teffer.dev
leadership

finally writing

by teffer · · 5 min

image: NASA / ESA / CSA — James Webb Space Telescope

I lead a large organization in AWS. Our mission is to invent the future of work with AI and human collaboration. On a typical Tuesday I’m in conversations about how agentic AI is reshaping the way teams do their jobs — what voice, chat, and case work look like when an LLM is in the loop with a customer or the human helping them.

On Saturdays I’m in a different conversation. At home, building things for myself with AI tools my team is using to build for our customers. A personal knowledge base. A fitness application. This blog. I write a prompt, an agent writes some code, I read it, iterate, repeat. Less than an hour from empty directory to live site for the page you’re reading now. Domain registered, DNS pointed, site built, this post drafted. I have thought about writing a personal blog for years, but AI encouraged me to get started.

The reason I’m writing publicly is that those two conversations — the leadership one at work, the practitioner one at home — keep me busy learning about this transformation of how we work with technology. AI doesn’t replace great leadership, the importance of critical thought, or an engineering mindset. I spend as much time today asking Claude Code how things work and to help me learn about concepts. This is my way of giving back. If any of it helps someone else, even better.

The principles that make a good leader help steer AI

I’ve spent years reading, learning, and refining a personal leadership philosophy. Sometimes learning from success as a leader, and sometimes from failures. I have around twenty principles, all earned along the way. Some examples: Give good data and listen for understanding. Push decision making down. Move with intent. Be curious, not judgemental. Amplify each other. Iterate to Awesome.

These were learned leading teams of people. But many of these map onto working with teams of agents.

A few that have helped me:

  • Give good data, listen for understanding. When I direct an agent the way I once directed a junior engineer — vague intent, poor mental model, no examples, no constraints — I get poor output back. Good data in, good output out. Bad data in, confidently wrong agent output out. The agent doesn’t push back the way a person will unless you ask it to. Ask good questions, encourage feedback, understand the whole context not just the specifics you prompted about.
  • Plan before you implement. Skipping the plan with humans can lead to wrong assumptions, poor designs, wasteful iterations. Skipping the plan with an agent costs you a fully-built, well-tested version of the wrong thing. The cost of unwinding that can be higher than the cost of unwinding broken code, because everyone — including you — may think the work is done, safe, and high quality.
  • Be curious, not judgemental. When something goes sideways the question isn’t “is Claude getting worse?” It’s “what context did I leave out?” Ask the model to give you suggestions on how to improve prompts or context. Learn as you work.
  • Push decisions down. With humans, distributed decision-making is the difference between an org that moves at the speed of one decision-maker and one that moves at the speed of hundreds. With agents, distributed decision-making is operating reality from prompt one — every choice you delegate is a choice you’ve decided you don’t need to make. The skill of which decisions to delegate well is the key.

Great leadership still leads to great teams, now those teams are composed of people and AI teammates.

What this blog will be

  • Leadership — what I’m seeing and trying while navigating this AI transition.
  • Building — what I’m working on personally, including the things I make on weekends.
  • AI tools — my experience using AI tools like Claude Code.
  • Stories — anecdotes and lessons from my personal and professional life.

I’ll be honest. I’ll be wrong about things. I’ll change my mind. I’ll prefer specifics over abstractions and personal anecdotes over predictions. I haven’t written a blog before, so this could go terribly. As with all things, I will do my best.

If any of that resonates — if you’re another leader trying to figure this out, an engineer wondering what your leaders think, or someone interested in the overlap between leading teams and building with agents — I’d like to find you. The RSS feed works. The about page has a way to reach me.

Welcome.