Lofty AI Sales Agent Now Calls Your Leads, But Should It?

THE APEX AI BRIEF

Yo, what up. Lofty's AI Sales Agent can now call your leads and hand you the phone while the call is still live. But should it? That is the real question, and I am still not using it myself. Here is what changed, what did not, and how to test it before you trust it with a lead.

Yo, what up. If AI touches your follow-up, your texts to buyers, or the way you work your leads in Lofty, this edition is about you. Not in some far off, sci-fi way. In a right now, your CRM, your phone, your license kind of way.

Here is where I stand, so you know where this is coming from. My earlier video on version 4.39 was called "Why I'm STILL not using the AI Sales Agent in Lofty," and that has not changed. I am keeping it real with you. The tech in 4.52 is seriously impressive, and live transfer fixes the exact moment where AI follow-up usually breaks. But just because AI can automate a call does not mean we should let it, not without a test, not without the human handoff, and not without thinking about what the lead on the other end actually experiences. Real estate is a relationship business, and our clients are making some of the biggest financial decisions of their lives. I broke the full update down, versions 4.48 through 4.53, in my latest video, Lofty CRM AI Can Now Call Your Leads... But Should It? This edition is the written version of that same question, with the guardrails spelled out.

Here is the pattern to watch this week. The AI got better at the opening move and worse at keeping secrets, both at the same time. Lofty's Sales Agent can now call a lead, talk to them, and hand the live call to you the second they ask for a human. And Apiiro just proved that a single lazy permission click can publish your passwords to the whole internet. Stronger tools, sharper edges. That is the week. And every one of those things points at the same job for you: give the agent narrow work inside Lofty, keep the human handoff switched on, test it before you trust it, and guard the keys like your license depends on it, because it does.

Story 1: Lofty's AI Sales Agent just learned to hand you the phone

Lofty shipped live transfer for its AI Sales Agent in version 4.52, published in the Lofty Help Center. Here is what it does, in plain English. Your Sales Agent, the AI that calls and texts leads for you, can now make outbound calls on its own. When a lead on that call asks to speak with a person, Sales Agent connects them straight to you, live, while the call is still going. There is no silent hold. The AI keeps talking with the lead while it calls you, so the lead is never sitting in dead air wondering if they got hung up on. You get a quick rundown of the call right before you join, so you walk in knowing what was already said. And if you are unavailable, Sales Agent books a time with the lead and follows up with you. That is the whole feature, and it is the one that matters most in this update, because it fixes the exact moment where AI follow-up usually breaks: the handoff to a human.

The same update carries two more changes worth your time. First, Sales Agent notifications got more reliable. You still get texts when a lead requests an appointment, and now you also see notifications in the Lofty app and in Lofty CRM, with email as a backup if a text cannot be delivered. You can approve an appointment request in one tap, and it clears on every device. The notifications still respect your working hours. Second, you can now copy and paste Smart Plan actions. A copied action keeps all of its original settings, so you can build longer plans faster without re-entering the same details. If you build your follow-up in Smart Plans, and you should, that is real time back.

Two honest caveats before you get excited, and this is where my skepticism lives. Lofty says feature availability varies by the package and add-ons you purchased and by your account type, so check your own account before you promise yourself this works on your plan. And this is outbound AI calling, which is regulated. Before you switch it on, check your brokerage's AI policy and your state's rules on automated calls, including consent and TCPA restrictions, and think about fair housing guardrails before any automated message goes to a lead. Keep the human handoff switched on. The transfer is the guardrail. Without it, you are an AI cold-calling your leads with no human in the loop, and that is not a system, that is a liability. One more honest check: a booked appointment is not a closed deal. Measure signed clients and closed transactions, not how many calls the AI made. For what it is worth, the part of this update I actually like is behavior-triggered follow-up, the AI reaching out when a lead clicks a property, because that is a lead telling you what they care about. Outbound calling is the part I am still skeptical about, and that is the part to test first.

What does live transfer change for your follow-up in Lofty?

Live transfer turns the AI Sales Agent from a voicemail machine into an opener. It makes the first call, qualifies the lead, and hands you the live conversation the moment the lead wants a human, with a rundown of what was already said, so you step in warm instead of cold. That is the split I teach inside Apex Elite AI, and it has not changed: the agent prepares, you decide. The follow-up is the relationship, and the relationship is the part that closes. The AI opens. You close. I covered the same split in last Tuesday's edition of The Apex AI Brief, where the best AI agent in the world finished 12 percent of real professional jobs end to end. Live transfer is Lofty building the product around that exact truth. Use it for the opening. Keep the conversation.

Would you switch it on?

Not yet, and not without a test first. I am still not using the AI Sales Agent myself, even with live transfer, because a booked appointment is not a closed deal. If you test it, keep live transfer on, set your hours and transfer number, listen to a test call, and measure closed deals, not calls made. That is my closing question in the video, and it is the question I want your honest answer on. Would you let AI call your real estate leads? And if you were the customer, would you want an AI calling you? A test call answers both. Run one call to yourself, listen to how it sounds, check the rundown you get before you join, and see whether the lead experience feels like your business or like a robot reading a script. If it does not feel right, do not switch it on for real leads yet. The feature will still be there next week. Your reputation with a lead will not wait.

