Artificial Insights: Conversations About AI

Regulation and innovation rarely move at the same speed. In this bonus clip, Daniel Manary and David Proulx talk about who should shape the future of AI. David argues that the people building modern systems understand both the risks and the capabilities far better than policymakers who lack technical grounding.

He contrasts Europe’s tendency to regulate early with the United States’ focus on innovation, liability, and builder responsibility. He points to the “Winning the AI Race” framework as an example of policy written by people who clearly understand the field, from guardrails to energy use to how a system should shut itself down if needed. For David, the real danger for countries like Canada and those in Europe is asphyxiating innovation before fully understanding the technology.

🎧 Want the bigger picture?

In the main episode, David explains how HoloMD built a clinical agent that has handled more than 100,000 mental-health conversations without hallucinating. He walks through long-term memory, safety testing with synthetic patients, and why incentives matter in healthcare AI.

Listen to the full episode: https://rss.com/podcasts/manaryhaus/2365531/

🔗 Resources & Links

• 🤝 Connect with David on LinkedIn: https://www.linkedin.com/in/davidproulxoui/
• 🌐 Learn more about HoloMD: https://holomd.ai/
• 📩 Subscribe to the Artificial Insights newsletter for key takeaways: https://manary.haus/podcast/#haus
• 👉 Have a guest in mind? Reach out to Daniel at [email protected]

🚀 Know someone thinking about AI regulation or policy? Share this clip with them.

One of the things David Proulx is most proud of is building an AI agent that has handled more than 100,000 mental health conversations without a single hallucination.

David is the Chief AI Officer at HoloMD, where an AI agent becomes a third party in the Therapeutic Alliance. The patient wants to feel better. The psychiatrist guides their care. HoloMD adds a daily point of contact that checks in, gathers context, tracks medication adherence, and offers support between visits. Psychiatrists stay fully in charge while gaining a clearer picture of a patient's life than a single appointment can provide.

In this episode, Daniel talks with David about his path from building a social network for moms with 600,000 downloads to locking himself in a room for nine months learning AI, then co-founding HoloMD, which won the 2025 Mental Health Tech Innovator of the Year Award from the GHP Mental Health Awards. They explore what it takes to engineer safe clinical agents, how long-term memory was solved in their system, and why guardrails matter more than engagement tricks.

🔑 What You’ll Learn in This Episode

• ✅ How HoloMD uses daily check-ins, mood scoring, and medication adherence tracking to strengthen the Therapeutic Alliance
• ✅ Why prompt, flow, and context engineering shape safety in high-stakes AI systems
• ✅ How HoloMD validated safety with synthetic patients, human-on-the-loop review, and a rigorously tested escalation pathway
• ✅ Why David sees over-regulation as a risk for countries that want to stay competitive in AI
• ✅ How real patients have used HoloMD to self-report to the ER and get timely care

🔗 Resources & Links

• 🌐 Learn more about HoloMD: https://holomd.ai/
• 🤝 Connect with David on LinkedIn: https://www.linkedin.com/in/davidproulxoui/
• 📩 Subscribe to the Artificial Insights newsletter for key takeaways: https://manary.haus/podcast/#haus
• 👉 Have a guest in mind? Reach out to Daniel at [email protected]

💬 Know someone building AI in a high-stakes domain like healthcare or public safety? Share this episode with them.

Most AI conversations describe personalization as if it were new. In this bonus clip, Alex Maier explains why the real shift is not personalization itself, but how agentic systems read context and respond to a person's actual state the moment they ask a question.

Alex describes how onWater uses prompts, sentiment, language patterns, and user actions to understand experience level and intent. He compares it to the way great salespeople have always read a person's posture, tone, and energy to meet them where they are. The technology is new. The underlying posture toward users is not.

🎧 Want the bigger picture?

This clip expands on one of the central ideas of the full episode: what it takes to design AI native systems that guide people safely, respectfully, and intelligently through complex real-world decisions. In the main conversation, Alex walks through how onWater stitches together conditions, regulations, maps, and community insight, and how his team builds guardrails for activities where bad information carries real risk.

Listen to the full episode here: https://rss.com/podcasts/manaryhaus/2328700/

🔗 Resources & Links

• 🤝 Connect with Alex on LinkedIn: https://www.linkedin.com/in/alexmaier2
• 🌊 Explore onWater Fish and Paddle: https://www.onwaterapp.com
• 📩 Subscribe to the Artificial Insights newsletter for key takeaways: https://manary.haus/podcast/#haus
• 👉 Have a guest in mind? Reach out to Daniel at [email protected]

🚀 Share this clip with someone thinking about AI, context, and what meeting users "where they are" actually requires.

Most companies stop at “ask the bot a question”. onWater wants to go further, building an app that can answer “Should I go today?” and then quietly handle everything that comes after yes.

In this episode of Artificial Insights, Daniel talks with Alex Maier, now President of onWater. His team has been working on a simple idea with complicated stakes: help people decide where to fish or paddle, understand the real conditions on the water, and act on that information without forcing users to dig through dozens of screens or data sources.

