Medigram, Inc.

At the TTIC | Trustworthy Technology and Innovation Consortium, we don't just talk about the multidisciplinary trust required in #healthcareAI. We did the work required to earn it across the disciplines. 👩⚕️Medicine builds trust through clinical evidence, professional judgment, patient safety, and accountable practice. 🔬Research builds it through methodological rigor, peer review, publication, and evidence that withstands scrutiny. 👷♂️Engineering requires architecture, interoperability, reproducibility, testing, and systems that work as designed. 🤖Cybersecurity assumes failure and adversaries. Trust requires controls, testing, monitoring, behavioral verification, and evidence of what actually happened. ⚖️Law, privacy, compliance, and governance require authority, obligations, liability, evidence, and accountability to be established and defensible. 👩💼Executives and boards ultimately have to determine whether all of this can be responsibly governed, financed, operated, and defended in production. We have spent years actually doing this work across those boundaries. And we are the only organization we know willing to do all of it. That is not a claim we make lightly. There are extraordinary people and organizations in every one of these domains. The problem is structural. Most people own a piece. The work between disciplines is difficult, expensive not just in time and money, professionally at high personal cost, often invisible with no credit. So it gets left undone. And the cost of leaving it undone is eventual failure. Organizations can have excellent clinicians, engineers, cybersecurity, research, governance, and executives and still fail because nobody owns the dependencies between them. Each function can succeed according to its own definition of trust while the system fails as a whole. #AI makes this more consequential at rapid speed because it crosses these boundaries simultaneously, at speed and scale. A model can perform well and still be clinically unsafe. A system can be secure and poorly governed. Something can be compliant and fail operationally. Rigorous research does not automatically become a dependable production system. Mitch Parker and I co-founded the Trustworthy Technology & Innovation Consortium because someone had to be willing to take responsibility for the work between the disciplines. We were willing. So we did it. And many extraordinary people have done it with us. We know how much work it takes, and we understand the cost of not doing it. Trust is not a label we put on technology. It is the result of work performed across disciplines, over the lifecycle, with evidence. No single profession can do all of it. Someone has to integrate it. That's why TTIC exists. Read the origin story here: https://lnkd.in/gAs65UbY Thanks to many advisors + champions Edward Marx David Finn Charles Podesta Khalid Turk MBA, PMP, CHCIO, FCHIME Noel Gillespie Steve Wilson and more #AIGovernance #AIStandards #Cybersecurity

I read a joke about the prevailing job market somewhere, that the professional world has landed in a strange place. 👉People are looking for jobs, and there are no jobs to be found. 👉Companies are looking for people, and there are no suitable people to be found. 👉And then there are people already in jobs, offering no value whatsoever. 🤣 Funny, until you sit in a leadership seat and realize it isn't really a joke about the labor market. It's a joke about capability. Then this landed in my inbox from ever so respected leader Sherri Douville, and it named the thing exactly: "We need people who can say: 'We don't know. I'll investigate, build a prototype, learn what I need, find the right people, test it, document what works, and come back with evidence.'" Sherri elaborates: "That doesn't mean everyone needs to be a frontier architect. Once those people establish workable systems, many others can become excellent operators within them. But somebody has to cross the unknown first. The instructions don't exist yet." The institutional challenge isn't just 'How do we train everyone?' It's 'How do we identify the people capable of learning and building when nobody yet knows what the organization is supposed to become, give them authority to experiment, and then rapidly convert what they discover into systems everyone else can use? That is a very different leadership problem. Read that twice. Almost every hiring rubric, training program, and performance review we run is built for operators: people who execute a known process well. That is not a criticism. Operators carry organizations. But every manual we hand them was written by somebody who once stood in front of a blank page with no instructions at all. With AI, we are all standing in front of that blank page. The scarce skill is not knowing the answer. It is the willingness to say "I don't know" out loud, and then go find out with rigor, evidence, and receipts. Which makes the leadership question sharp, and uncomfortable: 👉Can you spot those people? They rarely have the most polished résumé. 👉Will you give them real authority to experiment, including permission to be wrong? 👉Can you convert what they learn into systems everyone else can operate, fast? And Sherri's closing point is the one most organizations skip entirely: we also need the people who can build the procedures and process that drive trust across engineering, security, medicine, research, finance, and law, all at the same time. Discovery without that is just an expensive experiment. Somebody has to cross the unknown first. Then somebody has to build the bridge behind them. Does your organization know who those people are? Or are you still hiring for a manual you no longer have? #LeadershipInTheAgeOfAI #WisdomAtWork

