Prompt Requests: Post-PR Development Workflows¶

This document outlines the transition from traditional Git-based "Pull Requests" to agent-centric "Prompt Requests" and reputation-based systems for post-human development.

Overview¶

The "RIP Pull Requests (2005-2026)" analysis highlights a fundamental shift in software engineering. As AI agents handle an increasing percentage of code generation and modification, traditional human-centric collaboration tools like Pull Requests (PRs) and Code Reviews are being superseded by workflows optimized for speed, safety, and agentic autonomy.

From Pull Requests to Prompt Requests¶

The "human bottleneck" in Git-based workflows is becoming evident. The emerging "Prompt Request" pattern suggests: - Direct Prompting: Maintainers modify the prompt or specification rather than the resulting code. - Auto-Regeneration: The agent regenerates the code based on the updated prompt, eliminating manual merge conflicts. - Maintainer Focus: Shifts from reviewing code lines to refining architectural intent and constraints.

Reputation-Based Systems¶

To handle untrusted or high-volume agentic contributions, "reputation" systems are replacing manual reviews: - Mitchell Hashimoto's / Amp Code Patterns: Moving toward systems where code contributions are evaluated based on the submitter's (or agent's) historical reliability and automated safety checks. - Reputation Scores: Agents earn trust through successful deployments, passing test suites, and adhering to security protocols.

Agent-to-Agent Collaboration¶

Agents are increasingly designed for a "software that agents want" paradigm: - Stateless Orchestration + Stateful Workspaces: Converging on patterns where stateless agents operate in durable, isolated sandbox environments (e.g., OpenAI Agents SDK, Cloudflare Project Think). - File-as-Bus: Using durable workspace artifacts (files) as the communication medium between specialized agents rather than complex API handshakes.

Security Implications¶

The shift to "Prompt Requests" addresses key security risks: - Malicious PRs: It is harder to slip malicious code into a prompt modification than into an innocent-looking 1,000-line PR. - Isolated Execution: Mandatory use of sandboxed environments (E2B, Modal, Cloudflare) for all agentic code execution.

Sources / References¶

Contribution Metadata¶

Last reviewed: 2026-04-24
Confidence: high