"It's just not evenly distributed." — William Gibson
What's possible right now, if you're willing to commit.
01 / The New Competitive Advantage
The constraint is no longer knowledge. It's judgment about what matters.
Speed of iteration beats depth of planning. The gap between an idea and a working prototype is basically zero.
How many tokens per minute can you control with quality output? That's the new competitive advantage.
The old game was about being smart. The new game is about being willing. The tools are available to everyone. The gap between people who use them well and people who don't is already visible. It's about to get much wider.
02 / The Meeting vs The Build
"It now costs more to have the meeting about the feature than to just build the entire thing."
This is a complete shift in how work happens. Instead of scoping, debating, writing requirements docs, scheduling stakeholder reviews; build it. Ship it. Iterate. The cost of a failed experiment is measured in minutes now, not sprints. The companies that internalize this first will leave the rest behind.
03 / Workspace Engineering
A ladder. Most professionals sit on the first rung and don't know the rest exist.
You open ChatGPT without logging in. It starts from scratch every time. No memory, no context. Most of the room is here.
The AI knows your name, your role, maybe your preferences. Saves 30 seconds per conversation.
You have a prompt template you paste in every time. "You are an expert in X. I need to do Y. Here's the context."
Persistent workspace with shared context. Claude Projects, ChatGPT Projects, Gemini Gems. You stop repeating yourself.
Shareable AI assistants configured for specific workflows. Your whole team uses the same tool.
The AI has your entire codebase or project folder as its workspace. It can read files, edit them, run commands. This is where it starts feeling like a collaborator.
You give the AI its own computer. It wakes up every hour. It checks your email, reviews your calendar, reads transcripts, takes actions. "Make my life 1% better every day."
Most professionals are at Level 0. A few are at Level 2 or 3. The top 1% are experimenting with Level 5 and 6. The distance between them compounds daily.
04 / Self-Annealing Systems
Annealing is an old blacksmithing term. You heat metal, beat it, cool it. Each cycle makes it harder. Self-annealing AI systems work the same way.
"Over time, these become less probabilistic and more deterministic."
This is the compounding advantage nobody talks about. You don't just save time once. You save more time every month. The workflow keeps improving as long as you keep using it.
05 / Tokens Per Minute
Ask yourself: how many tokens per minute can you control without losing quality?
Can't use the tools
Output is zero. The room most companies are still hiring from.
One AI at a time
Can get meaningful value from a single assistant. Already ahead of everyone else.
Fleet operator
Running 4–5 AIs in parallel. Thousands of tokens per minute of output. Orchestrating fleets.
This isn't about prompt engineering. It's about becoming someone who can direct AI work at scale, in parallel, across contexts, without losing judgment. The people who can do this in 2026 will look like wizards in 2028.
06 / Auto-Research
Andrej Karpathy, former Tesla AI director and OpenAI co-founder, has been talking about auto-research: AI systems that iteratively improve themselves by designing their own experiments, running them, and updating their methods based on results.
This sounds futuristic. It's not. It's practical. If you have a workflow that runs once an hour, and each run makes the workflow 0.1% better, after a month, the workflow is 72% better than when you started. After three months, it's 3–4x more capable. This math is real.
07 / The Swarm
Start with one. A small computer in the cloud or on your desk. It wakes up every hour and does one useful thing. Checks emails. Reviews transcripts. Takes a small action.
Now you're not executing work. You're orchestrating a team that never sleeps. This isn't science fiction. This is Tuesday for the people already building it. The question isn't whether this is coming. The question is whether you'll be the person doing it, or the person it happens to.
"The constraint really is: run it more, or have more of these things going."
08 / The Gap
Because most people haven't started.
There's a whole room of people interested in AI who haven't touched it. That's the reality. The tools exist. The capability is here. The barrier is human; curiosity, commitment, time.
I spend every waking minute doing this. That used to be an embarrassing thing to say. Now I'm all in on it, and I still can't keep up. The field is moving that fast.
09 / Practical Takeaways
Find something you do at least once a week. Build a repeatable AI version of it. Commit to improving it every time you run it.
Get to at least Level 3 (Projects). Stop starting from scratch. Build context that persists across conversations.
The goal isn't to know everything. The goal is to know what matters and commit to acting on it. Ship. Iterate. Repeat.