Artificial Insights: Conversations About AI

This is Artificial Insights, where we talk to leaders and thinkers in AI about what it means to do AI right. Welcome to Season 3.

This season is a special “summer break” for the month of August. We’re moving from our usual focus on AI products to explore AI ethics. We’ll be examining the questions of purpose, trust, and responsibility that shape, whether consciously or not, every decision about AI.

AI brings up big, existential questions like no other technology ever has. We’re not just wondering whether we’ll find our next job, we’re wondering whether there will be jobs at all.

Join us as we explore these ideas and go beyond theory with real builders and leaders in AI!

🔗 Resources & Links

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

💬 Excited for the new season? Share this trailer with a colleague who wants to separate AI’s noise from its real impact.

AI is everywhere but not every implementation works.

In this Season 2 recap of Artificial Insights, Daniel revisits the most powerful moments from conversations with eleven leaders building AI in the real world. These guests have seen what happens when AI is rushed, misused, or built without purpose, as well as what it takes to create AI that lasts.

Across three themes, their voices reveal patterns worth paying attention to:

1️⃣ The risks of rushing into AI and what happens when pressure overrides purpose

• 🎧 Jennifer Moss, author of "Why are we here?", speaking on the false pressures executives are feeling to adopt AI before they have a plan. Full episode here.
• 🎧 Bijan Vaez from Merchkit, speaking on how feeding AI really bad data turned out to be the bigger problem. Full episode here.
• 🎧 Carlos Almeida from Optave, speaking on the pitfalls of the fast prototype and how you shouldn't depend on an AI solution built in a weekend. Full episode here.
• 🎧 Jonathan Green of Serve No Master, speaking on why you shouldn't just fire your entire customer support team and replace them with AI to save money. Full episode here.

2️⃣ How to build AI on solid ground to solve real problems and earn trust

• 🎧 Patrick Belliveau from GambitCo, speaking on how not solving for cancer, but instead a very specific problem, was the key for success. Full episode here.
• 🎧 Alex Millar from GovAI, speaking on how everything is, technically, a wrapper. The challenge is to find how you add value. Full episode here.
• 🎧 Mike Kirkup from Arlo, speaking on how easy it is to get to 50% with AI models, but the hard part is getting to 99%. Full episode here.
• 🎧 Tair Asim from Sync, speaking on how good inputs are non-negotiable when making RAG-based AI products. Full episode here.

3️⃣ Why rethinking work defines the future and why transparency matters

• 🎧 Atif Khan of MessagePoint, speaking on how becoming AI-first requires a major shift in thinking. Full episode here.
• 🎧 Chyngyz Dzhumanazarov of Kodif, speaking on how the future of customer service will probably be AI-first, with the human touch as a premium service. Full episode here.
• 🎧 Nicolas Tobis of Relias, speaking on how, with AI, we can now train and measure empathy. Full episode here.

This episode closes Season 2 with a big insight:

The companies that thrive aren’t asking, What can this model do?

They’re asking, What do we need to understand about our data, our customers, and ourselves to make this work?

AI is reshaping work in ways that are more complex than the headlines suggest. Nicolas Tobis has seen both sides: some roles shrink, others expand, and entirely new ones appear. At one education company, AI cut the content team by 80% while output soared to 35 times its previous level.

In this bonus episode, Daniel and Nico dig into what these shifts mean for workers and companies alike. They talk about when efficiency gains help teams grow, when they lead to cuts, and how roles in healthcare education are moving from creating content to validating it.

It’s an honest look at where jobs are being transformed and what skills will matter most next.

🎧 Want the bigger picture?

This clip is just a glimpse of Nico’s perspective. In the full episode, he shares how AI is being used to train empathy in healthcare, what natural language interfaces could mean for complex SaaS systems, and why the best AI tools fit seamlessly into the work people already do.

Listen to the full episode: https://rss.com/podcasts/manaryhaus/2143757/

🔗 Resources & Links

• 🤝 Connect with Nico on LinkedIn: https://www.linkedin.com/in/nicolas-tobis/
• 📓 Learn more about Relias: https://www.relias.com/
• 📩 Subscribe to the Artificial Insights newsletter for summaries and takeaways: https://manary.haus/podcast/#haus
• 👉 Have a guest in mind? Reach out to Daniel at [email protected]

🚀 Learn something new? Leave a review and share it with someone building in AI!

AI is now being used to teach and measure something most people think of as deeply human: empathy.

In healthcare, where conversations can be as critical as treatments, Nicolas Tobis, VP of Data and AI at Relias, and his team have been building AI training simulations for nurses and doctors. These tools let clinicians practice difficult conversations with AI-powered virtual patients and get feedback on how empathetic they were.