Check out this awesome new blog from Sherri Douville where she highlights how she is using both Praxen and Observra as part of the Medigram, Inc. strategy for highly secure medical AI systems. https://lnkd.in/g6Bgj7xT

Many thanks to Steve Wilson for founding and building this community and to Exabeam for supporting the Open Agent and AI Security Community and tools such as Praxen and Observra. Showing people the path is critical in raising morale right now when so few enterprises are getting real value out of production for #AI and #agents but feeling pressure from boards and investors for ROI. The work you're doing here accelerating paths to production for the people doing the real work is critical in the larger shift of this entry point of collecting telemetry for agents in the work of building operating models around observable evidence. The operating model is the point, not the model. Security, reliability, performance, and economics increasingly need to be evaluated together, particularly as agent activity scales. The hard part won't be generating more data. It will be determining which evidence matters, who is accountable for it, and what happens when it moves outside acceptable bounds. The Open Agent and AI Security Community should be helping lead this conversation together with other partners. Thanks to Stephen Moore's podcast and Meet Shah's question on Saturday's post who triggered this thought chain as well as Khalid Turk MBA, PMP, CHCIO, FCHIME cc Medigram, Inc. TTIC | Trustworthy Technology and Innovation Consortium

The new leader is not the person with the most information, meetings, or AI-generated output. In the #AI and #agentic era, information and output are becoming abundant. The scarce leadership capability is turning complexity into trusted outcomes. I think of the new leader as a catalyst: part architect, part coach, part point guard. They see the whole system, set direction, and design how people, AI, AI agents, tools, and organizations work together. They develop the team while maintaining enough selective depth to enter the play when their expertise, judgment, or authority can materially change the outcome. Most importantly, they close the loop. An idea is not an outcome. Neither is a strategy, presentation, AI model, policy, or piece of code. Leadership carries the work through idea → decision → ownership → execution → adoption → verification → outcome. The best leaders go further. They multiply capability. They build systems that allow people to do what previously seemed impossible, making extraordinary performance more repeatable, trustworthy, and scalable. Leadership in the AI era is increasingly not what you know. It is what you can responsibly cause to happen. See the system. Set the direction. Orchestrate the work. Enter the play when it matters. Close the loop. Multiply capability. Sports is a perfect metaphor because it’s the discipline and persistence that drives the growth and performance. Copying those who have helped shape the thinking Arthur Douville, Jr. MD Apurv Gupta, MD, MPH Steve Wilson Brian Yam Karen Jaw-Madson Khalid Turk MBA, PMP, CHCIO, FCHIME Anthony Lee Stephen Moore Medigram, Inc. TTIC | Trustworthy Technology and Innovation Consortium #AILeadership #AgenticAI #ArtificialIntelligence #Leadership #FutureOfWork #AIStrategy #AIGovernance #ResponsibleAI #SystemsThinking #AITransformation #ExecutiveLeadership #HumanAI #AITRUST #TTIC #Medigram

A huge thank you to the Exabeam team for doing something the #agentic ecosystem urgently needs: turning hard-won security knowledge into excellent open-source tools that agent builders and security teams can actually use. Praxen is the open-source reference implementation of Agent Behavior Verification. It compares what an AI agent is authorized and intended to do with evidence from its implementation and behavior. Observra, a second additional tool complements that work by capturing and normalizing agent runtime activity, including model calls, tool use, handoffs, cost, latency, and errors, so teams can observe what agents actually do once they're running. Together through the Open Agent and AI Security Community they help move agent security from assumed behavior toward verifiable, observable evidence. We were honored for our implementation at Medigram, Inc. to attain the highest Praxen behavioral score recorded to date, and even more grateful for the team's generosity in helping us reflect on what we learned from the experience. That conversation continued during the incredibly professionally run The New CISO Podcast, where the Exabeam team helped us organize our thinking around a much bigger question: what does the CEO, board, investor, and enterprise actually need from the CISO now that well governed, secure AI agents can execute work? The preparation for that conversation, expertly moderated by Stephen Moore and produced by Kelly Buckman inspired this new TTIC | Trustworthy Technology and Innovation Consortium blog, “The CISO Is the Defensive Captain Now, Not the Defensive Coach.” The premise is simple: as agents gain identity, permissions, tools, access, and the ability to execute consequential work, the security game changes. Configuration alone no longer predicts behavior. CISOs and their cross-functional partners increasingly have to verify what agents actually do, observe them in production, recognize abnormal behavior, intervene when controls fail, and own the security outcome. But you'll have to wait for the recording to hear our full take on why this game has changed for CISOs and their cross-functional teams, and what those of us actually driving agent fleets into production, with open-source security tooling such as Praxen and Observra are learning from the production surface. Thank you to Steve Wilson and the Exabeam team for sharing what you've learned, and for creating tools that give the rest of us something concrete to build, test, verify, and improve with. Read: The CISO Is the Defensive Captain Now, Not the Defensive Coach: https://lnkd.in/gydbhgnj Catch past episodes of the New CISO podcast here: https://lnkd.in/g4atqE4a #CISO #AgenticAI #AISecurity #Cybersecurity #AgentSecurity #AgenticSecurity #BehavioralVerification #AIObservability #Praxen #Observra #SecurityEngineering #TrustworthyAI