Alex explains how they approached AI native design, how they stitched fragmented environmental data into a single system, and what it took to add features that can safely guide beginners while still giving experts real depth. He also shares how onWater is using computer vision to turn fish photos into useful science and why personalization in an AI age still comes back to knowing what a person is actually trying to do.

🔑 What You’ll Learn in This Episode

• ✅ How Alex distinguishes AI assisted from AI native, and why that shift changes product design
• ✅ Why clean, structured water data creates a defensible moat when everyone has the same models
• ✅ How onWater estimates fish length and weight from photos and feeds that data into TroutSpotter
• ✅ Where safety guardrails matter when AI outputs could influence on-water decisions
• ✅ Why curiosity and skepticism both matter when experimenting with new AI tools

🔗 Resources & Links

• 🤝 Connect with Alex on LinkedIn: https://www.linkedin.com/in/alexmaier2
• 🌊 Explore onWater Fish and Paddle: https://www.onwaterapp.com
• 🐟 Learn about TroutSpotter community science: https://www.onwaterapp.com/features/troutspotter
• 🧬 Wild Me, the conservation AI nonprofit behind Wildbook and TroutSpotter: https://www.wildme.org
• 📩 Subscribe to the Artificial Insights newsletter for key takeaways: https://manary.haus/podcast/#haus
• 👉 Have a guest in mind? Reach out to Daniel at [email protected]

💬 Know someone building AI in high-risk, real-world environments? Share this episode with them so they can hear how onWater is approaching safety, data, and product design in the wild.

AI is currently in a transitional moment: a new tool being used to make old kinds of content. But, what if AI can do more than just help us create the same old things faster and better?

In this bonus clip, Daniel Manary and Ahmad Iqbal step back from product and design to explore how AI is a new creative medium that could change how people make, share, and experience ideas.

Drawing on the work of Marshall McLuhan, the Canadian media theorist who coined "the medium is the message", Ahmad explains how each technological leap (from the radio, to the television, to the internet) first copied old forms before finding its own.

The next creative leap will come when people start using AI to invent new formats altogether.

💡 Want to hear the rest?

This clip comes from Ahmad Iqbal’s interview with Daniel, where they discuss accessibility as strategy, AI adoption inside Canva, and how product design lowers the skill floor for creativity.

Listen to the full episode: https://rss.com/podcasts/manaryhaus/2309358/

🔗 Resources & Links

• 📖 Read about Marshall McLuhan: https://en.wikipedia.org/wiki/Marshall_McLuhan
• 📩 Subscribe to the Artificial Insights newsletter for key takeaways: https://manary.haus/podcast/#haus
• 👉 Have a guest in mind? Reach out to Daniel at [email protected]

💬 Thinking about what AI means for creativity? Share this episode with someone to spark a conversation!

What if "designing well" no longer requires years of training, and the real work becomes choosing the story to tell?

In this episode of Artificial Insights, host Daniel Manary interviews Ahmad Iqbal, product lead and Head of MENAP at Canva, about what it really means to design for accessibility in the age of AI. Ahmad helped launch Canva’s first generative AI features and shares how the company's guiding principle, making creativity accessible to everyone, drives its product, culture, and internal AI adoption.

The conversation traces how a 48-hour hackathon project became Magic Write, how Ahmad approached AI as a non-technical product manager, and why the best ideas at Canva don’t ship if they’re not simple enough for anyone to understand.

The episode closes with a wider reflection on how AI changes not just what we design, but who gets to create. Drawing on Marshall McLuhan’s writing, Ahmad describes a future where AI lowers the skill floor, turning consumers into producers and expanding who can participate in design and storytelling.

🔑 What You’ll Learn in This Episode

• ✅ Why "accessibility" is Canva’s north star for AI features
• ✅ How a 48-hour hackathon prototype became a core product surface
• ✅ What "truly local" means for regional expansion and product fit
• ✅ How to encourage teams to use AI internally to unlock time for harder problems
• ✅ Why McLuhan’s "the medium is the message" matters for AI-era design

🔗 Resources & Links

• 🌐 Explore Canva: https://www.canva.com
• 📖 Read about Marshall McLuhan: https://en.wikipedia.org/wiki/Marshall_McLuhan
• 🎧 Check out the Student Specials from S3: https://rss.com/podcasts/manaryhaus/2185718/
• 📩 Subscribe to the Artificial Insights newsletter for key takeaways: https://manary.haus/podcast/#haus
• 👉 Have a guest in mind? Reach out to Daniel at [email protected]

💬 Building AI into real products or localizing at scale? Share this episode with a PM or regional lead who will benefit from Ahmad’s playbook.

Most people talk about "data privacy" like it’s a legal checkbox. For Josh Müller, it’s part of faithful stewardship.

In this bonus episode, Josh and Daniel pick up on a single idea from their main conversation: data management as Kingdom stewardship. Josh shares how his once “off-the-cloud-at-all-costs” mindset softened as AI made it possible to do genuinely good things with data.

The hard part now isn’t the tech. It’s building the habits and structures that keep data private, local, and used for good.

