Ask HN: Another AI agent, this runs CRM follow‑ups (200msg in prod). Worth it?

1 points by ebenes 259 days ago | 1 comment

I’m a software developer hacking on a proof‑of‑concept for a friend’s sales agency. The idea is a lightweight agent that plugs into their existing CRM (AgencyZoom) and handles repetitive lead follow‑ups.

It’s already running in production in a semi‑automatic way (every action still goes through human approval). So far the system has sent about 200 texts. Too early to measure impact, but it’s been enough to see interesting behavior.

=== How it’s built:

Almost stateless: core data lives in the CRM, with SQLite for caching / minimal state. Written in Typescript, using a lot of functional programming. Trying to follow Domain Driven Design patterns, which keep the logic and bounded contexts manageable. Rate limits and random API errors were a pain: solved with higher‑order functions wrapping API calls. Challenges so far:

Prompt tuning: lots of “close but not quite” AI decisions, especially edge cases where I’d act differently as a human. Infrastructure: significantly more code around state, retries, and API integration than around the actual LLM.

=== My question to you:

Does this sound like something worth pursuing further? Is there a genuine gap here, or am I just reinventing existing automation tools?

I’d love to hear perspectives from people who’ve built or used similar systems.

Emma_Schmidt 256 days ago |

From my perspective, there’s definitely something interesting here. Most existing CRM automation tools feel rigid they’re either rule-based or too generic. An AI driven follow-up system that learns tone, context, and timing could add real value, especially for agencies where every lead touchpoint matters.

That said, the challenge isn’t just sending messages it’s making them feel human, relevant, and trustworthy. If you can nail personalization while keeping approvals in the loop, it could stand apart from typical “drip” automation. The real test will be scalability and measurable ROI. If early results show higher engagement compared to standard workflows, then yes, there’s a gap worth pursuing.