Josh Johnson

Notes from JJ

Week 21November 2, 2026 · 5 min read

But Seriously... What Are You Actually Doing With It?

Pick your lane, question what you're paying for, and build something useful.

A hand arranging a nine-cell grid beside connected task cards on cream drafting paper

The projects at MathCo are taking shape, and I'm enjoying getting into the work. It's also changing what I want to talk about here. Less “this model is incredible.” More: what are we actually doing with all this capability? Last week, I wrote about separating sourcing from execution. This week, I'm asking a more fundamental question: what's worth building?


Pick a Lane. Then Pick Your Next Move.

Colleagues organizing ideas together on a whiteboard

Coming into a new role, there's a question underneath the title: what do I want people to count on me for?

In consulting, that might be strategist, project lead, or the person who gets close to the problem and builds the thing. I think those are useful lenses well outside consulting, too. Not a prediction that every job disappears into three boxes. More a way to decide where you want to contribute as AI becomes part of the team.

  • Do: direct the tools and produce the work.
  • Manage: coordinate people, agents, handoffs, and delivery.
  • Lead: use domain knowledge to decide which problems matter and what good looks like.

You can occupy more than one lane. The interesting question is your capability in the one you want to develop. Here's the nine-box I've been sketching:

YOUR AI CONTRIBUTION
Find your lane. Make your move.
LEVELDoManageLead
HighTeachExplain your checksTeachShare the processTeachMake judgment reusable
MediumApplyShip a small taskApplyRun a workflowApplyTest a decision
LowLearnFollow a small buildLearnMap a handoffLearnStudy the problem
Choose a role across the top. Locate your capability down the side.
Each box gives you a next move—not a permanent label.

Low means learn. Medium means apply. High means teach. Not permanent identities; next moves.

The scale is relative to the work, not the loudest person on the internet. A strong engineer can be learning how to manage agents. A domain expert can lead an excellent AI project without being its best coder. Rate yourself against an actual task: can you explain the result, catch a bad one, and repeat the good one?

Then look outside your own setup. Find someone in your field doing this well, study their process, and compare outcomes—not subscription lists. Pick one cell and one thing you'll practice this week. That's more useful than declaring yourself “good at AI.”


AI Can Be Good for Users and Rough on the Business Model.

Financial charts and market data displayed on a monitor

I don't write much about stocks here, but I watch them closely. Last week's selloff made the gap between “AI is useful” and “AI is good for this company's economics” hard to ignore.

On October 29, Reuters reported a sharp software selloff, including ServiceNow, after results disappointed investors already worried about AI disruption. Yet ServiceNow's own release reported subscription revenue up 21% year over year. Falling shares don't automatically mean a disappearing business.

Full disclosure: I participated in the ServiceNow selloff with my own money. Bought in, promptly took an L, and I'm still in the red. I believe in the company, so apparently I'm a long-term investor now. Otherwise known as a bag holder.

The big-platform story was different. Reuters reported that Alphabet fell 12.6% on October 29, then another 1.4% on October 30; Amazon finished October 30 down 1.4%. Heavy AI spending and cash-flow concerns were central to that coverage. These weren't all double-digit drops, or all evidence of the same kind of cannibalization.

My broader question is about what happens when the task gets cheaper than the tool built around it.

If the AI subscription I already use can help me make a simple site or a graphic, I might skip another specialist subscription. That doesn't make it a complete replacement for Figma, Canva, Wix, or Squarespace. Hosting, collaboration, support, and reliability still count. But for my particular task, the extra bill now needs a better explanation.

That's the tension I'm watching: AI can expand what customers can do while putting pressure on what vendors can charge. Useful for thinking about your career and what you build—not a stock tip or proof that every software company is finished.


Fewer Apps. More Inventions.

A hand holding an analog wristwatch in warm natural light

I'm done asking AI to impress me. I'm starting to ask for what I actually want.

I've spent plenty of time testing the limits: how beautifully it can design, how much it can code, how elaborate I can make the next little thing. Cool. I believe it works. I don't need to code my mom an AI-generated greeting card to prove it again.

Sometimes what I want is much simpler:

  • “Check Rolex Explorer prices on Reddit, eBay, and Facebook Marketplace. Tell me when something worth a look shows up.”
  • “Email my mom. Tell her I love her.”

Neither takes rocket science to say. Neither needs a new app. These are things I'll keep wanting to do, and somehow we still forget to ask for them while we're busy finding out whether AI can build an entire company before lunch.

Muse has made me more willing to just say the thing. Give it the details, tell it what a useful result looks like, and let the tools I already have do some work for me. A price check that comes back when it matters. A little note that actually gets sent.

That's where I'm leaning in: fewer demonstrations of what's possible, more small things I actually wanted done. I'm saying that shii.

What do you keep meaning to do but never think to ask your agent for? Reply and tell me. It doesn't have to sound impressive.


The Takeaway

  • Choose your contribution. Do, manage, lead—or a deliberate mix.
  • Match the next move to your capability. Learn, apply, or teach, using real work as the test.
  • Build around the outcome. A useful workflow matters more than another software subscription.

— Josh

Quick Links