From this morning’s conversation at Medigram, Inc. Being an AI & Agentic Systems Engineer in 2026 is basically professional dog walking, except every dog can call an API, access a database, spend money, trigger a workflow, create a security incident, or wander into production. 😂 Knowing Python, SQL, RAG, LLMs, ML/DL, (machine learning, deep learning) cloud, and system design is increasingly just the beginning. Production #AI and #agentic systems also require requirements engineering, testing and evals, security, privacy, governance, observability, data quality, CI/CD, incident response, human review, change management, cost management, regulatory compliance, verification, clinical validation, and agent orchestration. And someone still has to answer the least glamorous but most important questions: Who owns this? What happens when it fails? Did it actually improve anything? Meanwhile, the CFO wants the budget controlled, the CISO wants the risk controlled, the CIO wants it integrated, the CMIO wants it clinically sound, General Counsel wants the liability understood, the COO wants it operational, the investor wants returns, and the Board would very much like to know what exactly everyone has unleashed. The model may increasingly be the easy part. Engineering a trustworthy system around it is the work and the profession. Adapted with appreciation from a Python developer group meme that made us laugh this morning because the dogs needed a few more leashes. #AgenticAI #AISystemsEngineering #AIEngineering #AIAgents #ResponsibleAI #AIGovernance #AITrust #HealthcareAI

I crossed 100 commits in GitHub again for July, bringing me to more than 200 commits across June and July. I'm celebrating the milestone, but not because I'm trying to become the most prolific engineer. I'm not. The best engineers operate at an entirely different level of output than what I'm doing. What this represents for me is selective depth, a leadership capability I increasingly believe will be mission critical in an AI and agentic economy. Leaders cannot become experts in everything, nor should we try. But we can identify the domains closest to the outcomes for which we're accountable and deliberately go deeper. For me, development is one example. Working directly with the technology changes the questions I ask. It changes what I can inspect and verify, how I assess difficulty, how I evaluate talent, and ultimately the quality of the decisions we make. This becomes even more important as #AI and #agentic operating models increase how much execution we can delegate. The more we delegate, the more important it becomes to know where we need depth ourselves. Selective depth will look different for every leader at different times. We will need to go deep in all these areas depending on what's happening. It might be software development, finance, clinical operations, cybersecurity, regulatory affairs, manufacturing, sales, or another domain central to the outcomes we own. The goal isn't to do everyone's job. It's to go deep enough in the right places to recognize quality, to enable, to challenge assumptions, verify what matters, and make better decisions. AI expands what leaders can delegate. Selective depth determines how well we can lead it. I’m not just talking about or writing standards and policy. I’m closing the loop on them: translating them into executable systems, implementing them in code, and verifying that they actually work as intended. An important nuance to preserve: selective depth doesn't require 100 commits/month. Our commits are one observable manifestation of it. 💸 A CFO might demonstrate selective depth by personally interrogating financial models and agent-generated forecasts. 👩⚕️A physician executive might go deep into clinical workflow and AI evaluation. 👾 CISO might personally examine traces, controls, attack paths, and evidence. So I would not turn “200 commits” into the standard. Your standard is: Go deep enough in selected consequential domains that you can independently recognize quality and verify what matters. That is the leadership capability we're illustrating. 🚀 Define it. Build it. Execute it. Verify it. #AIGovernance #ResponsibleAI #AITrust #AIStandards #AIVerification #AIInfrastructure #AgenticAI #HealthcareAI #AISafety #TrustworthyAI

