Via Every

Finally, the other day I just said to Claude Code, ‘Here’s my Gmail credentials. I want you to build an email triager. Here’s how I want it to work.‘ And it basically spit out a pretty web-based version of Superhuman that met my exact workflow, integrated with Google. It’s not perfect, but, in like a week, it’s gotten to the point where I can use it every single day. And that is astounding. Like anyone who’s technical knows how astounding that is and how frustrating it is to build an email client.

— How Andrew Wilkinson Uses Opus 4.5 in His Work and Life at 8:25

---

Most organizations treat AI like a task optimizer—using it to answer individual questions or automate single steps. But entrepreneur Andrew Wilkinson demonstrates a fundamentally different approach: he architects systems where multiple AI capabilities work together to solve entire problems.

This architectural thinking comes through in two areas: email management and personal development.

Email management example at 15:27.

Instead of asking AI to perform tactical interventions like “draft a possible reply to this email” (a task), he built a three-layer system:

• First layer: routing: Automated routing that immediately categorizes incoming messages and forwards most to the right people, cutting his volume in half.
• Second layer: triage and ranking: An intelligent triage interface that ranks remaining messages by priority, urgency, and importance, matching his decision-making priorities.
• Third layer: response options: Initially, he had AI-generated response options that he can approve with a single keystroke. For more complex emails, he could go into Q&A mode to provide answers one at a time.

The result? He transformed a problem that required one, sometimes two, full-time assistants into a manageable 15-minute daily review. Each layer reinforces the others: routing reduces what needs to be triaged, triage informs which emails get responses, and he can approve proposed responses with a single keystroke.

Personal development example at 19:01 and at 22:25.

Wilkinson applies the same systems thinking to personal development. Instead of asking AI for occasional coaching advice (a task), he built a multi-layer system:

• First layer: assessment: A comprehensive personality assessment tool that combines multiple clinically validated tests into a 40-minute quiz, generating a detailed psychological profile covering personality traits, attachment styles, and compatibility patterns.
• Second layer: assessment results become AI context: The psychological profile is used to generate system instructions for a custom AI coach that understands his specific patterns, triggers, and growth areas.
• Third layer: pattern analysis: The AI coach analyzes his meeting transcripts over time to identify behavioral patterns and relationship dynamics as they unfold.
• Fourth layer: applied guidance: These insights generate relationship advice he and his partner can reference during actual conflicts, plus shareable cards that help colleagues understand how to work with him.

The result? The system predicted the fights he and his girlfriend have in their relationship with “scary” accuracy. When they use the custom AI coach during arguments, it provides contextually relevant guidance based on their combined profiles. Each layer feeds the next: assessments inform AI context, AI context enables pattern recognition, patterns generate actionable guidance.

Beginners to AI focus on which tools to use or how to write the perfect prompt. Power users think architecturally and ask: “What workflow can I redesign end-to-end?” rather than “What single task can I automate?” They look for leverage points where multiple AI capabilities, when used together, can compound, creating solutions that are greater than the sum of their parts.

This is the difference between using AI tactically and using it strategically. Tasks get automated. Systems get transformed.