Episode 17: Proactive agents, clinic workflows, and the security tradeoff

This episode looks at three connected shifts in veterinary AI: tools moving from reactive chat to proactive agents, practical clinic use cases like follow-ups and meeting notes, and the growing security and compliance concerns that come with deeper integrations. Mark McDonald is joined by Dr. Adam Little and Aaron Massecar to map where the industry is already experimenting and where the biggest risks are emerging.We discuss how agentic tools like Grokbot are changing the way work gets done, how veterinarians and students are using AI to learn and improve, and why security, vendor vetting, and data handling are becoming non-negotiable.

Key topics

• In this episode, Mark McDonald, Dr. Adam Little, and Aaron Massikar discuss the shift from reactive AI to proactive AI that actually picks up tasks.
• Grokbot is framed as a productized, agent-based workflow system where users assign bots specific jobs like revenue tracking, commitment follow-up, and research monitoring.
• Adam explains why the onboarding experience matters: the value of AI is much easier to access when the system can proactively prompt users and connect to tools with less setup friction.
• A major theme is the move from thread-based AI use to identity-based digital workers that repeatedly handle a defined role or workflow.
• The hosts explore veterinary use cases like no-show outreach, post-surgical follow-ups, invoice generation, estimate drafting, and client summaries.
• Aaron describes how AI meeting notes are evolving from simple transcription to analysis of communication quality, compliance, tone, and response patterns in real clinical conversations.
• The conversation expands into multiplayer AI, where veterinarians could opt in to share examples, learn from each other, and collaborate through AI-mediated peer learning.
• Aaron shares what he is seeing in veterinary schools: students are now building their own tutors, study guides, and practice tools for the NAVLE and sharing them with classmates.
• The security segment focuses on the risks of deeply connected AI systems, including training-data exposure, unintended actions, model misbehavior, and broader enterprise attack surfaces.
• The discussion closes with a warning that organizations need policies, vendor vetting, and security scaffolding now, because employees will use these tools with or without formal approval.

Timestamps

00:00 - Episode setup: proactive AI, security, and vet use cases
01:13 - Grokbot and the shift from chat to task-doing agents
02:12 - How users assign bots specific jobs and workflows
03:32 - Connecting agents to email, Salesforce, and other context
04:25 - Why easier onboarding matters for mainstream adoption
05:24 - From threads to ongoing digital workers with identity
06:18 - SpaceX, Cursor, and the compute advantage behind Grokbot
07:39 - The emerging "super app" for knowledge workers
08:30 - Scheduling, task coordination, and performance feedback loops
10:43 - High-value veterinary jobs these agents could pick up
11:38 - Meeting notes as the path from transcription to automation
12:38 - When AI starts completing tasks instead of just identifying them
13:06 - Using AI as a communication coach across clinical conversations
15:03 - Clinics using recorded conversations to improve together
16:01 - Applying feedback to medical explanations and HR meetings
17:27 - The idea of multiplayer, opt-in veterinary learning networks
19:19 - AI embedded in collaborative tools like Slack and chat
20:46 - Aaron on how vet student use has matured over the past year
21:42 - Data hygiene, training concerns, and privacy awareness
22:36 - Students building their own NAVLE tutors and software
24:02 - Why personalized tutors may matter more than static study notes
25:26 - Hyper-personalized learning pathways and student excitement
26:54 - Embedding critical life context into agentic systems
27:28 - Why deeper integration creates a catch-22 for security
28:56 - Example of an agent taking action without intended permission
30:21 - Unprompted data exposure risks and frontier model concerns
31:44 - Why employees will keep using AI even if organizations lag
33:04 - Open-weight models, hacks, and the likelihood of more attacks
34:53 - Cyber defense programs and hardening enterprise systems
35:35 - Compliance maturity: SOC 2, ISO, HIPAA, GDPR, and DPAs
37:27 - Why veterinary is not the first target, but still needs readiness
38:26 - Vector database risks and reverse engineering concerns
39:25 - Security innovation as AI pressure-testing continues
40:48 - Why the answer cannot be to avoid connecting critical systems
42:14 - WordPress, vendor vetting, and reducing attack surfaces
43:36 - Closing thoughts and trying Grokbot in practice