Four AI agent workflows that held up after six months
After six months of trying eight s, four stayed useful in real work. The first is using an AI agent as the first reviewer. CodeRabbit checks every pull request for routine mistakes, while the human review focuses on design and architecture.
The second is generate-then-curate: the AI creates many options, and the human keeps and improves the useful ones. This works because generating is cheap, and choosing from options is easier than starting from nothing. The third is , where the AI can act freely only inside clear limits, such as certain files and commands, with no power to deploy.
The fourth is escalation: the AI tries the task, but hands it to a human when confidence is low. The failed patterns included fully without checkpoints, multi-agent chains, and AI-driven . can lose direction and waste money, multi-agent chains can amplify one early mistake across later steps, and AI-driven lacks and workplace realities.
Key points
- The four useful patterns were first review, generate-then-curate, , and escalation to a human.
- CodeRabbit was used on every pull request to catch routine review issues before a human looked at architecture.
- Generating many options can be valuable because selection is often easier than creating from scratch.
- AI agents need firm limits on files, commands, and risky actions like deploys.
- Fully and multi-agent chains can waste money or spread mistakes.