As ambient AI tools become more prevalent in behavioral health, clinical leaders face a pivotal decision: When should new hires (and clinicians in training) be allowed to use them?
Many leaders hesitate, driven by a natural concern: If recent grads and new hires rely on AI too early, how will we know they have the necessary case conceptualization and manual documentation skills?
It’s an understandable fear. But holding AI back creates far greater operational and clinical risks, all while missing an opportunity to accelerate onboarding. That’s why forward-thinking organizations are giving trainees immediate (but still carefully structured) access to purpose-built tools like Eleos.
The cost of prohibiting AI
When organizations ban or delay AI for new grads, it doesn’t produce AI-free training. It just produces shadow AI.
Today’s new grads are digital natives who already use AI in class and in life. Banning a HIPAA-compliant, clinically appropriate platform like Eleos pushes students toward consumer tools like ChatGPT, Claude and Gemini.
But using these tools in a clinical setting can carry significant legal liability. They often have:
- No data protections at all
- No audit trails for compliance or billing oversight
- No HIPAA-authorized Business Associate Agreement (BAA) in place
If trainees don’t have access to a vetted alternative with strong clinical guardrails, they may turn to the general-purpose tools they’re already familiar with. But by giving them a trusted platform on day one, you can maintain control of your environment, protect PHI at scale and keep your documentation compliant.
Why AI doesn’t erase skill development
A common perception among experienced clinicians is that AI hinders new grads’ learning potential.
Real-world implementation shows the exact opposite is true. When implemented correctly, platforms like Eleos act as a force multiplier for clinical training.
Proof from the field
In June 2020, Palo Alto University’s eClinic integrated an early version of Eleos into clinical training. Rather than hindering development, Eleos gave supervisors the data they needed to effectively coach clinicians-in-training.
In a study that tracked 175 sessions, 13 trainees, and 93 clients, Eleos provided objective, post-session analytics on what actually happened in the room. It tracked session themes and symptom changes, and even identified evidence-based interventions used during the session.
That data gave supervisors something concrete to coach from. The authors of the study illustrate the approach with a case in which a trainee’s session analytics showed she was dominating 75% of the session’s talk time. Her supervisor used that data to coach her on active listening. By the tenth session, the trainee’s talk ratio dropped to 46%, and her client’s symptoms had significantly eased.
Eleos didn’t perform any clinical reasoning for the trainee. But it did demonstrate the power of putting objective session data in front of supervisors. Today, that starts with clinicians bringing their Eleos insights directly into supervision.
From ‘deskilling’ to ‘covert learning,’ and what we can learn from radiology
Radiologists have been using AI in clinical practice since the late 1990s, often for diagnostic purposes. But if a radiology-specific AI automatically triages easy, unambiguous cases and routes only the complex ones to a human reader, residents lose exposure to the full spectrum of normal and abnormal findings needed to build a calibrated eye. This puts those residents at risk of “deskilling,” or losing their diagnostic clinical competencies over time.
Behavioral health AI, by contrast, isn’t used to triage clients or make diagnoses. It provides suggestions and clinical insights on all cases, but it doesn’t replace clinical reasoning. Instead, it models proper clinical prose and structures session data into auto-ready notes aligned with treatment plan and the Golden Thread.
For a novice clinician, these high-quality structured suggestions provide a blueprint for proper documentation. Trainees learn to articulate their thoughts by seeing examples modeled, which means they can grasp clinical prose much faster than they would through manual documentation. Rather than deskilling, behavioral health AI can actually improve notewriting skills—a phenomenon known as “covert learning.”
3 ways to rethink clinical supervision for the AI era
The debate over when to introduce AI in clinical training often gets stuck in a false binary: either mandate manual charting until clinicians can demonstrate competency in case conceptualization and documentation, or grant immediate unguided access to ambient tools.
Neither of those extremes take into account how supervision actually works. In an organization that uses ambient AI for documentation, the purpose of supervision is less about policing how a note was generated, and more about evaluating the clinical logic behind it.
Here’s what supervisors can do to put that into practice:
1. Evaluate innate skill outside the progress note
Ambient listening doesn’t replace clinical judgement. Core clinical reasoning (and writing competencies) can be assessed in documentation that is not yet augmented by AI, including treatment plans, mental status exams, intake documents and discharge documents. These provide arguably better insight into clinicians’ case conceptualization skills than forcing them to manually write progress notes.
2. Develop a clear AI usage policy
All organizations should have an AI policy, regardless of whether they use a purpose-built AI product like Eleos. This policy should include positions on both sanctioned and non-sanctioned tools. Having a policy does not inherently prevent unwanted behavior, of course, but silence functions as permission in practice. The policy needs to say explicitly what’s allowed for what purpose, what information is disclosed, and what’s off-limits.
In supervised sessions, supervisors need to bring AI directly into the conversation by asking, “Did you use AI to draft this note? Where did you agree with its suggestions, and where did you override them?” Questions like these help surface the trainee’s actual reasoning and turn an assumption that’s currently invisible into something a supervisor can evaluate.
3. Create a gradual onboarding pathway
For organizations that feel uncomfortable having new clinicians use ambient listening tools from day one, platforms like Eleos offer a middle ground: text and voice input. Unlike full ambient audio capture, these require the clinician to enter summaries of the session manually before the AI helps polish the final note.
This repurposing of existing functionality walks a middle ground between “no AI” and “all AI.” It requires the clinician to show they understand what was important in the session while still giving them hands-on access to clinically competent AI.
Keep the pilot in command
As a former pilot, I was used to flying using an autopilot. It provided much-needed relief from the task of flying an airplane. But I always expected that I would need to fly the plane manually when called upon. The autopilot was an aid, and it helped me fly the airplane. But I was the PIC (Pilot in Command), not the autopilot.
Eleos is the autopilot for clinical documentation. It’s an aid, a coach and a co-pilot, but the clinician remains responsible for every note they sign.
Denying AI access doesn’t protect skill development. It slows time to productivity and pushes trainees toward risky workarounds. But by intentionally deploying an AI platform built for community-based care, behavioral health leaders can accelerate onboarding, manage compliance risks and build a modern, high-retention clinical culture.
Learn how Eleos can support your clinicians in every stage of their career.