💡 Want the full conversation?

This clip comes from Josh Müller’s conversation with Daniel, where they dive into AI-powered translation, offline models, and how Waha uses automation to multiply impact across 43 languages.

Listen to the full episode: https://rss.com/podcasts/manaryhaus/2282863/

🔗 Resources & Links

• 🤝 Connect with Josh on LinkedIn: https://www.linkedin.com/in/thejoshmuller/
• 🌐 Explore Waha: https://waha.app/
• 🌐 Josh’s site: https://joshmuller.ca/
• 📩 Subscribe to the Artificial Insights newsletter for key takeaways: https://manary.haus/podcast/#haus
• 👉 Have a guest in mind? Reach out to Daniel at [email protected]

💬 Enjoyed this one? Share it with a builder wrestling with data stewardship and privacy.

What if a six-month production bottleneck became a 16-minute job?

Josh Müller, director at tech nonprofit Waha, tells Daniel how forced alignment and smart automation turned a tedious audio-editing gauntlet into a repeatable pipeline that ships Bible study content across dozens of languages. The result is more time for actual disciple-making and less time in Audacity.

They dig into privacy-first analytics in risky contexts, when to choose local models over cloud tools, and why data management is Kingdom stewardship.

🔑 What You’ll Learn in This Episode

• ✅ How timestamping with forced alignment collapsed months of manual audio work into minutes
• ✅ Practical ways nonprofits can use agents and local LLMs without leaking sensitive data
• ✅ A privacy posture that starts with “we can’t leak what we don’t have”
• ✅ When to reach for tools like Whisper for offline transcription, and agentic coders for data chores
• ✅ Why disciplined data stewardship unlocks future AI use while protecting people at risk

🔗 Resources & Links

• 🤝 Connect with Josh on LinkedIn: https://www.linkedin.com/in/thejoshmuller/
• 🌐 Explore Waha: https://waha.app/
• 🌐 Josh’s site: https://joshmuller.ca/
• 📚 Meta’s Massively Multilingual Speech (forced alignment): https://ai.meta.com/blog/multilingual-model-speech-recognition/
• 💻 OpenAI Whisper (speech-to-text): https://github.com/openai/whisper
• 📩 Subscribe to the Artificial Insights newsletter for key takeaways: https://manary.haus/podcast/#haus
• 👉 Have a guest in mind? Reach out to Daniel at [email protected]

💬 Enjoyed this one? Share it with a builder wrestling with data stewardship and real-world multilingual workflows.

When is the right time to start a company in AI? Adeel Zaman argues it comes down to breakthroughs.

In this bonus clip, Adeel explains why "why now" matters more than ever in AI. Without a recent step forward, most ideas have already been tried and blocked by the same old limits. He points to Jeff Bezos in 1995, who spotted the internet’s growth curve and built Amazon around books, an application that fit the technology’s strengths at the time.

🎧 Curious where Adeel takes this next?

In the full conversation, he and Daniel talk about building embodied intelligence, why voice feedback may be the missing ingredient in reinforcement learning, and how personalized AI could change both privacy and performance.

Listen to the full episode: https://rss.com/podcasts/manaryhaus/2238950

🔗 Resources & Links

• 🤝 Connect with Adeel on LinkedIn: https://www.linkedin.com/in/adeelzam/
• 📩 Subscribe to the Artificial Insights newsletter for key takeaways: https://manary.haus/podcast/#haus
• 👉 Have a guest in mind? Reach out to Daniel at [email protected]

💡 If this sparked a thought, share it with a founder or builder who’s weighing their own "why now" moment.

What happens when AI doesn’t just live in text and screens, but begins to reason and act in the physical world?

Adeel Zaman, CTO and Co-Founder of DOZR, has spent his career moving from deep learning research to startups, scaling companies and tackling cold-start problems with machine learning. Now, backed by the HF0 residency, his focus is on "Embodied Intelligence" and how foundation models can learn physical tasks, adapt through feedback, and interact with humans in real time.

In this conversation, Daniel and Adeel explore why embodied AGI may be a prerequisite for true general intelligence, how voice interaction could change human-machine collaboration, and what it means to give individuals, not just big labs, the ability to teach and shape their own AI models.

🔑 What You'll Learn in This Episode

• ✅ Why embodied intelligence could be a prerequisite for reliable long-horizon agents and true "AGI"
• ✅ How real-time "reasoning out loud" can change human-robot collaboration on job sites
• ✅ Why reinforcement learning from language feedback is key when rewards aren't cleanly verifiable
• ✅ The case for individually owned AIs vs. globally shared weights

🔗 Resources & Links

• 📄 Google DeepMind’s RT-2 research overview: https://robotics-transformer2.github.io
• 🤝 Connect with Adeel on LinkedIn: https://www.linkedin.com/in/adeelzam/
• 📩 Subscribe to the Artificial Insights newsletter for key takeaways: https://manary.haus/podcast/#haus
• 👉 Have a guest in mind? Reach out to Daniel at [email protected]

💬 Inspired by the idea of embodied AI? Share this episode with someone else who would be interested in the conversation!