Modus Flows
Designing an AI voice agent that knows exactly where to stop.
Capable is easy. Restraint is the design.
Go ahead. Call one. It picks up.
The call that goes to whoever answers first
A homeowner's AC quits. They search "AC repair near me," pick the two or three with good reviews, and start calling down the list. The first contractor who picks up and sounds competent usually takes the work. Getting a second quote is a hassle, so homeowners rarely call back the contractors who missed them.
The job wasn't lost at the top of the funnel. It was lost the moment nobody picked up.
Voicemail is the thing callers are already hanging up on. An answering service is a stranger reading from a card who can't book anything. A receptionist doesn't cover nights and weekends, which is when emergencies happen. Modus Flows is the company I built solo on HighLevel. Its product is an AI voice agent for home-service contractors: it answers the calls the owner can't, picks up, captures the caller's details, handles basic questions, and books appointments.
Tension
AI beats voicemail, not humans.
One positioning decision the whole product lives or dies on
Quotes the job, handles pricing, works the objection, closes the sale. Easier to sell, easier to charge for. It demos like the future, and it breaks down the moment a real caller feels like they're being sold to, not helped.
Contractors don't fear voicemail. They fear handing their business to a robot. Every added capability is one more way for a machine to damage a relationship while wearing someone else's name.
Most of the design work was deciding what the agent would never do.
Tune every voice by ear, not by template
HighLevel offers a menu of preset voices, and each starts from a different place. One is flat, one runs too hot, one carries an accent that changes how it reads. The settings that govern behavior, how long it waits through a pause, how fast it talks, whether it matches the caller's pace, aren't portable. What brings one voice to a respectable bar makes another worse.
Most conversational design isn't what gets said. It's what gets allowed to happen between sentences.
No portable configuration, no shortcut, fresh tuning work for every new voice. Accepted on purpose, because the portable version produces agents that sound like a machine reading a card, the exact thing the caller was hanging up on.
Cap the agent at capture-and-handoff, and let the rest go
The agent's job is drawn tight, and everything from the impressive version is deliberately out of bounds. Those aren't the risky tasks by accident. They're the owned tasks, where the contractor makes their margin and builds the relationship. The line where the agent stops is drawn precisely where the human's value begins.
- Name and number
- Address
- The problem, briefly
- A day and a time
- Books the appointment
- Quote the job
- Handle the pricing
- Work the objection
- Close the sale
- Own the relationship
Where the agent stops, the human's value begins.
There's a second reason to keep it short, and it's mechanical. Every extra field and every extra minute is another chance to fail: to repeat itself, stall, wander down a rabbit hole, or say something untrue. A longer call isn't more impressive. It's more exposed.
The summary does double duty: it's the lead, and it's a receipt. A skeptical owner can read back what the agent said on any call and confirm it stayed in its lane.
After a first-ever conversation with an AI, they walk away knowing they were heard and someone is coming on a real date, instead of a message left in the dark.
A quoting, closing agent is more impressive and easier to sell, and that was given up on purpose. It's the same boundary a lot of companies are now redrawing the hard way. Modus Flows started from that line instead of backing into it.
Call it. It's live.
The agents are live. Call one cold right now and you get a conversation that holds: it asks its questions, captures the lead, books the appointment, and hands off. Both decisions are audible in that call, the by-ear tuning in how the agent sounds, and the restraint in what it reaches for and what it leaves alone.
The proof isn't a screenshot. It's a phone call.