Engineering notes and case studies on autonomous business agents — self-improving loops, voice support, and email operations.
Serious buyers browse listings at night, and portal leads go cold in minutes. Where after-hours buyer and seller inquiries go to die, what an answering layer actually has to do for a licensed agent — minutes-fast response, structured qualification, deterministic scoring, no property advice, no outbound — and the honest limits of what runs today.
Read the post →Water does not check posted hours. Where after-hours tenant calls go to die, what an answering layer actually has to do under a management agreement — emergency classification, vendor dispatch, timestamped records, no maintenance advice — and the honest limits of what runs today.
Read the post →Arrests, accidents, and served papers do not keep office hours. Where after-hours calls to a law firm go to die, what an intake layer actually has to do at 11pm — screening, conflict-check capture, no legal advice — and the honest limits of what runs today.
Read the post →Tooth pain does not keep office hours — the calls that land after dark are often the highest-intent ones a dental practice gets. Where after-hours calls go to die, what an answering layer actually has to do at 9pm — triage, emergency classification, rescheduling, no clinical advice — and the honest limits of what runs today.
Read the post →A furnace dies and a pipe bursts on their own schedule — the calls that land after dark are often the highest-value jobs a home services company gets. Where after-hours calls go to die, what an answering layer actually has to do at 2am — emergency triage, dispatch capture, timestamped records — and the honest limits of what runs today.
Read the post →An honest side-by-side on cost, hours, consistency, escalation, and audit trails — where a human answering service still wins, what AI changes, and why hybrid triage usually beats both.
Read the post →Metric ingestion, week-over-week deltas, anomaly flags routed to owners — the draft-to-owner workflow and guardrails that keep an automated weekly report honest.
Read the post →Deterministic invoice aging, dispute detection routed straight to humans, payment-promise pauses, and hard touch caps — how to automate dunning without ever threatening a customer.
Read the post →Deterministic request-and-nudge ladders for tax docs, contracts, and KYC files — extension pauses, hard touch caps, a tone gate on every draft, never chasing received items, and a 30-day human handoff.
Read the post →Booking, rescheduling, no-show recovery — the four jobs of a scheduling agent, the deterministic guardrails that keep it safe, and the four metrics that prove it pays for itself.
Read the post →Triage first, draft second, send third — the staged playbook with approval gates, audit trails, and the four metrics that prove your shared inbox automation works.
Read the post →An honest guide: what voice AI does well, what it shouldn't touch, how to read the pricing, and a five-point checklist before you sign anything.
Read the post →Why every agent cycle should leave the system smarter than it found it — loop logs, lessons, nightly quality gates, and why improvement compounds.
Read the post →After-hours calls are customers quietly shopping your competitor. What good voice AI coverage looks like — and how we build it to be swappable and accountable.
Read the post →Transparent 0–100 scoring rules, hot/warm/cold tiers, and routing instead of outreach — how to automate lead triage with a hard no-contact guarantee that keeps humans in charge of first touch.
Read the post →A guardrail-first dunning agent: deterministic aging ladder, disputed invoices frozen and routed to humans, payment-promise pauses, and a banned-phrase tone gate on every draft. Results framework to be filled from real pilot data only.
Read the case study → Case studyGuardrail-first appointment automation: how a scheduling agent books, reschedules, and refuses — with zero double-bookings in our simulated dry-run.
Read the case study → Case study · SkeletonA draft-and-approval-gated Email Operations Agent for a two-person team drowning in 400 emails/day. Results framework to be filled from real pilot data only.
Read the post → Case study · SkeletonA deterministic weekly reporting agent: WoW deltas, ±20% anomaly flags routed same-day, and missing metrics surfaced instead of silently omitted. Results framework to be filled from real pilot data only.
Read the case study → Case study · SkeletonA no-contact agent that scores every inbound lead hot/warm/cold with an explicit reason and routes only qualified interest to sales. Results framework to be filled from real pilot data only.
Read the case study → Case study · SkeletonA guardrail-first document-chase agent for onboarding: request→nudge→firm ladder, extension-promise pauses, never chasing received docs, day-14 human escalation, and a banned-phrase tone gate. Simulated dry run: 79 docs / 37 tone-gated chases / 29 human escalations.
Read the case study → Boundaries · Service fitWhy inbound-only agents are structurally incapable of outreach — no newsletters, no social posting, no cold contact by design — and why that same architecture is exactly what makes them safe to point at your customers.
Read the post → Voice · Revenue recoveryA $19-a-month auto-text is a fine commodity purchase if all you need is an apology SMS. Here is where static text-back stops, and when answering, qualifying, and booking the recovered caller end-to-end is worth stepping up to.
Read the post → Scheduling · No-showsActive confirmations beat passive reminders, sequences beat single pings, and code-enforced touch caps keep reminders from turning into nagging — plus booking-side prevention. Every industry figure attributed to its source.
Read the post → Quotes · Follow-up disciplineA deterministic follow-up ladder for sent estimates — scheduled touches, promise pauses, same-day human routing on objections, structural touch caps — plus the honest fit test for when your FSM's built-in reminders are enough.
Read the post →