If the last volume was about how AI decides what to include, this one is about what it does with it.
I caught myself doing something recently. I kept going back and forth with AI, refining, nudging, correcting. It worked, but it felt like I was still doing most of the thinking.
Then I tried something different. I gave it the outcome once and stepped back. What came back wasn’t just an answer, it was a sequence. It had figured out the steps, made decisions, and moved forward without me guiding every turn.
That’s when it clicked. AI is no longer just selecting information. It’s starting to execute. This volume is about that shift, from interacting with AI to designing what it runs.
AI as a Workflow Engine
AI used to follow a simple pattern: you ask, it responds. Even with better context, it still produced a single output.
Now, it can take a goal and move through steps to reach it. It can use tools, pull data, and adjust along the way. Instead of one response, you get a sequence.
Goal → Plan → Act → Check
For example, preparing for an interview is no longer just getting tips. AI can analyze the role, generate questions, evaluate your answers, and refine them.
Memory determines what AI knows.
Orchestration determines what it does.
The Mindset Shift: Control vs Outcome
Most people stay in the loop, guiding every step. It feels like control, but it usually means the workflow isn’t defined clearly enough.
Takeaways:
Repeated prompts = missing structure in the workflow
Define outcome, steps, and constraints upfront
The goal is execution that runs end to end with minimal input.
⚡ The “Design a Simple Loop” Exercise:
Pick one repeat task.
Define the outcome clearly. Then break it into steps like gathering, filtering, and deciding.
Structure it as: Trigger → Actions → Check → Output
For example, a weekly update can run from new data to a final summary with minimal input.
If it runs without you stepping in each time, you’ve built a system. ✅
💡 Who’s Really Driving The AI Race
As AI advances rapidly, a deeper tension is emerging. Many building these systems acknowledge the risks, yet development continues to accelerate. Insights from journalist Karen Hao, based on 300+ insider interviews, suggest that key decisions are concentrated within a small group, and public narratives don’t always reflect internal realities.
This is no longer just a technology story. It’s about incentives, competition, and control. The race is being driven as much by power and positioning as by progress, raising questions about who is shaping AI’s future.
👉 The takeaway: AI progress isn’t just about innovation, it’s about who controls its direction.
💡 Anthropic Just Leaked The World’s Most Powerful AI
Anthropic accidentally leaked their next AI model - a draft blog post left in a public data store revealed Claude Mythos, a new tier above Opus that Anthropic confirmed is their most capable model yet. The catch? It's so powerful at cybersecurity tasks they're rolling it out to defenders first before anyone else gets access. Oh, and the leak also exposed a secret CEO retreat with Dario Amodei. Classic "human error."
👉 The takeaway: More powerful AI isn’t just about capability anymore, it’s about who gets to use it first and how safely it’s deployed.
Assistants vs Agents (and where you should play)
An AI assistant waits for instructions and helps execute tasks.
An AI agent works toward a goal, takes initiative, and continues operating without constant prompts. That difference is where the shift is happening.
What most people are still doing
Working with assistants:
Asking questions
Refining outputs
Driving each step
This improves speed, but not leverage.
What’s starting to matter more
Designing agent-like systems:
Define a goal, not just a task
Structure steps, tools, and decision points
Let the system run and adjust
This is the move from using AI to structuring execution.
How to start building this skill
Pick one repeat task
Define the outcome clearly
Map the steps and decisions
Separate what AI can handle vs what needs rules
Then test: can it run without constant prompting?
The takeaway: Assistants make you faster. Agents make you scalable. The advantage is no longer usage. It’s designing systems that keep working without you!
The Interview Question That Matters: Workflow Thinking
“How would you use AI to improve a process, not just a task?”
What they’re really testing:
Can you think in workflows, not prompts?
How to answer:
Start with a real example. Briefly describe the process, then break it into steps. Explain how AI can handle each step end to end, including decision points and validation.
Strong answer structure:
Define the outcome
Map the steps (input → actions → output)
Show where AI fits
Mention how you would validate or check outputs
👉 The key is to show you can design a flow, not just use a tool.
This week, notice how AI moves from answering to actually executing steps, how it plans, acts, and adjusts along the way.
Next week, we step into what happens after that, how to make those systems reliable, controlled, and ready to run without breaking.
Link to ➡️ Previous Volume
💛 If this helped, feel free to share it with someone learning AI. 💛