It’s not just roleplay. The AI reacts naturally, evaluates empathy, and even gives feedback. As one nurse said, “I really wanted to hold the AI’s hand.”

In this episode, Daniel and Nico talk about how AI is changing education, what a natural language interface could mean for complex SaaS systems, and even what happens when job candidates use AI avatars in interviews. They explore why the best AI tools feel like they were part of the workflow all along.

🔑 What You’ll Learn in This Episode

• ✅ How empathy can be measured and why it matters
• ✅ How AI simulations can transform healthcare training
• ✅ What a natural language interface means for the future of SaaS
• ✅ How AI helps companies do more, not just faster
• ✅ How to start building AI tools without being overwhelmed

🔗 Resources & Links

• 🤝 Connect with Nico on LinkedIn: https://www.linkedin.com/in/nicolas-tobis/
• 📓 Learn more about Relias: https://www.relias.com/
• 📩 Subscribe to the Artificial Insights newsletter for summaries and takeaways: https://manary.haus/podcast/#haus
• 👉 Have a guest in mind? Reach out to Daniel at [email protected]

🚀 Learn something new? Leave a review and share it with someone building in AI!

Some people try to engineer perfect prompts. Others just talk and get better results.

Jonathan Green has seen how a conversational, almost playful approach to AI changes the results you get. The people he calls master prompters name their AI, treat it like a collaborator, and, often without realizing it, unlock responses others struggle to reach.

In this bonus episode, Jonathan shares why these habits are hard to teach but powerful to watch in action, and why the best results often come from those who think they’re “not very good” at AI.

🎧 Curious what else Jonathan has learned?

This clip is just a glimpse of Jonathan’s perspective. In the full episode, he talks about designing AI agents with personalities and skill trees, why most companies don’t actually need more AI, and how “friendship-style prompting” changes outcomes.

Listen to the full episode: https://rss.com/podcasts/manaryhaus/2120044/

🔗 Resources & Links

• 🤝 Connect with Jonathan on LinkedIn: https://www.linkedin.com/in/servenomaster/
• 📓 Learn more about Jonathan: https://servenomaster.com/
• 📩 Subscribe to the Artificial Insights newsletter for summaries and takeaways: https://manary.haus/podcast/#haus
• 👉 Have a guest in mind? Reach out to Daniel at [email protected]

🚀 Learn something new? Leave a review and share it with someone building in AI!

“The best AI users are always women, and they’re always super high creative types.”

Jonathan Green has seen what many AI builders overlook: the most effective AI users aren’t technical experts: they’re creative communicators who treat AI like a teammate. They name their agents, talk to them conversationally, and rarely get frustrated when something doesn’t work.

After being fired from his university job in a snowstorm, Jonathan built a business around real-world AI use. From a tropical island, he now consults on AI automation and business workflows, helping teams solve problems without overengineering.

In this episode, Daniel and Jonathan talk about why most companies don’t need more AI, what customer support teams reveal that bots can’t, and how AI agents with personalities, skill trees, and logic chains are more effective.

🔑 What You’ll Learn in This Episode

• ✅ Why the best AI users may not look like who you expect
• ✅ How personality and skill trees make agents more effective
• ✅ Why customer support teams should never be replaced outright by AI
• ✅ How “friendship-style prompting” outperforms command-and-control approaches
• ✅ What most companies really need when they ask for AI

Resources & Links

• 🔗 Connect with Jonathan on LinkedIn: https://www.linkedin.com/in/servenomaster/
• 📖 Learn more about Jonathan: https://servenomaster.com/
• 📩 Subscribe to the Artificial Insights newsletter for summaries and takeaways: https://manary.haus/podcast/#haus
• 👉 Have a guest in mind? Reach out to Daniel at [email protected]

🚀 Learn something new? Leave a review and share it with someone building in AI!

Investors often ask Carlos Almeida the same question: if your AI is built on someone else’s model, what’s your moat?

For him, the answer starts with building what the model can’t: controls that keep AI aligned with a company’s goals, interfaces that let humans guide its decisions, and ways to manage and expand its usable data.

In this conversation, Daniel and Carlos explore the risks of relying on a model’s default behavior, the challenge of separating fact from fabrication, and why adding trust and control might be just as valuable as improving accuracy.

They dig into the less visible work required to turn a general AI into something a business can rely on.

🎧 Want the bigger picture?

This clip is just a glimpse of Carlos’s perspective. In the full episode, he shares how copilots can outperform chatbots, the traps companies face when building their own support AI, and why automation might expand demand for customer support instead of replacing it.

