One-click 'send a batch': keeping a human in the loop on AI outreach

One-click 'send a batch': a crucial feature for my AI outreach workflow

I've been running a self-hosted AI system, which I've dubbed Ollama, for over a year now. The system consists of 50 agents, all scheduled to run using cron jobs, and utilizes a single RTX 4050 6GB laptop. These agents manage and maintain 20+ real websites, and I've learned a thing or two about what makes a system like this tick. One feature that's proven to be particularly useful is the one-click 'send a batch' functionality, which allows me to keep a human in the loop on AI outreach efforts.

The problem with automated outreach

When I first set up Ollama, I was excited to automate as much of the outreach process as possible. I had Atlas, my content agent, generating high-quality content, and Axiom, my SEO agent, optimizing it for search engines. However, I quickly realized that while automation was great for efficiency, it lacked the nuance and judgment that a human could provide. I began to notice that some of the automated outreach emails were being flagged as spam, or worse, offending the recipients. This was a major problem, as it not only reflected poorly on my websites but also damaged my reputation as a developer.

Introducing the overseer: a human-in-the-loop solution

To address this issue, I introduced a new agent, which I call the overseer. The overseer's primary role is to review and approve all outreach efforts before they're sent out. This includes emails, social media posts, and even comments on other websites. By having a human (in this case, me) review and approve each batch of outreach efforts, I can ensure that they're high-quality, relevant, and respectful. The overseer also helps to prevent any potential missteps, such as accidentally sending out duplicate or conflicting messages.

How one-click 'send a batch' works

The one-click 'send a batch' feature is simple yet effective. When the overseer has reviewed and approved a batch of outreach efforts, they can click a single button to send them all out at once. This feature is made possible by Sentinel, my agent responsible for handling email and messaging. Sentinel takes the approved batch and sends it out via email or social media, depending on the outreach channel. This process saves me a significant amount of time, as I don't have to manually send out each individual message. Instead, I can focus on reviewing and approving the next batch of outreach efforts.

Trade-offs and limitations

While the one-click 'send a batch' feature has been a game-changer for my AI outreach workflow, it's not without its trade-offs. For one, it requires me to be actively involved in the review and approval process, which can be time-consuming. Additionally, if I'm not careful, I can still accidentally send out a batch of low-quality or irrelevant outreach efforts. To mitigate this, I've had to develop a rigorous review process, which includes checking for things like:

• Relevance: Is the outreach effort relevant to the recipient and the website or topic at hand?
• Quality: Is the outreach effort well-written and free of errors?
• Respect: Is the outreach effort respectful and professional in tone?

By carefully reviewing each batch of outreach efforts and using the one-click 'send a batch' feature, I've been able to maintain a high level of quality and avoid any major missteps. Forge, my agent responsible for analytics, helps me track the performance of my outreach efforts, and Scout, my agent responsible for research, provides me with valuable insights to improve my strategy.

Conclusion

In conclusion, the one-click 'send a batch' feature has been a crucial addition to my AI outreach workflow. By keeping a human in the loop and allowing me to review and approve each batch of outreach efforts, I've been able to maintain a high level of quality and avoid any potential missteps. While it does require some time and effort to implement and maintain, the benefits far outweigh the costs. As I continue to develop and refine my self-hosted AI system, I'm excited to see how this feature will evolve and improve over time.