"The money is made in the follow-up, and automation is key to getting your time back. But automation with a human handoff is how you keep the deal and your license."

Story 2: "Don't ask again" just leaked 2,539 secrets onto GitHub

Security firm Apiiro published its research on November 5, 2026, and the finding is the kind that should change a habit the same day you read it. When a developer using Claude Code, Anthropic's AI coding assistant, picks "Yes, and don't ask again" on a permission prompt, Claude Code saves the exact approved command, word for word, as a standing allow rule in a project file called .claude/settings.local.json, according to Apiiro's report. If that command had a password, an API key, or a token typed into it, the secret now sits in plain text in the project folder. And that file is supposed to stay on one machine, but nothing in the repository itself stops it from being committed. Once it is pushed to a public GitHub repository, anyone can read it.

Apiiro searched public GitHub between November 1 and 3, 2026, and the scale is what makes this a story instead of a bug report. The firm found 104,185 public copies of that settings file across 103,617 repositories. In 1,122 of them there was at least one real secret, 1,379 distinct secrets in total. When Apiiro widened the search to older git history, the shared settings file, and saved agent transcripts, the count rose to 2,539 secrets across 1,994 repositories and 63 organizations. The biggest categories were 379 database passwords and 122 Supabase service role keys, plus 43 SSH logins for public servers, 42 GitHub, GitLab, and npm tokens, and 23 AWS keys. And here is the part that should worry you most: the standard secret scanners missed a lot of it. Of the 1,122 files with a real secret, gitleaks flagged 564 and TruffleHog flagged 643, and 215 files slipped past both tools. A green scan is not proof you are clean.

To be fair about it, Apiiro did not test whether the credentials still worked, and nobody has reported them being abused. One organization the firm notified did confirm its exposed key had been live. And Boris Cherny, who leads Claude Code at Anthropic, replied on Hacker News on November 7 that the team is looking into what it can do to prevent this. This is not a story about a hack. It is a story about a convenience button doing exactly what it says, forever, in a file nobody checks. And it lands in the same week the EU AI Office opened its first enforcement action under the AI Act, which I covered in Thursday's edition: the regulators are now asking the labs how their agents are contained and monitored. The permission you grant is becoming a legal question, not just a settings question.

What should agents do before they let an AI make calls or touch their accounts?

Before any AI agent touches your Lofty account, your email, or your phone, check three things: how it got access, whether you can take that access back, and whether anything client facing gets a human yes first. If the access came from a password you handed over, cut it and reconnect through the official integration instead. Here is the realtor translation of the Apiiro leak. You probably do not run Claude Code, but you do the same thing every time you paste a password into a tool, click "remember this," or let an agent log in as you. A password gives the agent every permission you have, and the log records everything it does under your name. Never put a password, a key, or a token into a command or a prompt an AI agent runs. Reference secrets as variables the agent never sees. If you want AI working inside Lofty, use the sanctioned connection, Lofty's own MCP server, which gives the agent scoped access with its own identity, not your password. And remember the guardrails: no Social Security numbers or full financial records in any AI tool, no client information in an unvetted tool, and no negotiating position to a third-party AI without the client's informed written consent. You are responsible for your clients' information. Know where the data goes, know whether you have permission to put it there, and protect it like your license depends on it, because it does.

Frequently asked questions

Will Lofty's AI Sales Agent call my leads without me knowing?

No. Sales Agent works the follow-up you set up, and the 4.52 update makes it easier to stay in the loop, not harder. You get texts, app notifications, and email backup when a lead requests an appointment, and you approve it in one tap. When a lead asks for a human on a call, the AI transfers them to you live and gives you a rundown first. You set the working hours it respects. The agent opens. You decide.

Is it safe to let an AI agent call my leads?

It can be, when the guardrails are set before the first call, and I am still testing it myself before I trust it with my own leads. Check your brokerage's AI policy and your state's rules on automated calls, keep the live human handoff switched on, and never let the AI send or promise anything on your behalf without a human yes. The risk is not the call. The risk is an agent with permissions nobody checked, calling leads nobody reviewed, with your name on it. Set the limits first, test it, then decide.

What is the biggest risk when giving an AI agent permissions?

The biggest risk is a permission you cannot take back. Apiiro's November 5, 2026 research showed Claude Code saving approved commands, secrets included, into a project file that ended up public on GitHub, exposing 2,539 real secrets across 1,994 repositories. The lesson applies to every agent, not just that one: never type a password, key, or token into a command an agent runs, approve narrow patterns instead of "don't ask again" on everything, and connect tools through official integrations with scoped access instead of handing over your login. If a tool holds your password, that access gets cut today.

Bottom line: Lofty's AI can now open the call and hand it to you live, and that live transfer is the only reason it is worth testing. I am still not using it myself, and I would not switch it on without a test call, your brokerage rules checked, and a plan to measure closed deals, not calls made. And 2,539 leaked secrets just showed what happens when you hand an agent a permission you cannot take back. Give AI the opening, keep the conversation, and guard the keys.

Adam Gillespie, Apex Elite AI. The money is made in the follow-up, and automation is key to getting your time back.