Hi there,

Happy new year! In 2026, I'm writing more about becoming an AI-fluent knowledge worker. You can expect our editions to be 70% practical, 30% meta.

Today, in 10 minutes or less, you’ll learn:

• 👻 Why "PhD test passing ghosts with adolescent-like cognitive abilities" perfectly captures today's AI tools

• ⚡ How I'm leveraging LLMs to design UI, write SQL, and fix bugs as a generalist

•  🔥 What finally made work feel magical again after feeling stuck and uninspired

🚀 Live Workshop: Claude Code for Non-Coders 101 (RSVP)

Learn how to build reusable AI workflows with Claude Code slash commands - no coding required.

I'll demo my actual content and event automation, then we'll build your first command together.

What We'll Cover

• Live demo: My slash commands for newsletters, workshops, and events

• Hands-on: Build your first slash command for a task you do weekly

• Q&A: What works, what doesn't, and how to iterate

Who It's For

• Knowledge workers who want to automate repetitive tasks

• Anyone curious about AI agents but unsure where to start

To get your brand in front of 7k+ professionals, founders, and consultants, fill this out and let's chat.

👨‍💻 How I'm using agentic tools in January 2026

In the past month, multiple business owners told me AI keeps them up at night.

To start, I'm blown away by last year’s LLM advancements. I have written about using tools like Claude Code and Relay to do things in minutes that would’ve taken me days before - like synthesizing user research. I’ll share more examples later.

On the other hand, countless media outlets predict the death of knowledge workers (which they also did 10 years ago. Hmm). Many of my peers have admitted feeling afraid of losing their jobs or being replaced by technology. And who can blame them?

I'm a knowledge worker too. Everything feels like it’s changing.

I don't know what the future will bring, but I try to focus on what's in my power to shape my own future.

One thing I’m doing is obsessively learning and practicing new technology. Leading with curiosity. I treat it a bit like a game.

My current learning agenda: trying to get more out of agentic tools like Claude Code and Cursor.

LLMs are powerful, but also finicky and unreliable. Andrej Karpathy, former OpenAI co-founder, had this sticky line where he describes LLMs like ghosts that can pass PhD tests but still possess adolescent-like cognitive abilities.

My hope is that by pushing the boundaries of the edge tools habitually, I can figure out how to best apply them to interesting problems. Like wiping the fog off the window just enough to catch a clearer glimpse of the future.

Here are four ways I'm leveraging agentic tools to accelerate my work:

1/ Filling in gaps as a generalist

The 2010's were defined by technical specialists.

Information asymmetry and domain knowledge scarcity meant specialists led organizations, commanded high salaries, and found themselves in high demand.

In the late 2020's, we can now use LLMs to quickly fill our knowledge gaps faster than ever before. We’re one step closer to being like Neo instantly uploading kung fu skills in the Matrix.

Let’s take vibe coding. Yes, there’s quality issues, but I’m increasingly seeing non-technical friends building productive apps for their work. The inverse is also happening - I see more engineers filling the gaps of product/business people in small teams.

15 years ago, Chris Dixon said, "The next big thing will start out looking like a toy." (which he borrowed from Clayton Christensen).

I think we're seeing it here too.

How I'm implementing this:

In my pod, I'm filling in the gaps wherever possible to relieve bottlenecks:

• When our designer was still deliberating her offer, I stepped in to design a few UI/UX flows to unblock development.

• When our engineers were heads down on crucial tech milestones, I fixed bugs myself and submitted PRs for review.

• When our firm needed to automate reports, I wrote the SQL queries instead of waiting for us to hire a data analyst.

Charles Munger, famed investor, lamented often about how our top educational institutions failed to train our grads in interdisciplinary thinking.

In this era, working across disciplines is not a luxury, but a necessity.

2/ Delegating repetitive work

Learning how to automate the energy-sucking parts of your job is becoming table stakes.

Take doc-writing for a traditional product manager.

Writing PRDs, planning docs, tickets, specs, and Slack messages used to consume dozens of my hours every week. And I’d still be drowning in more work.

Today, writing docs and Slack messages is 3x-10x faster. Notion AI helped, but setting up Claude Code slash commands + Notion MCP led to a much greater lift for me.

As a result, I freed up more of my time to interface with portfolio firms, problem-solve, and prototype/validate my ideas. (the fun stuff!)

How I'm implementing this:

Prototyping used to be something I only got around to once in a blue moon.

I remember at Dropbox a decade ago, I depended on my designer to create UI mockups for customer feedback. But they were so busy, we only did it once a month at our best.

