Salesforce AI Researcher's Annual Review: Shifting from Hand-Coding to Orchestrating Multiple Coding Agents
I use AI (Oct 2026)
A researcher at Salesforce AI Research published an annual review of their AI usage, noting that their daily work has fully shifted from manual coding to orchestrating five to six coding agents—such as Claude Code—simultaneously, with machines now writing over 90% of the code. The author highlighted that while agents dramatically speed up prototyping, they still lack architectural design capabilities and introduce managerial fatigue. Furthermore, while LLMs excel at tutoring and search, they remain unable to replace humans in complex original research and independent writing.
The author systematically examines the practical workflows and limitations of front-line engineers managing multiple coding agents, providing readers with insights into defining the division of labor when deploying LLMs in software development.
Source:Hacker News · AI · ben.stolovitz.com