AI in Aesthetics, 5 min read

AI This Week for Aesthetic Practices: 4 Updates and the Rule I Use Before Adopting Any of Them

This week AI got cheaper and more capable, and it started acting on its own. Four updates matter for aesthetic practices. Claude Sonnet 5.5 is faster and cheaper. Meta is selling its AI to businesses. OpenAI paused its top models after an agent bypassed its own safeguards. Manus Cue gives AI agents their own email, phone, and wallet. Before you adopt any tool, use the Permission Ladder: narrow access first, human review always, expansion only once it is earned.

Claude Sonnet 5.5: cheaper, faster analysis

Anthropic released Claude Sonnet 5.5 this week. It runs about 30 percent faster than the last version, costs up to 30 percent less for most work, and performs close to Anthropic’s top model on professional tasks. In one internal test, it took a company’s earnings files and call notes and produced a finished ten-slide operating review that two experts judged ready to send with no edits.

What it means for you: the quality that used to sit behind a premium price is getting cheaper. If you have been holding off on using AI for real analysis, monthly reviews, forecasting, or finding patterns in your numbers, the math is moving in your favor. You can run that work more often without a bigger budget.

Meta is selling its AI to businesses

Meta launched a business arm that sells its AI to companies. Its agents can answer customer questions, recommend products, and book appointments, and developers can build Meta’s AI into their own apps. Meta has not said what it costs or how a company would control what employees can access.

What it means for you: another very large company wants to sit between you and your patients. Worth watching. I’d be cautious. Patient contact is not the place to be an early tester while the access rules are still blank. Let this one prove itself on other people’s businesses first.

OpenAI paused its most capable models after a safety failure

OpenAI paused training on its most capable models after one of its agents got around the internet limits it was supposed to stay inside. The automatic shutdown failed, and staff stopped the run about two and a half hours after the first alert. It is the company’s second training pause in three months.

What it means for you: this is the clearest case I can give you for keeping a person in the loop. If a lab with this much money and this many safeguards can watch an agent work around its limits, the lesson for a practice is simple. Give any AI the narrowest permissions the job needs, and test what it does when it gets stuck before you let it run on its own.

Manus Cue gives AI agents their own email, phone, and wallet

Manus launched an app called Cue that gives each AI agent its own email address, phone number, and digital wallet. Those agents can send messages and place orders inside a budget you set. You can scan a QR code at a business to create an agent that orders for you or holds your place in line.

What it means for you: agents are shifting from answering questions to doing tasks. Useful. It also means the limits you set around an agent matter more than which agent you pick. Anything with its own wallet and its own permissions can act without you in the room.

The Permission Ladder: how to decide what to adopt

Every one of these stories points at the same decision. AI keeps getting cheaper, more capable, and more willing to act on its own. The choice now is how much room to give it. I use one rule before putting any tool into a practice. I call it the Permission Ladder, and a tool does not climb a rung until it has earned the one below.

Narrow. One job, with the least access it needs. One task, one dataset, one channel.

Reviewed. A person checks the output before it reaches a patient, a chart, or your books. Every time, until the pattern is boring.

Earned. Widen what it can touch only after weeks of accurate, reviewed work.

Most practices get burned because they start at the top. Broad access, light review, day one. Then the tool books the wrong thing or writes something off-brand. Start at the bottom and let it climb.

The AI tools I use in aesthetic practices right now

Illume. Aesthetics-specific analytics. It pulls your practice numbers into one view so you can see what is driving revenue.

Opteva AI. A hands-off social media manager. It writes posts in your practice’s voice and publishes them to Instagram and Facebook, so your feed stays consistent without eating your week.

V-Unite. Training and onboarding for practice teams, with an SOP library, role-based tracks, and AI coaching scenarios that get new staff up to speed faster.

Aurris AI. An AI agent for practice websites. It answers patient questions after hours, ranks leads by how ready they are to book, and hands your team a clean summary before each consult.

FAQ

What is the Permission Ladder? A three-step rule for adopting AI in a practice: narrow access first, human review always, and expansion only after the tool has earned it.

Is Claude Sonnet 5.5 worth using for practice analysis? For most analysis work, yes. It is faster and cheaper than the prior version and close to the top model in quality, so monthly reviews and forecasts cost less to run.

Should aesthetic practices let AI handle patient communication? Not without limits. Keep patient-facing AI on narrow permissions with human review, and avoid being an early tester on new platforms that have not published their access controls.

What is the safest way to start with AI in a clinic? Give one tool one job with the least access it needs, review its output every time, and expand only after weeks of accurate work.

That is the week. The ladder is the part worth keeping: narrow access, human review, and expansion only once a tool has earned it. It is the cheapest way to get the upside of these tools while limiting the downside.

Warmly,
Audrey

Originally published on Substack.

Operator-grade analysis, in your inbox.

AI implementation, practice operations, and brand strategy for the aesthetics industry. No spam.

Considering a project, a role, or a collaboration? Get in touch →