Today we can build software, analyze data, create videos, conduct research, design experiences. Make ideas real. All for $20 or $30 a month. Or free. I'm literally posting this using a PR & Marketing app I built on @Replit. So I'm going to keep building. It's a magical time.

The economics of AI are changing. Accenture is limiting Claude usage. I can hear it now: You call customer service. The bot answers. Tick. Ask it a question. Tick. Ask another. Tick. Finally get to a human and… POW. The real charges start. thestateofai.com/news/accent…

I'd prefer not to get so much love from @nassaupartners please. 7 "professionals" hit me this week using rented LinkedIn profiles to spam me about raising capital. As someone who designs AI workflows: your technology isn't broken, your interaction design is. True AI optimization requires designing the governance and deciding where the humanity belongs. Blasting empty, deceptive interactions at scale completely destroys trust. I can help you build actual human-centric workflows that respect user experience. (Hit my DMs!) But seriously, make it stop. #Founders #LinkedINSpam #ArtificialIntelligence #AIAutomation #ResponsibleAI #VentureCapital

“To thine own self be true.” My mother quoted Hamlet to me when I was young. Nearly 40 years after her passing, I feel those words more deeply than ever, as a founder, following the right path and being my authentic self.

"An idiot in motion is better than a genius at rest." (Often attributed to P.T. Barnum). A reminder to myself and others as we build: Momentum beats perfection every single time. Get moving.

I’ve been quietly building Caidir this year, to help my healthcare clients work through a question that keeps getting more important: Where should AI act, and where do people need to lead? That work has grown into Caidir, an AI decision platform for healthcare, now being refined through pilots before we open more broadly. Today, I wanted to share a little of its front door. Draw a line and a neighborhood comes to life: people walking dogs, visiting friends, caring for one another, and sometimes needing healthcare. That’s intentional. People have whole lives. The care we design, and the AI we introduce, need to fit into them. While we finish the pilots, I wanted to give you somewhere to explore, create, and get a feel for what Caidir stands for. Take a moment to draw your own journey, and leave your email to hear when we launch. caidir.com Embrace AI. Keep life at the center. #HealthcareAI #ExperienceDesign #Caidir @Replit

Excited to speak at the Applied AI Summit with @JohnSnowLabs. Identify where in your workflow to automate AI, or not. Register now: appliedaisummit.org

It's exhausting reading healthcare's endless stream of "the other group is the problem." Doctors blame administrators. Clinicians blame payers. Administrators blame regulation. Generations blame each other. And now we have groups talking about redesigning healthcare and AI while entire groups affected by those decisions aren't even in the conversation. Has healthcare accumulated too much administration and regulation? Absolutely. But do we really believe the answer is doctors without administrators? Nurses without physicians? Administrators without clinicians? AI without any of them? Some administration is bureaucracy. Some exists because we decided someone should be accountable for safety, quality, privacy, finances and whether people actually get good care. I've seen this dynamic before in M&A work. Bring two organizations together and they can spend enormous energy explaining why the other one is the problem. Introduce a third competitor and suddenly yesterday's rivals discover how much they have in common. Healthcare feels like it's entering that moment. There is real pain underneath all of this. Roles aren't working. Workflows aren't working. People are exhausted. And AI is about to change who does what yet again. But the next healthcare system won't be designed by one tribe winning. Maybe we need to stop asking who healthcare needs less of and start asking what healthcare needs each of us to become.

Minsky was right, and it's measurable. Serotonin transporter binding follows a seasonal cycle that shifts risk perception, satisfaction, and judgment. Same event, different neurochemistry, different decision. The evidence shows up clearly in healthcare. A physician or nurse delivers the same care to a patient in January and again in April. Yet the April patient reports lower satisfaction and has worse outcomes. Demoralizing for the clinician, and the patient is genuinely experiencing it differently. Their biology is working against contentment and changes decision making. And it's not just healthcare. I've built this into my AI governance platform (built on @replit) and wrote about it here: joankelly.group/april-effect…

In addition, hospitals don't even know all the services being rendered. They see pieces of the system: transactions, encounters, surveys, complaints, claims. The patient experiences the whole journey. And inherently, hospitals don't really know the totality of that patient experience. Neither do health insurers. You can't redesign what you can't see.

