Hacker News front pageJul 30, 2026, 03:07 UTC+8

How much to delegate to agents depends on the task, not the model

How much can you delegate to agents?

Jina Yoon from PostHog offers a two-factor framework: how easy is it to verify the agent's output, and how cheap is it to undo mistakes. Tasks fall into four levels—from assistant-only for hard-to-check, costly-to-undo work, to full self-driving when both are easy. A real example is a feature-flag engine migration where risky core changes stayed manual while SDK propagation was handed off to agents.

Why it matters: A first-person framework from a PostHog engineer with concrete quadrants and examples—not fluff. Capped at 72 because it's a single blog post, not a product launch or research breakthrough.

Read the original ↗Export Markdown

Take

This is worth a click because it gives you a delegation framework that doesn't depend on chasing model benchmarks. Jina Yoon splits tasks along two axes: how easy it is to verify the output, and how cheap it is to undo mistakes. Hard-to-check, costly-to-undo work stays at assistant level—think core feature flag engine changes where a human writes the risky parts and agents handle SDK propagation. Easy-to-check, hard-to-undo is the default ceiling for most dev work today. Easy on both fronts can go full self-driving.

I'd save this decision tree. It doesn't argue from model capability but from engineering controllability—you can map any task to a level yourself. The feature flag migration example is honest: core logic stayed manual because the blast radius was huge, not because the model couldn't handle it.

What's missing is the concrete bar for "easy to check"—what test coverage or integration test speed counts? You'll need to fill that in yourself.

Source: Hacker News front page · newsletter.posthog.com
Published: Jul 30, 2026, 03:07 UTC+8 · Jul 30, 2026