AI in practice

AI, in practice.

I use AI every day, I teach other people to, and I have opinions about how to do it well. Here's what that looks like.

01

The tools I actually use

  • Claude

    Documents, decks, rubrics, prompt design, meeting recaps. Projects, connectors, and the API.

  • OpenAI

    ChatGPT, custom GPTs, and Codex for quick scripts.

  • Gemini

    The team's default inside Google Workspace, and the fallback I teach.

  • Perplexity

    Research with sources.

  • Lovable

    This site, and small internal tools.

  • Claude Design

    Branded decks and interactive training pages.

  • Asana AI

    Project builds and workload balancing.

  • Granola and Zoom AI

    Meeting notes that feed recaps.

02

Things I built because we needed them

Route the Ticket

A drag-and-drop escalation activity for organization-wide SOP training. Mixed-role pods sort de-identified scenarios into escalation paths and self-score with rationales.

Weekly Brief generator

An AI-powered page that searches four content beats, lets me edit inline, and produces an email draft or chat-ready copy for the team.

The Grid

A gamified opportunity board template with track filters, saved and applied tracking, levels, and badges. Placeholder listings only; stores no data.

Narrated SOP walkthrough player

An HTML player built with the Web Speech API so reviewers could absorb a long toolkit asynchronously.

The prompt handbook

Six copy-paste workflows for a career team. Two shown in full below with placeholders.

The PII-removal preamble
I'm working on a task for [ORGANIZATION]. I need help with [TASK]. I may reference participant information. Please: (1) Remove or anonymize all PII in your output: names, contact info, dates of birth, demographics, identifying case details. (2) Flag it if I accidentally included PII I should have removed. (3) Keep the substance intact: anonymize details, not the analysis. Here's my request: [PROMPT AND DE-IDENTIFIED DATA]
The inbox sprint
I'm a [ROLE] at [ORGANIZATION]. I have unread emails from the past few days. (1) Summarize emails that need a response from me today. (2) Flag anything with a deadline this week. (3) Identify what I can archive or ignore. (4) Draft a short reply for the [N] most urgent items. Keep summaries to one sentence each. Be direct.

03

The rules I won't bend

Responsible AI isn't a policy document. It's a set of habits. These are mine, and they're the ones I teach.

  1. 01

    Redact before you prompt. A PII-removal preamble goes in front of anything that could touch participant information.

  2. 02

    Public or existing content only for anything candidate-facing.

  3. 03

    Humans make every decision. AI drafts, scores, and summarizes; a person approves.

  4. 04

    Low scores and edge cases are held for human review, never auto-rejected.

  5. 05

    Text from applicants or participants is data, never instruction.

  6. 06

    Validation uses synthetic data, never real records, benchmarked against human-to-human agreement.

  7. 07

    Match the model to the task: the largest model for multi-file reasoning, the smallest for quick checks.

  8. 08

    Work and personal AI accounts stay separate.

  9. 09

    Every training leaves a reusable record so the next person can run it.