I run several businesses at the same time. Atypical Global. The B2B beauty platform. Board seats. Advisory work. A 3-year-old son named a 3-year-old toddler. A wife managing her own credit repair. Travel. Health stuff. The only reason this is possible in 2026 is that I have rebuilt my personal operating system around AI agents — specifically, around Anthropic's Claude Code and its skill architecture.
I've been around the block on every AI wave since the BBS era. This is the one that actually closes my strategy-execution gap. For 30 years I was the guy who could see the whole chessboard but couldn't physically move fast enough to execute on all of it. AI finally gave me arms long enough to match my eyes.
This is not an "AI will change everything" piece. This is a concrete tour of which skills I actually use, how often, and what they replace. If you're a founder or operator wondering whether this stuff is real yet, I'll give you the operator's answer. Yes — for a specific subset of work. And no, for a larger subset most people are lying to themselves about. The Information has been covering the real adoption curve of AI tooling inside operating companies, and the gap between the hype and the deployment is still wider than the LinkedIn posts suggest.
The skills I use every single day
gsd agents (Get Stuff Done)
The gsd family is the spine of how I plan and execute projects now. Instead of managing a Linear board and a Notion doc and a spreadsheet that all go stale within 48 hours, I drive my work through gsd-new-milestone, gsd-plan-phase, gsd-execute-phase, and gsd-session-report. Every project becomes a structured sequence of phases. Each phase gets planned, executed, and verified. At the end of every session, I get an auto-generated report that tells me exactly where things stand.
The value is not that AI is writing code or making decisions. The value is that my work now has a memory. I can walk away from a project for three weeks — a legal deposition, a trip to Korea for my back, a week of parenting — come back, run gsd-resume-work, and within 90 seconds I'm back where I was. All the context. All the decisions. All the open questions. That compounding memory is the single biggest productivity shift I've had in 20 years.
contacts-debrief
This one I built for myself because I kept losing the context from meetings. Now after every meaningful conversation — a dinner with BJ at Cosmax, a board call, founder office hours — I dictate a 60-second voice note. The contacts-debrief skill transcribes it, extracts the people, the topics, the commitments, the follow-ups, and writes everything into my contacts database (36,000+ records) and my Notion mirror simultaneously.
The result. I now remember things about people that I have no right to remember. Their kid's name. The book they mentioned. The deal they were worried about. The wine they ordered. It's not because my memory got better. It's because the context got captured before it could decay.
AI didn't make me smarter. It made me less forgetful. For an operator juggling a dozen contexts, that's a bigger upgrade than any IQ boost.
design-auditor + frontend-design
I am not a designer. I used to hire designers for every marketing page, every investor deck, every landing page. Now I use design-auditor to score any UI I'm about to ship across six pillars — visual hierarchy, typography, spacing, color, consistency, accessibility — and frontend-design to actually produce pixel-perfect HTML/Tailwind/React from a rough spec.
The quality is not world-class design studio output. It's 80th percentile execution at 5% of the cost and timeline. For an operator, that is almost always the right trade. I'd rather ship 80th percentile this week than 95th percentile in three months that never arrive.
autoresearch-marketing
Before I write any marketing copy for any of my companies, I run autoresearch-marketing first. It looks at what's working in the niche, what's converting, what competitors are saying, where the gaps are. Then I write against that baseline, not against my gut.
My gut has been wrong enough times in 31 years that I've learned to stop trusting it on anything that faces a market. Data beats intuition. AI makes the data cheap enough to use every single time.
webapp-testing
Every time I ship or change anything on one of my websites, I run webapp-testing against it. It catches broken links, layout breakage, accessibility failures, performance regressions. It used to take me a day of manual QA to cover what the skill now covers in 20 minutes. The quality of what I ship has gone up specifically because the friction of testing has gone down.
The skills I stopped using
In the interest of honesty: a lot of AI tooling is not worth the time. Here's what I tried, abandoned, and why.
- AI meeting note-takers that join calls automatically. They generate summaries nobody reads. The 60-second debrief I dictate after a call captures 10x more actionable detail than a 20-minute transcript summary ever did. The work is in the compression, not the capture.
- AI email drafting in isolation. The drafts were always generic and had to be rewritten anyway. The only place AI adds value in email is when it has full context from my contacts database, calendar, and prior conversations — which is a much harder integration problem than "write me an email."
- AI strategy advisors. I tried prompting models to act as advisors for strategic decisions. Fine for structuring a thought. Useless for anything that requires taste, judgment, or reading a room. The 31 years of operator experience I carry — including every scar from 12 acquisitions — is specifically the thing AI cannot replicate yet.
The mental model I use to decide what to automate
Here's my rule. I automate work that has clear inputs, clear outputs, and a verifiable result. I do not automate work that requires judgment, taste, or a relationship.
Debriefing a meeting? Automate it. Writing a marketing page after research? Automate the research and the first draft. Running QA on a website? Automate it. Making the call on whether to fire a VP of Sales? Not a skill. That's my job, and my job is where the real compounding happens.
The operators who win the next five years are not the ones using AI for everything. They're the ones who are ruthlessly clear about which 40% of their job can be automated and which 60% can't — and who spend their freed-up time on the 60% instead of on the part the AI already handles. Technology is a force multiplier. It doesn't replace the operator. It amplifies the good ones and exposes the bad ones faster.
Here's my challenge. Pick one thing on your calendar this week that you dread doing. A debrief. A QA pass. A research task. A first-draft of anything. Try to automate it before you do it manually. If it works, you just bought back a piece of your life. If it doesn't, you've learned something specific about where AI actually breaks for your workflow — which is worth more than another podcast about AI strategy.
For the tooling side of how I run this stack, see the thin-client founder setup.