Contrast that to now: I can whip up a prototype, deploy to Vercel or a staging server, and get client feedback in 1-2 days.

Here's a subset of the tasks I'm delegating to Claude Code, Gemini, etc.:

• Writing SQL queries

• Synthesizing research

• Making presentation graphics and slides

• Debugging code

• Drafting PRDs and specs

• Creating prototypes

The list keeps growing every week.

3/ Investing upfront time in context scaffolding

I’ve realized the quality of my AI output is directly proportional to the quality of my input.

Garbage in, garbage out.

However, if I input structured context, the output quality is wildly better.

I'm now spending more upfront time collecting, organizing, and structuring context than before.

This feels a bit counterintuitive - shouldn't AI reduce my task checklist? But I think of it like onboarding a new employee. You need great onboarding for them to ramp up faster.

How I'm implementing this:

I’m obsessive about creating a "data layer" that feeds my LLM tools:

• Notion as a knowledge base - We use Notion at work, and I use it for Portfolio Path too. Every meeting note and project doc lives here. I then use Notion AI as a chief of staff to surface what I need. When I need Claude to help with a coding/writing project, I can use Notion MCP to feed Claude the full context in seconds.

• Structured templates - I've created templates for common documents (PRDs, specs, retrospectives). The structure makes it easier for AI to parse and generate consistent outputs.

• Voice notes → transcripts → organized notes - I capture ideas via voice, transcribe them, then organize into my knowledge base. This raw material becomes fuel for Claude to help me create content briefs and outlines.

This has shifted my mental model:

I'm not just picking up new AI tools. I’m building personal data scaffolding that gives these tools significantly more leverage.

It's like the difference between asking a new hire to do something before vs. after they’ve onboarded properly. Same skills, but vastly better results.

4/ Practicing meta-skills differently

Data fluency, management, communication, and learning to learn.

These skills matter more than ever. But they don't look like how I viewed them traditionally.

How I'm implementing this:

Data fluency - This used to be about basic data analysis, Excel, and reporting. Now it's about setting up context scaffolding to feed into my LLMs and agentic tools. I spend more time organizing my knowledge base (Notion, markdown files, project READMEs) than creating excel analyses.

Management - This used to be about hiring and building a large team. Now it's veering towards managing the AI/agentic tools I have to scale my team's output.

Communication - This used to be about communicating up, down, and sideways in complex cross-functional situations. Those situations still exist, BUT I also have to master LLM prompting and context engineering to get the results I want.

Learning to learn - The most meta skill. This used to be about learning one skill intensely in a continuous stretch of time. Now I'm learning a skill enough to clear bottlenecks before switching to what's demanded by the next bottleneck.

Admittedly, this feels a bit like a dance right now. It started off feeling random, but gradually picked up on the coherent rhythm and pattern behind the movements.

The most important thing I’ve discovered

I know I've talked a lot about output and productivity.

But the most important part about all of this isn't what I'm doing, achieving, or even learning.

It's that I'm having more fun with work than recent years.

As a kid, I was obsessed with using new web technologies for the first time. It felt magical. Like seeing into the future.

25 years later, I feel like I've rekindled some of that spark I had from my first web impressions.

Candidly, before this, I was feeling hesitant about working in the tech industry again. It started to feel a bit repetitive and uninspiring.

But for now, I'm having a great time. I hope it'll last for a while.

In Summary

Here are four ways I'm using agentic tools in January 2026:

1. Becoming a generalist - Filling knowledge gaps with AI to relieve bottlenecks (design, code, data)

2. Delegating to automation - 3x-10x faster docs, prototypes in days not weeks, growing task list for AI

3. Investing in context scaffolding - Building a personal database (Notion, markdown, templates) that makes AI tools higher leverage

4. Refining meta-skills - Data fluency, management, communication, and learning to learn - but done differently for the AI era

These edge tools are still finicky. Ghosts that can pass PhD quizzes, but possess elementary student like cognitive abilities.

But by using them daily, I can wipe the mist off the window just enough to catch a clearer glimpse of the future.

P.S. Thanks for reading! Are you using agentic tools in your work? What's working, what's not? Reply and let me know.

💎 Last Week’s Gems

• Foundations: My 1999 (and part of 2000) (Michael Burry)

• Is it a Bubble? (Howard Marks)

• AI tools are overdelivering: results from our large-scale AI productivity survey (Lenny Ratchisky & Noam Segal)

• 6 Surprises From 2025 (Ben Carlson)

• Australia Just Kicked Kids Off Social Media. (Is the U.S. Next?) (Cal Newport)