TTL 57

The other day, I was on a bus trying to figure out whether I should have taken the tram instead to get to my destination quicker.

Somewhere between two stops, I thought someone should have invented an app for this.

And then, in a second, my brain snapped: I described the idea to an AI on my phone and, by the time I got off the bus, I had a design mockup.

Less than twenty-four hours later, I possessed two working prototypes, despite having invested only two hours of my own time.

Here’s the interesting part: I’m not a developer. I’m a product manager.

Which means I’m not an engineer, but I am technical enough and I know how to coordinate work well. And that’s exactly why AI is perfect for me.

The Misunderstanding About AI

There’s a joke going around about AI in my bubble.

Someone says: “Can you believe it? I can just speak English and get things built!” And someone else replies: “I’ve been doing that my whole career. I’m a product manager.”

For years, building software required you to translate ideas into code, or to work through layers of specialists who could. Now, that translation layer is collapsing.

And the people who are best positioned to take advantage of this aren’t the prompt engineers. It’s product managers. People who already know how to describe outcomes, set constraints, delegate work, and evaluate results.

The Evolution of AI

To understand how to manage this technology, we must look at how it evolved.

• We started with the Chatbot phase, which functioned as a conversation with a “brilliant amnesiac.” It was useful for quick lookups, but little else.
  Prompt Engineering was the role of the year.
  
• We then moved into Context Engineering, discovering that context is king and the primary skill is briefing the AI.
  Many people remain stuck in this stage, complaining that the technology lacks memory.
  I see this lack of memory as a feature, not a bug. No memory means no historical baggage: every conversation starts with a clean slate, free from office politics or old grudges.
  Because there is zero context bleed, the exact same system can simultaneously work on a compliance report, a customer email strategy, and a board presentation without interference. Try asking a human to do that.
  
• And this leads us to Agents with tools. Right now, most AI adoption looks like a feature squeezed into existing tools—like a copilot in a spreadsheet. But putting AI in a spreadsheet turns an incredible technology into a glorified calculator.
  The real shift is the reverse: giving tools to the AI and letting it decide what to use. The difference is the gap between having a calculator with financial functions, and hiring an expert analyst and handing them a financial calculator.

---

The Manager’s Toolkit: AI is Staff, not Software

If we accept that AI functions as staff rather than software, we must apply a management framework to lead it.

1. Define the Agent’s Soul (Onboarding)

You wouldn’t hire a person and expect them to figure everything out without a job description and guidance.

You must provide agents with clear roles and constraints.

Take the example of OpenClaw (formerly Maltbot/Clawdbot) out of Vienna. It became the fastest-growing open-source project in history—over 300k GitHub stars in 60 days—because it has personality.

OpenClaw introduced the SOUL.md file: a document defining who the agent is, its voice, its values, and its constraints. When agents have distinct personalities, people stop treating them as tools and start treating them as team members.

2. The Trust Ladder

How do you adopt this safely? With a four-rung trust ladder:

• Observer: Watches, summarizes, and suggests.
• Advisor: Provides options and recommendations for human approval.
• Operator: Executes tasks within a defined scope under your supervision.
• Autonomous: Acts and manages entire workflows independently.

You wouldn’t give a new hire the company credit card on day one, and you shouldn’t grant an AI autonomous power immediately either. Start at the Observer and Advisor levels—which are perfectly viable today because a human remains in the loop—and promote based on performance.

3. Performance Reviews

You must check your agents’ work and give feedback. That feedback goes directly back into the system prompt, allowing the agent to improve. Unlike some human employees, AI actually reads its performance review.

4. Shared Context

Another massive advantage is perfectly shared context. Imagine a system where an agent picks up a project exactly where another left off. No handover meetings: every new “team member” inherits the full institutional knowledge of every decision ever made.

5. Agent Coordination

Once you have multiple agents, they can communicate with each other, hand off tasks, and escalate when needed. An orchestrator organizes the team and distributes work, ensuring the team follows protocols and never has a turf war.

This brings us to the core insight of this entire shift.

The skill that matters most in the AI era is not prompt or context engineering. It is management: delegation, clarity, accountability, and systems thinking.

When people hear “AI productivity,” they often assume it means working longer hours. But the shift is about leverage per unit of time. AI doesn’t add hours to my day; it gives me infinite junior team members to help me. It allows me to execute complex, multi-layered projects from a train, a mountain, or a remote cafe. The constraint is no longer access to resources, but clarity of intent.

---

Stop prompting, start managing. Your digital team is ready when you are 😉

If someone forwarded this to you, you can subscribe.

I also publish on paolo.blog and monochrome.blog.