08-ai-at-work-guides.md — This closeout