I Stopped Doing Work a Few Weeks Ago

I stopped doing work a few weeks ago.

Not in the way you think. I still show up every day. I’m still responsible for SecOps, DevOps, compliance, AI governance, and IT operations at PKWARE.

But I stopped executing work. I build the systems that execute it , and now I spend my days giving feedback and making decisions.

That shift happened in a few weeks.

What I Built at Work

It started with my SecOps backlog. Dozens of tickets in various states, inconsistent formatting, unclear acceptance criteria. The kind of operational debt that every IT leader knows and nobody has time to fix.

So I built skills , stateless, repeatable AI capabilities , that groom the entire backlog to conform to standards. Every ticket gets consistent structure, clear criteria, and proper categorization. Not a one-time cleanup. An ongoing capability.

Then I went further. I built agents that handle the full planning lifecycle: when a Jira ticket arrives, AI generates plans, architecture documents, decision logs, specs, and research , all automatically during the planning phase. Then it executes against those artifacts in a structured, stateless way. Human approval at every gate.

I’ve now sent dozens of Jiras through this pipeline end-to-end. The work gets done. The documentation exists. The decisions are tracked. And my role shifted from “person who does the work” to “person who reviews the work and makes the calls.”

That’s not a productivity improvement. That’s a different job.

The Day I Got a Wall of Thumbs Up

We were onboarding people to a new system. Normally this means writing instructions, sending them out, and then fielding a stream of individual questions , each person with slightly different circumstances, slightly different issues.

I built a Teams listener and autoresponder hooked into my own account. It sent onboarding instructions to each person, then automatically replied to their individual follow-up questions , addressing their specific situations, not canned responses. Every reply went through me for approval before sending. Nobody knew it was AI. They just knew they got fast, accurate, personalized answers.

I got a wall of thumbs-up emojis that day.

Email at Zero

I did something similar for email. My inbox is now at zero , not because I declared bankruptcy, but because an AI agent watches, files, prioritizes, and predrafts my responses.

It knows who I need to talk to and who I don’t. It surfaces actionable items. It drafts replies for my approval. I review, adjust if needed, and send. The cognitive load of email , that constant background tax on every knowledge worker , is gone.

The Same Patterns at Home

I brought the same approach home because the patterns are universal.

I used 1Password’s secure CLI (op) to fully analyze my family’s password vaults , identifying duplicates, weak passwords, inconsistent naming, orphaned entries. The CLI handles all credential access locally; no passwords ever touch the AI model. The AI sees metadata , titles, URLs, vault assignments, password strength scores , and makes organizational decisions from that. Then I reorganized everything with standardized naming conventions across vaults. The kind of digital housekeeping that everyone needs and nobody does because it’s hours of tedious work.

I built a multi-platform email agent that watches, organizes, drafts, and responds across my Gmail, Outlook, and iCloud accounts. Three platforms, one unified workflow.

I even organized my bookmarks , cross-referencing browser usage data with my password vault to categorize, deduplicate, and structure years of accumulated links.

Each of these would have been a weekend project. Together they took days.

Security Isn’t Optional. It’s Structural.

Here’s the part that matters most to me professionally: every single thing I build goes through automated STRIDE threat modeling and security review. Not as an afterthought. As part of the design itself.

When AI is reading your email, responding to your messages, and accessing your credentials , security can’t be a policy document you wrote last quarter. It has to be structural. Baked into the architecture. Enforced by code, not by trust.

Prompt injection defenses, input validation, output boundaries, human approval gates , these aren’t features I add at the end. They’re constraints I define before the first line of code.

This is what I mean when I talk about intent engineering. You define the outcome, the constraints, and the verification criteria , and then you build systems that can’t violate them, even if they wanted to.

What a Few Weeks Taught Me

The biggest lesson isn’t about AI capabilities. It’s about what happens to your role when you stop being the executor and start being the reviewer.

You make better decisions because you’re not exhausted from doing the work. You catch more issues because you’re reading output with fresh eyes instead of debugging your own code at 2 AM. You move faster because you’re not context-switching between planning and execution.

I will say , I traded work fatigue for decision and feedback fatigue. Turns out “just review everything and make the calls” is its own kind of exhausting. My brain is tired in a completely different way now. Less “I’ve been staring at YAML for six hours” and more “I have approved or rejected 200 things today and I need to go touch grass.”

The organizations that figure this out first will have a structural advantage that compounds. Not because their AI is better , but because their people are operating at a different level.

A few weeks. Same title. Different operating model.

Can’t wait to see what comes next. See you at AgentCamp Milwaukee.