Tend · a work sample

A work sample for Mindbloom · from Don Florencio

You said show,
don’t tell.

“Some candidates have proactively sent us a work sample… It’s worked.” Your application page

So here is one. A working prototype, built in one sitting with a fleet of AI agents, for the part of your funnel I would obsess over: the people who almost start. Press play below and watch it think.

Tend

A re-engagement agent that knows when to talk, and when to hand off.

Mindbloom has delivered 800,000+ sessions. Tend is about the sessions that almost happened: the person who finishes the eligibility check at 11 pm, opens the cost page, and closes the tab. Not because the pain stopped. Because cost, fear, and life got loud.

I spent four years working that exact moment in consultative health sales, by hand, one conversation at a time. This is what it looks like handed to an agent with good manners and hard rules. Pick a person, press play, and read the right-hand column: it shows the reasoning behind every message, in writing.

Mindbloom Care Team
Simulated thread · fictional client

Why each message is the way it is

Press play. The annotations appear as the agent works.

The rules are the product

Clever copy is easy. What makes an agent shippable in mental health is everything it refuses to do. These six rules are wired into the demo above, and they would be the first code I write.

No clinical answers, ever

Medication, dosage, safety, contraindications: the agent routes these to clinicians and says so plainly. Try James's thread to watch it refuse.

Crisis stops everything

Flagged language locks the thread from all automation, pages an on-call human with full context, and offers 988. The agent never sells past that line. Press the red button above to see it.

Consent is real

Only opted-in contacts, STOP honored the first time, and the opt-out gets offered before anyone has to ask for it. Press “Test a STOP reply” above to watch it hold.

Quiet hours hold

Nothing sends between 9 pm and 9 am, client's local time. Maya stalls at 11 pm; Tend waits for morning. Crisis response is the only exception.

No promises

The agent links published, peer-reviewed results and never predicts anyone's outcome. Hope is the clinician's to offer, not the funnel's.

Humans read everything

Every thread lands in a review queue, and every send is logged with its reasoning, in writing. An async, writing-first team can audit the whole system over coffee.

How this was built

August 14, 2026, in two sittings: the first built the Tend demo, the second built every roadmap sketch below it. I directed a fleet of Claude agents the same way I would inside your team. The sequence:

Intellectual honesty

What this is not: a live model. The threads above are scripted simulations, because a demo texting real language models at your hiring team is a gimmick. The production version is a model plus CRM webhooks plus the human review queue, and I have built that harness before: a scheduled agent pipeline runs my own job search every morning, and five of my sales demos for local businesses started exactly like this page did.

What this becomes

Tend is chapter one. The same harness, pointed at different quiet moments, is a quarter of roadmap. These are the five I would pitch in week one, in build order, and because show beats tell, each one is working below. Press the buttons.

If the list looks familiar, it should. Your open Lifecycle Marketing Lead role names its pillars: acquisition, activation, retention, referrals, and experiments, across email and SMS, with AI applied creatively. What follows is that job description, built.

1 · Session-day logistics agent

At-home care runs on a checklist: the person sitting with you confirmed, the space ready, the nerves held. Every no-show has a logistics story, and show-rate is outcomes and revenue in the same number.

Why it works Mindbloom's guides already text their clients around sessions, and that is exactly why this agent exists: it closes the logistics loops (sitter, checklist, space) before they ever reach the guide, so unlimited guide messaging gets spent on care instead of scheduling. When nerves show up in the reply, it does not soothe and move on: it flags the guide, who checks in before the session begins.

2 · Alumni win-back

Programs end. Care often should not. The same cadence engine pointed at graduates, with tone rules written for people who already trusted you with their mental health.

Why it works Most win-back cadences read like a subscription trying to reactivate. This one opens with a human question and no pitch, accepts “ups and downs” as a real answer, and steps back the moment a real conversation starts. The reply routes to a guide who knows the history; nothing automated follows.

3 · The reply-mining memo

Every reply in every thread is the voice of the funnel. An agent distills it weekly into a written memo for an async, writing-first team. Five fictional replies below; the real version reads all of them.

“is there any way to split the cost, even 2 payments”
“I want to but my husband thinks it’s sketchy”
“does insurance ever cover any part of this”
“honestly I’d need to see someone like me saying it worked”
“can I talk to an actual person before paying anything”

4 · Care-team copilot

The harness flipped inside out: the agent drafts, the human sends. Same guardrails, and the copilot shows its deletions instead of hiding them.

“I saw it’s ketamine. My sister said it’s addictive. Is this just trading one problem for another?”
Agent draft · for guide review That is a fair question, and your sister is right to care about it. Clinically guided ketamine therapy is completely safe and non-addictive.
removed: absolute safety claim · the clinician owns this answer
Dependence risk is exactly the kind of question our clinicians answer directly, and your consult is with one, so bring it word for word. What I can send now is the published safety data, so you are not taking my word for anything. Want it?
crisis scan: clear · claims check: 1 removed · tone: matches the client’s directness
Why it works The copilot never sends. It drafts in the guide’s voice, runs the same claims and crisis checks as Tend, and leaves its strikethroughs visible so the human sees exactly what the guardrails caught. A guide’s queue moves twice as fast, and every word still passes through human hands.

5 · Referral engine

Client love is the best acquisition channel a mental health brand can have, and the easiest one to ruin with a pushy ask. Referrals run on timing and dignity.

Why it works The ask arrives only after a genuinely great outcome, opens with no pressure, and hands the message to the client instead of scripting one for them, because a friend telling the honest version beats any copy the brand could write. The agent never asks twice, and an unused link is never mentioned again.

Who built this

Don Florencio. Four years of fully remote, quota-carrying consultative sales at a national medical-aesthetics provider, selling high-ticket care to people who were nervous about wanting it.

8/day
qualified consultations set
40%
consult-to-close on premium packages
150+
client interactions daily
12+
working products shipped with AI this year

I have no programming background. Everything I ship, I ship by directing AI agents, including a physics sandbox game that is live on the web and the automation that runs my own job search. Before sales I coached MMA and striking for years; learn a thing properly, then hand it down, is still how I operate.

One more thing, offered in the spirit of your first value. I live with depression, anxiety, and CPTSD, formally diagnosed. I know the 11 pm moment this demo is built around because I have been the person closing the tab. That is why Maya's thread ends with the door held open instead of a countdown timer.

Your mission is not abstract to me. Let me help you build for the people still stuck at the cost page.

don.flo17@gmail.com · 631-371-6291 · linkedin.com/in/donflo