Are AI agents reintroducing problems software engineering already solved?

Market insights

Stack

Apply

Upload your CV — we will connect you with the employer directly through our pool.

Description

Are AI agents reintroducing problems software engineering already solved? Working with agent workflows lately, I've started feeling like we're just reintroducing a bunch of problems software engineering already spent years solving. Once an agent gets past the "Hello World" stage, its behavior depends on a mix of prompts, tool permissions, memory, retrieval settings, and whatever model endpoint happens to be up. A lot of that state is runtime-driven or buried inside framework abstractions. Trying to reliably review, reproduce, or audit it becomes much harder compared to the static code workflows most of us are used to. We've spent decades building mature workflows around version control, CI/CD, PR reviews, rollback capability, and environment separation so you actually know what binary is running in prod and what changed since the last incident. With agents, a lot of behavior still seems to be assembled dynamically at runtime instead of being treated as a properly versioned artifact. How are teams actually handling this in production? Are people moving toward declarative, git-based definitions for agent workflows, or is the ecosystem still too fragmented and framework-specific for that to work cleanly? GitHub Next shipped Agentic Workflows, gitagent exists, and Claude Code already leans heavily into git-native workflows. The direction clearly has traction now, even if the ecosystem hasn't converged yet. [link] [handle]

Employer contacts (email/phone/telegram) are hidden from the public preview — send your CV, and we will connect you directly.