Why the best developers are writing less code than ever
Lena Hall, Sr. Director of Developer Relations at Akamai on why the most valuable developers in the AI era are the ones who've stopped typing.
A conversation with Lena Hall, Sr. Director of Developer Relations at Akamai, ex-AWS, ex-Microsoft Research.
Something quietly shifted in the last 12 months.
The senior engineers moving fastest right now - the principal engineers, the architects - are spending 80% of their time writing specs, not code. They’re defining inputs and outputs, mapping component contracts, eliminating ambiguity before an agent ever touches the implementation.
If that makes you uncomfortable, it should. It means the game has changed - and what made you good yesterday may not be what makes you valuable tomorrow.
I sat down with Lena Hall to talk about this. Lena has built distributed systems at Microsoft Research, led developer relations across AWS, and now drives AI infrastructure strategy at Akamai. She’s watched this shift happen in real time.
“Code is becoming like binaries. We don’t manage binaries - they’re generated. Code is heading the same way. The question is: what does that make you?”
- Lena Hall
The Role Is Changing. Here’s What It’s Changing Into.
Developers aren’t becoming obsolete. They’re becoming architects - and the best ones are operating more like CTOs. They own the logic, the system design, the edge cases. They define what needs to be built with enough clarity that an AI agent can execute it reliably.
But here’s where Lena’s technical background adds a layer most people miss: AI agents are non-deterministic by nature. If you don’t control them structurally - with structured outputs, phased execution, and human checkpoints at high-stakes decision points - they will break in ways you can’t predict or explain.
She calls this pragmatic AI: architecture matched to the stakes of the business problem. Low-stakes tasks can tolerate some ambiguity. High-stakes tasks - financial decisions, healthcare workflows, anything with legal exposure cannot. The expert must be in the loop before the system ships, not after.
What You’ll Take Away From This Episode
Why the Two Generals Problem from distributed systems applies directly to every LLM call you make
The 3-tier framework for deciding how much AI control your use case actually needs
Why excluding domain experts from your AI workflow is the #1 mistake teams make
The one habit that separates developers who ship reliable AI from those who don’t: fix the spec, not the output
WATCH THE FULL EPISODE - AgentBuild Expert Exchange
Whether you’re writing code every day or managing teams that do - this one reframes how you think about where your value actually lives in the AI era.
Thanks, and please leave some comments on the video.
-Sandi.
Ask your friends to join.
More valuable content coming your way.