johnmaconline chain logo johnmaconline / The250
Daily one-pagers about AI, leadership, technology, and being human.

johnmaconline

I'm writing to think, learn, and remember in public. I'll be here everyday.

August 14, 2026 1 minute read

An Optimistic AI Future from Zuck

AIHumanLeadership

Throw a rock and you’ll hit a doomsday AI prediction.

Dario from Anthropic leads the fearmongering by basically telling everyone who’s just coming out of school or who works in a white-collar industry, “Hey, Claude’s gonna take your job. Sorry.”

Sam from OpenAI is less overt, and way more savvy (cagey?), but I think inside the OpenAI fortress of $100M engineers, he agrees.

You know me. I’m not a fearmonger. Also, I’m not closing my eyes to the reality that AI is and will continue to change how the workforce looks. AI has totally transformed how I do my work and how those on my team do their work.

But I’m an AI optimist.

And now so is Zuck.

Here’s the TL;DR:

Our AI future can be one where superintelligence is widely distributed and turns billions of people into more capable inventors, learners, and entrepreneurs. AI can make individuals dramatically more capable, rather than simply replace them.

To help usher that in, Meta will be open-sourcing their models and tools. More eyes. More ideas. More people. Less friction.

Kudos to Zuck.

I’m on board. I can’t wait for the future.

August 13, 2026 1 minute read

It’s Not You

HumanLeadershipThe250

If you hear that, assume it’s you.

Not because the person who said it is lying. They might be, but also, they might be completely sincere.

However, if you just assume it is you, then you can purposefully and definitively move on.

Go where you’re valued.

August 12, 2026 2 minute read

Self-Awareness

AIHumanLeadership

Who are you in the world of AI?

If AI can do what you do, you have two choices:

  1. Do it better.
  2. Do something else.

Right now we’re in a phase where people still want to do one because that’s what they know or have always done. And also, think they can indeed do it better.

It’s possible. Studies still show that AI can’t do it better than the very best humans (but not necessarily faster or cheaper) for now. This won’t last long, however. If you want it this way, you must lean into the art of what you do. Find the art. Master the art. Better is in the art.

Two is probably a better path for most of us.

Pick a trade or human-facing work, or there is plenty of work in genres that can’t or won’t lean into AI yet due to security or other concerns.

Another way is to recognize that AI will need a boss. It will always need a boss at some level. AI rewards people with ideas and the gumption to try stuff.

Figure out how to get AI working for you or you’ll be working for AI.

August 11, 2026 1 minute read

UFOs

HumanThe250

The government just released more UFO files.

UFOs are fun, and I’ve been thinking about them since I was a kid.

If you don’t want to dive into all of the info, here’s a summary:

Yes, people really are seeing things.
No, we cannot identify all of them.
And no, we still do not have publicly verified alien spacecraft.

That’s all. This is just a UFO appreciation post.

August 10, 2026 1 minute read

Doing It or Helping You Do It

AIHumanTechnology

There’s a difference between when you hire someone to do the work for you and when you hire them to help you do the work.

“I’m looking for someone to build my detached garage.”

That’s different than, “I’m looking for someone who will dig and pour the foundation for the detached garage that I’m building.”

Your reasons might be cost, time, expertise, certifications, trust, design, or simply whether you want to do some or all of the project yourself. Regardless of your reasons, you think about the projects differently.

Currently, AI is better at helping you do the work.

But it’s getting closer to doing the work for you.

When you’re hiring AI, you must apply the same logic.

Are you hiring AI to do the work or help you do the work?

August 9, 2026 2 minute read

The New johnmaconline.com

AILeadershipTechnology

If you’re reading this on johnmaconliine.com, you might notice that it looks very different from what it did a few days ago.

The redo is 100% Codex-developed. Not the content, of course, but the website itself.

I wrote zero lines of code. However, I went round and round with first GPT-5.5 and then Terra and Sol to make it what you see. My workflow was:

  1. Develop a documented plan — I did that by describing to Codex what my goals were in free-form text.
  2. As per the plan, developed a requirements document — I did that by listing out all of my known requirements and having Codex walk me through a Q&A to develop a fuller set.
  3. Fed the plan and requirements doc into Codex and asked it to develop a local prototype and test it. It built everything from the ground up and tested it (as best it knew how).
  4. Iterated on that local prototype with Codex for a couple weeks by giving it feedback such as “I don’t like the way the sidebar is ordered. Change it to…” and “Look at this screenshot (pasted screenshot). The title is in the wrong spot and the wrong size. Fix it.” and “Let’s try a different color scheme.”
  5. Once it looked OK, I had Codex publish it to v2.johnmaconline.com so I could test both the publishing automation to my host (I currently use Bluehost) and all the functionality, such as email subscription, search, stats, etc.
  6. Iterated for a couple more weeks with Codex to get all of the functionality correct.
  7. The last step before publishing live was to have Codex develop automation that could “undo” the publishing of the new site. I wanted a failsafe so that if I published and it fell on its face, I could immediately restore the “v1” version.
  8. When I was satisfied it all worked, I told Codex to go ahead and publish.