One of the biggest changes in the economy is being mistaken for an AI productivity story. It isn’t. The economy used to have enough slack to support people who produced excellent work while others turned that work into outcomes. Researchers researched. Physicians practiced. Analysts analyzed. A whole layer of coordination made it work. AI and economic pressure are reducing that slack. A weak operator says, “Look how productive I am. I made 100 artifacts instead of 10.” A strong operator asks: What are they for? Who uses them? What decisions do they drive? What downstream work did I create? Who owns the outcome? The first sees output. The second sees the system required to absorb it. That distinction matters because AI can generate work faster than organizations can integrate, verify, and act on it, while many organizations are simultaneously reducing coordination capacity. So the economics of talent are shifting. We are moving from rewarding output inside a function to rewarding ownership of outcomes across functions. You can see it in conferences too. Grants, institutional budgets, sponsors, administrative support, and travel funding historically helped absorb the economics and coordination around participation. As that support tightens and, in many cases, collapses, the system becomes visible. Someone has to fund the room, attract attendees, create value for sponsors and participants, and make the whole thing work. The same is happening inside organizations. This doesn’t mean everyone becomes an administrator or hands critical work to AI. Particularly in regulated environments, AI and agents must be governed, secure, managed, maintained, monitored, and verified. It means thinking beyond your workstream. Before you ship something, ask: What outcome is this driving? What happens downstream? Who closes the loop? And when you use AI, don’t measure leverage by how much you produce. Measure it by how much outcome you can reliably own. The scarce valuable person isn’t the one who produces the most. It’s the one who makes the whole thing work. #ArtificialIntelligence #AIProductivity #AgenticAI #OutcomeOwnership #AILeadership #AIGovernance

Our last post was about why speakers don’t build conferences. Ecosystems do. It prompted a follow-up question: How will you earn your seat? Seats at important tables aren’t participation trophies. They’re earned through contribution. And what counts as contribution is changing fast. #AI is reducing the value of isolated work products and increasing the value of people who can integrate across boundaries and own outcomes. For a long time, many organizations trained people to operate in workstreams: do your part, produce the analysis, write the report, hand it off, check the box. In many expert-driven fields, the system was designed so that highly specialized people could focus narrowly while others handled everything around them. That model is shifting fast. The emerging expectation is: I don’t just complete work. I ensure the outcome happens. Well governed, secure, and maintained #AI and #agents will increasingly absorb parts of the “hand-off” work, but not the judgment, coordination, verification, or accountability that makes outcomes real. That’s still human work. And it’s getting more valuable and brutal at the same time. Which creates tension: the environment is now changing faster than most institutions can train for. So waiting for formal training is no longer a strategy. That leads to a simple shift in mindset: don’t define your job by your workstream. Define it by your outcome. Don’t outsource your development to your employer. Don’t confuse training completion with readiness. Because the person who throws work over the fence is competing with an agent. The person who owns the outcome, who connects the dots, who makes decisions, removes blockers, and closes the loop is not. That’s the real filter now. And it applies to everyone: executives, experts, partners, speakers, rising talent. Me. So what can you do tomorrow, Monday? 1. Redefine your job as an outcome. Ask: What is supposed to be different because I exist in this role? Then trace it end to end, not just your piece. 2. Close one gap. Find one capability your environment now needs that you were never formally trained for. Learn it. Apply it. Get feedback. Iterate. Don’t wait for permission or a course. 3. Remove one handoff. Find where work stalls between people or functions. Step into the gap. Clarify ownership. Drive it to completion. Close the loops, all of them. Do this consistently and something changes: you stop being known for what you produce. You become known for what you make happen. And that’s the real shift. Because conferences, promotions, and opportunities don’t reward participation. They reflect contribution. So instead of asking, “How do I get a seat?” ask: “What outcome will I own?” “How will I earn my seat?” And once you’re there: “What will be better because I was in the room?” That’s not just a conference question. It’s a career one. #HowToImplementAIinHospitals #HowToTransformHealthcareSystems #HowToOwnClinicalOutcomes