Listen to the full episode: https://rss.com/podcasts/manaryhaus/2082407/

🔗 Resources & Links

• 🤝 Connect with Carlos: https://www.linkedin.com/in/carlos-alm/
• 🤖 Learn more about Optave: https://optave.com/
• 📩 Subscribe to the Artificial Insights newsletter for summaries and takeaways: https://manary.haus/podcast/#haus
• 👉 Have a guest in mind? Reach out to Daniel at [email protected]

🚀 Learn something new? Leave a review and share it with someone building trust into AI systems!

AI customer support can improve service without replacing people, and Carlos Almeida believes that starting with automation is the wrong approach.

At Optave, the company he co-founded, the focus is on AI copilots that help enterprise customer support agents work more effectively, with automation only when the conditions are right. That means mapping every step of a support journey, building AI quality controls so responses meet standards, and giving companies the ability to guide what the AI is doing as part of a broader enterprise automation strategy.

Carlos has worked in customer support long enough to see the gap between marketing promises and what AI can actually deliver: from overhyped automation to quick prototypes that create more problems than they solve.

In this episode, he talks with Daniel about the traps companies face when building their own customer experience AI, why control and trust are as important as accuracy, and how automation might expand demand for customer support instead of replacing it.

🔑 What You’ll Learn in This Episode

• ✅ Why AI copilots can be a better first step than full automation
• ✅ How to build AI quality controls that keep responses on track
• ✅ Why companies overestimate what AI can do in support
• ✅ The hidden costs of quick AI prototypes
• ✅ How automation could expand demand for customer support rather than replace it

🔗 Resources & Links

• 🤝 Connect with Carlos: https://www.linkedin.com/in/carlos-alm/
• 🤖 Learn more about Optave: https://optave.com/
• 📩 Subscribe to the newsletter: https://manary.haus/podcast/#haus
• 👉 Have a guest in mind? Reach out to Daniel at [email protected]

🚀 Enjoyed this conversation? Pass it along to someone building AI systems their customers can trust.

AI hype cycles come and go, but the difference with this AI wave is how widely it touches every corner of business and work.

Mike Kirkup has seen big shifts before, from the rise of smartphones at BlackBerry to the explosion of web apps, but he says AI feels different because “everybody’s working on it,” even if not everyone realises it yet. That ubiquity creates both urgency and overwhelm: whole functions like customer success may need to be rethought, while companies like Klarna are already reversing course on automation-first strategies.

In this bonus clip, Daniel and Mike talk about why AI’s spread feels like a social experiment playing out in real time, what makes this moment unlike mobile or Web 2.0, and why sometimes the best sales script might start with: “Hi, I’m a human.”

🎧 Curious for more?

This clip is just a glimpse of Mike’s perspective. In the full episode, he shares how Arlo spun out of EnPowered, what it takes to build AI into lending workflows, and why domain expertise is as important as the AI itself.

Listen to the full episode: https://rss.com/podcasts/manaryhaus/2061508/

🔗 Resources & Links

• 🤝 Connect with Mike on LinkedIn: https://www.linkedin.com/in/mikekirkup/
• ✨ Learn more about Arlo: https://usearlo.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]

💬 Found this clip useful? Share it with someone rethinking where AI fits in their business.

AI is changing the way lenders and brokers process deals, and it’s not by replacing people, but by rethinking how the work gets done.

Mike Kirkup didn’t set out to build a product for the finance industry. As CTO at EnPowered, he spent too much time on the repetitive work of gathering and verifying loan documents. Solving that problem led to building what would become Arlo, an AI-powered loan processing platform for commercial lending and document automation.

Today, Arlo is used by lenders and brokers across North America to streamline loan document collection and verification. Instead of asking borrowers to upload documents one at a time, it lets them “just dump all of the documents.” The AI then identifies, extracts, and verifies what’s needed, improving speed and accuracy in loan processing workflows.

In this episode, Daniel and Mike discuss designing AI in finance with domain expertise, what happens when AI starts asking its own questions, and how building for AI from the start changes the borrower experience. They also explore the balance between giving AI more room to “play” and making sure it doesn’t ask a “stupid question” in front of a customer.

🔑 What You’ll Learn in This Episode

• ✅ How Arlo started as an internal tool and why partners asked to use it
• ✅ How AI changes the borrower experience in commercial loan processing
• ✅ Why domain expertise matters as much as the AI itself
• ✅ How to design for AI from day one instead of bolting it on later
• ✅ The trade-offs between control, unpredictability, and trust in AI

🔗 Resources & Links

• 💼 Connect with Mike on LinkedIn: https://www.linkedin.com/in/mikekirkup
• 🏦 Learn more about Arlo: https://usearlo.com
• 📩 Subscribe to the Artificial Insights newsletter for summaries and takeaways: https://manary.haus/podcast/#haus
• 👉 Have a guest in mind? Reach out to Daniel at [email protected]

⭐ Found this useful? Leave a review and share it with someone building in AI!