I sell residential HVAC, electrical, and plumbing in Tampa Bay, and I build AI systems for trade shops through Eureka Pathways (flagship is OnCall Desk — 24/7 AI phone answering + booking).
I’ve been using Cursor agents to build and maintain the unglamorous parts of a home-services sales stack: intake → follow-up → dispatch handoff → rehash on quotes that didn’t close. Same job I’ve done for 27 years in someone’s driveway, just with agents doing the remember-to-run-it part.
Curious what others are doing with Cursor agents in this domain (or adjacent field sales):
1. Voice / SMS intake agents that book real appointments into a CRM
2. Automating rehash / follow-up without spamming homeowners
3. Keeping “human handoff” clean when a call needs a live person
4. Anything that broke when you tried to automate an in-home sales process
Happy to share what’s worked (and what I refused to build because the process wasn’t written down yet). If you’re in trades or building for them, drop your stack / lessons.
Hey Kevin, this is a great case. “Agents take on the remember-to-run-it part” really captures where a lot of users are heading with the automation stack.
A couple things from Cursor that map directly to what you’re describing:
Automations: for recurring or triggered tasks like rehashing open quotes and follow-ups. The agent runs on a schedule or an event, not manually. Automations | Cursor Docs
Cloud Agents: if you want part of the pipeline to run autonomously, not inside an open IDE. Cloud Agents | Cursor Docs
Rules: to keep the agent consistently following your rules, like “don’t touch a process that isn’t documented yet,” tone when talking to the homeowner, and human handoff criteria. This helps a lot with clean handoff. Rules | Cursor Docs
On your point 4 about what broke when automating in-home workflows, I’d love to hear specifics from others in the thread too. Where exactly does the handoff fail, and how are users setting guardrails so follow-up doesn’t turn into spam. That’s usually the hardest part to formalize.
Let me know if you want to go deeper on any of these scenarios. Hopefully we’ll get more input from other users in trades.
Hey Dean… thanks. The three you named map cleanly to how I think about the stack.
Where in-home sales automation actually broke for me (and why I refuse to build certain agents yet):
1. Follow-up without a written cadence turns into spam.
If rehash rules aren’t explicit (day 1 / day 3 / day 7, channel, stop conditions), an agent will “helpfully” keep pinging. I only automate what I’ve already run by hand for months.
2. Handoff fails when context doesn’t travel.
Voice/SMS books the appointment, but the live person gets “new lead” with no transcript, homeowner intent, or objection. Guardrail I use: no appointment is “done” until a short handoff packet exists (who, what they want, urgency, quote status, last objection).
3. CRM truth vs driveway truth.
Agents are great at remembering to run the process. They’re terrible when the process in Jobber/ServiceTitan doesn’t match what actually happens on the visit. Automating a dirty pipeline just scales the mess.
4. Tone + consent.
Homeowner SMS is not Slack. Rules have to encode quiet hours, no double-text after no reply, and “human only” triggers (price fight, complaint, elderly homeowner who asks for a person).
Cursor-wise I’m using Automations for the remember-to-run-it layer and Rules for the “don’t touch undocumented process / don’t spam / escalate when X” layer. Cloud Agents for longer pipeline work that shouldn’t depend on my laptop being open.
Happy to go deeper on the handoff packet format or the stop-conditions I use on rehash if useful to the thread.
Hey Kevin, this is exactly the kind of specific detail these threads are for. Point 3 in particular, “automating a dirty pipeline just scales the mess”, is something worth putting on the wall. Agents don’t fix a process, they scale the one you already have, with all its holes.
Yeah, let’s do both breakdowns, the handoff packet format and the stop conditions for rehash. I think that’ll be the most useful thing for everyone else in the thread, because this is where most setups fall apart.
A couple questions to pull the most value for readers:
Handoff packet: do you keep it as a structured template in Rules so the agent always collects the same fields, like who, intent, urgency, quote status, last objection? Or is it a separate step in the pipeline right before an appointment gets marked done?
Stop conditions: do you encode quiet hours and “no double text after no reply” in Rules, and the “human only” triggers like price pushback, complaint, elderly homeowner through the same Rules, or is that a separate escalation layer?
The way you separate “remember to run it” Automations and “don’t touch undocumented, don’t spam, escalate when X” Rules is a solid pattern that applies way beyond trades. If you can write out the two formats, I think others building a field sales stack will jump in too.