That’s the new software development paradigm with AI.

As developers, we’re no longer writing code. We’re guiding and directing the agent to do it. I’m in charge of requirements, how it’s architected and implemented, and ensuring it meets the requirements. It’s in charge of syntax.

This isn’t “vibecoding.” This is pro software development.

I’m a pro, and my AI agent is a pro.

Welcome to the new johnmaconline.

August 8, 2026 2 minute read

The Distraction of Effort

HumanLeadershipThe250

We often think about work to be done from the standpoint of how hard it will be, and we usually equate how hard it will be to how many hours it will take, either to do the work or to become the expert.

Our industrial complex for the last couple hundred years has taught the average worker to think about their worth in terms of how much effort and time.

I get paid $X per hour, therefore, this is what it’s — and I’m — worth.

But in reality, the purchaser of the work doesn’t care about how hard it will be or how many hours it will take you. We care about the outcome.

If you’re paying someone to dig a ditch, you don’t care about how many rocks they’ll have to deal with. You care that when they’re done, you have a ditch.

If you’re paying someone to teach your children Algebra, you don’t care about the behavior they’ll have to manage. You care that when they’re done, your children understand Algebra.

If you’re paying an accountant to do your taxes, you don’t care about the nuances of the tax system they’ll need to know. You care that when they’re done, your taxes are filed correctly.

The effort required is a distraction when the outcome matters.

Doctor’s don’t deserve to get paid more because of the 12 years of schooling and residency they’ve put in. They deserve to get paid more when they’ve saved or enhanced a life.

When you’re deciding how much you’re worth, don’t get distracted by the effort required.

August 7, 2026 1 minute read

Eyes on the Prize

HumanLeadershipThe250

When you’re keeping the spinning plates from falling, it’s hard to keep your eyes on the prize.

But wait, what is the prize?

Because your prize might not be your colleague’s prize, or the media’s prize, or your neighbor’s prize. Many will try to distract you so you care about their prize.

Always know your prize so you can keep your eyes on it.

August 6, 2026 2 minute read

Somebody Who Cares

HumanLeadershipThe250

We found ourselves in a situation.

Rented an apartment in Philly. Showed up to move in. It was the wrong apartment.

So let me clarify that. We signed the lease on the apartment that they gave us, but that was NOT the apartment we had been shown nor the one we wanted.

It was bait and switch, but (probably) an honest mistake, not malicious.

Why did this happen?

Because the rental agent dropped the ball. Didn’t do her job. Provided the paperwork for the wrong apartment. Didn’t check. Didn’t ensure. We trusted, and had no choice but to trust given the situation. Because how would we know?

So here we are. Full truck parked at the loading dock on Ludlow St with the flashers on.

First call was to the rental agent.

“No, that’s the apartment you wanted.” Then, “But you said…” Then, “I have in my notes…”

All wrong. All BS. All verifiably incorrect via the text message string. All her trying to push the blame onto us.

Then finally, “Well, that’s the lease you signed.” Unfortunately, the only truth of the conversation.

All we needed — all we wanted — was somebody who cared and would try to help. We weren’t moving into this apartment. So help us figure this out.

She had an opportunity here. She had an opportunity to help. She had an opportunity to advocate. She had an opportunity for true customer service.

She had an opportunity to care.

But she didn’t.

If you’re in the business of customer service, it’s gonna pay to be the person who cares about your customers.

August 5, 2026 1 minute read

Asking Questions

AIHumanTechnology

“Do AI models ask questions?”

ChatGPT:

“Not usually in the way a person asks questions.

During standard training, the model is mostly given examples and learns to predict the next token. It does not pause, notice confusion, and ask the trainer for clarification. The training system selects the data, calculates the error, and adjusts the model automatically.

Questions can appear in the training data, so the model learns how and when people ask them. Later stages may also specifically reward good clarification questions through human feedback or reinforcement learning.

Some newer training approaches let models generate questions for themselves, create practice problems, critique their own answers, or request information from tools. But that is a designed training loop—not spontaneous curiosity.”

Claude and ChatGPT don’t ask questions. They aren’t curious. They don’t care.

You ask questions. You are curious. You care.

Keep asking questions.

Pin It on Pinterest