Digital Social Hour – Episode #1768

“Alan from Tryolabs – AI Is Not Magic: Here’s What Real Deployment Looks Like”

Host: Sean Kelly
Guest: Alan (Tryolabs)
Date: January 19, 2026

Overview

In this engaging conversation recorded at the AI4 conference, Sean Kelly and Alan (CEO of Tryolabs) dive deep into the realities of deploying AI in real-world scenarios. Rather than centering on hype or sci-fi speculation, Alan shares concrete examples—from lion conservation and wildfire detection to sustainable fishing and enterprise solutions—illustrating both the power and the challenges of genuine AI implementation. They discuss distinguishing between meaningful AI innovation and smoke-and-mirrors demos, the importance of reliable data, and why the future of AI will still require a strong human touch.

Key Discussion Points & Insights

1. What is Tryolabs? (00:49)

• Tryolabs is an AI consulting and services company operating for 15 years, well before AI was mainstream ("AI was called machine learning back then").
• Originally served Silicon Valley startups, now focuses primarily on enterprise and large non-profit clients.
• Their key offering: building custom solutions to generate business or societal outcomes using clients’ large, often underutilized datasets.

“It’s not just a technical aspect… you need to make AI fit into those complex systems and actually provide business value.” (Alan, 01:29)

2. AI for Good: Real-World Impact Projects

• Satellite Imagery for UNICEF (02:02)
  Tryolabs created AI to find unmapped schools in developing countries using satellite imagery, helping UNICEF’s GIGA project connect these schools to the Internet.

  “In most developing countries, the schools are not 100% mapped by the government… UNICEF’s intention is to find these schools and go and connect them to the Internet.” (Alan, 02:41)

• Sustainable Fishing & Nature Conservancy (03:05)
  AI on fishing vessels classifies intended catch vs. bycatch in real time, helping address global overfishing and meet new retail regulations.

  “We’re shortening the cycles and making sure that this ship can report independent metrics in real time.” (Alan, 03:45)

• Lion Identification for Conservation (04:33)
  Collaborating with Lion Guardians, they use computer vision to non-invasively identify individual lions from photographs (using unique whisker patterns), tracking their movement across Africa.

  “We can actually use AI to match [a photo] to the existing database… this lion was actually found in another reservoir hundreds of miles away and now it’s moved here.” (Alan, 00:10 & 04:33)

3. AI & Surveillance: The Double-Edged Sword (06:06)

• Recognition of AI’s potential for both beneficial and dystopian uses (e.g., mass surveillance in China vs. “AI for good”).
• Discussion on “edge AI”—on-device computation for analytics in, e.g., retail—showing not all AI equates to invasive surveillance.

  “The fact that there’s a camera does not mean that you’re being surveilled. This can be something that runs on device and… just reports aggregate statistics.” (Alan, 06:29)

4. Wildfire Detection (07:17)

• Tryolabs helped a startup develop AI for early wildfire detection using a camera network, triangulating early smoke signals for rapid firefighting response.
• Challenge: distinguishing smoke from fog, BBQs, or cloud patterns, and managing false positives.

  “If your system generates too many false positives, then nobody will pay attention to that.” (Alan, 08:43)

5. The Toughest AI Models to Build (09:12)

• Wildfire detection stands out for its need for vast, diverse data and low tolerance for false alarms, due to many unpredictable edge cases (birds, snow, lens flare).
• Other tough projects: demand forecasting for retailers, where stakes and complexities are high.

6. The State of AI in Enterprise (11:42)

• Tryolabs builds industry-specific solutions, from airline fuel optimization (helping one carrier save $120M) to generative tools for developers, and manufacturing process improvements.
• The real challenge: most enterprises’ data infrastructure lags far behind the latest tech headlines.

“We started a project for a massive retailer…the way we got access to the dataset was a hard drive mailed to our office…” (Alan, 12:51)

7. Systemic Challenges and Advice to Aspiring AI Entrepreneurs (13:50)

• For newcomers: Don’t get discouraged by “imposter syndrome.” The landscape’s speed and hype can be overwhelming, but most flashy demos don’t scale to real solutions.

  “Don’t think you are the imposter… play with the tools, get things done, and build that mental model of how the future might look like and enjoy the ride.” (Alan, 14:19)

8. AI Agents Aren’t Magic – Reliability Is Key (15:14)

• Enterprise AI apps require ultra-high reliability—99% isn’t enough in most mission-critical settings.
• Silver bullet thinking is misleading; practical adoption is messy and gradual.

  “An agent that is 99% reliable on a business might be completely useless… Getting to the 99.9999999% you need…is really, really hard.” (Alan, 15:34)

9. Are We in an AI Bubble? What’s Next? (16:45)

• Alan doesn't see an internet- or crypto-style “bubble,” but rather a period of “realization” as companies move from hype to understanding AI’s real utility and limits.
• Human judgment and expertise will remain vital for years—even as advanced AI tools mature.

“The real world is super messy and doesn't run like that... the industries will take time to adapt.” (Alan, 16:51)

Notable Quotes & Memorable Moments

| Timestamp | Speaker | Quote | |-----------|---------|-------| | 01:29 | Alan | “You need to make AI fit in those complex systems and actually provide business value.” | | 02:41 | Alan | “In most developing countries, the schools are not 100% mapped by the government… UNICEF’s intention is to find these schools and go and connect them to the Internet.” | | 03:45 | Alan | “We’re shortening the cycles and making sure that this ship can report independent metrics in real time.” | | 06:29 | Alan | “The fact that there’s a camera does not mean that you’re being surveilled. This can be something that runs on device and… just reports aggregate statistics.” | | 08:43 | Alan | “If your system generates too many false positives, then nobody will pay attention to that.” | | 12:51 | Alan | “…the way we got access to the dataset was a hard drive mailed to our office…” | | 14:19 | Alan | “Don’t think you are the imposter… play with the tools, get things done, and build that mental model of how the future might look like and enjoy the ride.” | | 15:34 | Alan | “An agent that is 99% reliable on a business might be completely useless… Getting to the 99.9999999% you need…is really, really hard.” | | 16:51 | Alan | “The real world is super messy and doesn't run like that... the industries will take time to adapt.” |

Section Timestamps

• [00:49] – What is Tryolabs? & Alan’s background
• [02:02] – UNICEF: Mapping schools with satellite AI
• [03:05] – Sustainable fishing & real-time catch monitoring
• [04:33] – Identifying individual lions for conservation
• [06:06] – AI for surveillance: Good, bad, and the nuanced middle
• [07:17] – AI wildfire detection: Challenges and impact
• [09:12] – Toughest models to build
• [11:42] – AI deployments across various industries
• [13:50] – Data infrastructure and realistic AI in the enterprise
• [13:50-15:14] – Advice for newcomers; battling imposter syndrome
• [15:14-17:53] – Why “AI magic” won’t work at scale
• [16:45-18:01] – Bubbles, reality, and the future of AI

Conclusion

This episode is a reality check on AI: it’s not magic, but powerful tools can yield dramatic results—if organizations do the hard work (and get their data in order). Alan and Sean strip away the hype, offering a candid view of what works, what’s hard, and why the future belongs to persistent problem-solvers who embrace both technical and human complexity.

To learn more about Tryolabs and their work:

• Find them on LinkedIn (Tryolabs) or visit tryolabs.com