AI Engineering

What the sources say

These sources explore the transition from traditional coding to agentic software engineering, focusing on autonomous workflows, specialized models, and the systems needed to manage them.

Experts agree that the developer role is changing from writing code to managing AI agents. Mihail and Ram Koning note this requires building agent friendly codebases where humans orchestrate systems^[5]. Boris Cherny from Anthropic adds that tools like Claude Code shift engineers to reviewers, moving work away from traditional environments^[6]. Andy Dev Dan highlights Stripe's Minions as a prime example, showing how enterprise teams build deterministic layers around AI to handle thousands of code changes autonomously^[7].

Sources differ on where engineers should focus their attention. Theo and Ben actively compare the capabilities of unreleased models like OpenAI Saul and Anthropic Fable^[3]. However, Matt argues in an interview with David that chasing new models is a mistake and developers should focus on the surrounding architecture instead^[4]. Similarly, while some agents handle complex tasks, decision models like Jev remove conversation and web search entirely to provide fast judgments directly to code^[1].

Readers can take away practical strategies for adopting these agentic workflows. They can look to Matt Pocock's repository to integrate specific skills into their development process^[2]. Ultimately, as Andy Dev Dan points out, teams must stop coding the application directly and start building the system that writes the application^[7]. This requires prioritizing the software harness over the engine^[4].

Written by an AI model from Google, from the analyses on this page.