When the Littles come to visit & discover your label maker, poster board & markers! 🤣

Patients are clear they want connection with their Doctors & Nurses. Their humanity. Yet they need navigation in the process of care. That's the AI goldmine. Not replacing the human connection.

Hospital executive bonuses are often tied to how patients rate the hospital. The #1 driver, on a question CMS only started asking in 2025: did the care team work well together. Not one great clinician. Everyone knowing what everyone did.

Last night my niece's fire alarm went off at bedtime. Her husband was traveling, so her mother and I ran over. There was no fire, thankfully. But the alarm kept going off unexpectedly. My 4-year-old great niece was in my arms crying and was able to tell me exactly what was frightening her: the noise, yes, but also not knowing when it was going to happen again. It made me think about something I've spent much of my career working on: No Surprises. Not the good kind, like a surprise trip to Disney. The uncertainty that comes from not knowing what's happening, what comes next, or what you're supposed to do. I knew that feeling as a child. My father had MS and was a quadriplegic. I'd walk home from school holding my breath as I came down the hill, not knowing whether he'd be home or whether I'd find the front door wide open because an ambulance had taken him away. Years later at Humana, we heard a version of this again while talking with seniors and redesigning the Medicare Advantage experience. Different circumstances. Same human need. Help me understand what's happening. Tell me what comes next. Don't make me guess. The challenge was that everyone working on one piece of the experience didn't necessarily understand how their piece affected the member's whole experience. And now I'm seeing the same pattern with AI. We're installing AI solutions without always designing the experience around them. Does the person understand what just happened? Do they know AI was involved when they need to? Do they know what happens next? Does the next person in the workflow know what happened before? And do we even know whether the experience worked? Patients don't experience our departments, technologies or organizational charts. They experience one journey. The organization owns that journey. As we redesign healthcare with AI, we should design the workflow around one of the most basic human needs: Reduce unnecessary uncertainty. No surprises.

Hospitals already knew. Press Ganey’s instrument carried a version of the teamwork question for years. Vendor research had already identified it as among the strongest drivers of likelihood to recommend. What changed in 2025 was not discovery. It was standardization, public reporting, and payment. If it took nineteen years to nationally measure something the field already knew, the odds that anyone promptly measures what AI does to team coordination are not good. The lag is the governance problem.

Agree with 5. The human-comparison benchmark is saturating and we need better measures. Here's one that just became available. CMS added a teamwork item to HCAHPS in 2025. First national data: whether the care team worked well together is the #1 predictor of overall hospital rating. All five tasks in the paper are bounded by the encounter and measure information flowing in, eliciting history, differential, test selection. Nothing measures information flowing out: what the patient understood, who owns follow-up. In Medicare episode data, communication measures predict ~nothing during the index admission and the entire effect in the 30 days after. The paper concedes the mechanism: "transmission of information between the LLM and the user is a particular point of failure," and notes vignette studies can't detect it.

In 2025, CMS began asking patients whether their care team worked well together through HCAHPS. On the first national data it ranked as the number one predictor of overall hospital rating. Health systems are deploying AI into team coordination faster than anyone is measuring what it does there. Those two facts are on a collision course.

Such a creative and engaging way to showcase your work. Congratulations!!

When you need a little magic in your life watch Field of Dreams. Some movies entertain you. Some remind you to believe in things you can't quite explain.

What if you could test a healthcare improvement before implementing it? AI is making scenario planning much more practical. Working on reducing readmissions, I can now model a patient scenario, change interventions and assumptions, and see how different decisions could affect the outcome. Instead of: design → implement → wait for the data. We can begin to: model → test → compare → then design. Built this Readmission Scenario Planner with @ReplitDesign and using it for a client with their own data. This changes how we can approach healthcare improvement. @MannyBernabe @raymmar @Franciscocrz