The Tech Trek

AI agents are easy to demo. They are much harder to trust, maintain, govern, and put into production.

In this episode of The Tech Trek, Amir Bormand talks with Lucas Thelosen, CEO and cofounder at Gravity, about the agent economy, AI analytics, and what changes when analysts move from doing every task themselves to managing AI systems that create more bandwidth.

Lucas shares why Gravity built Orion, an AI analyst, after years working in analytics, product, and data teams at companies like Looker and Google. The conversation gets into the messy middle of AI adoption, why so many agent projects struggle to make it into production, and how context may become one of the most valuable assets a company owns.

Practical Takeaways

• Agent prototypes are easy. Production agents require support, maintenance, accuracy checks, and clear ownership.
• Not every company should build every agent internally. If the capability is not core to what you sell, buying may be the faster path.
• Context matters because it lets humans critique AI output with business judgment, not just technical review.
• Analysts may shift toward data architecture, governed data models, and internal product management for analytics.
• AI does not remove human responsibility. It raises the bar for review, delegation, and decision making.

Timestamped Highlights

00:40, What Gravity is building with Orion, an AI analyst designed around the work analytics teams already know well.
02:28, Why mature companies still miss major insights, even when they already have data teams.
03:43, The agent economy reality check, easy prototypes, hard production, and the gap between demo and durable system.

06:28, Why companies still build agents internally, even when many projects never reach production.

09:48, The case for experimenting now instead of waiting for the AI stack to settle.

11:40, How AI shifts people from doing the work to managing the work.

18:50, What the future analyst role may look like as AI takes on more of the execution layer.

One Line That Stuck

"Where previously you were the person doing the work, now you're the manager."

Subscribe to The Tech Trek for more conversations on how technical teams are building, hiring, operating, and adapting around AI, data, platform, product, and engineering execution.

Yahoo is not just adding AI on top of existing products. It is using AI across product experiences, internal tools, engineering workflows, and modernization efforts.

In this episode of The Tech Trek, Lee Zen, CTO at Yahoo, joins Amir Bormand to talk about modernizing at massive scale, moving from on prem infrastructure to the cloud, rebuilding internal tools with AI, and how engineering organizations need to rethink process when agents can move faster than people.

Lee also shares how Yahoo views AI as a coworker, not just a tool, and why the next bottleneck in software delivery may be human judgment.

Practical Takeaways

• Modernization at scale often means operating in two worlds at once, keeping proven systems running while new cloud based services move faster.
• AI can help teams move past legacy tools by reverse engineering requirements and rebuilding modern versions from scratch.
• The real unlock is not only code generation. It is connecting agents to documents, chats, emails, production context, and internal knowledge with the right permissions.
• As agents speed up execution, engineering teams need to rethink where human approval, judgment, and review should live.
• The build versus buy equation is changing because some tools that were too expensive to build before may now be realistic to create internally.

Timestamped Highlights

00:31, Yahoo’s mission and why the internet still feels hard to navigate
02:01, Where AI fits across Yahoo products and engineering work
03:30, The challenge of moving from on prem data centers to cloud based infrastructure
05:27, How Yahoo has used AI to rebuild internal tools and leave technical debt behind
07:25, Why agents need access to engineering context, not just code
10:20, AI as a coworker and the shift from human speed to machine speed
16:27, Why parts of the SDLC may need to change as AI increases delivery speed

One Line That Stuck

“AI as a coworker, not just as a tool.”

The Tech Trek is for technical leaders thinking through how teams build, operate, modernize, and adapt as AI changes the work. Subscribe or follow for more conversations with engineering, product, data, and technology leaders.

Mike Choi wanted to work at Apple for years. Then he got there and had the moment many ambitious builders eventually hit.

Is this the thing I was sprinting toward?

In this episode of The Tech Trek, Mike Choi, co founder at Koah, shares his path from Korea to the United States, mandatory military service, Apple, Twitter, and eventually building Koah, an AI monetization company helping AI app builders create sponsored experiences.

The conversation is less about the glamour of startups and more about what founder work actually demands: making decisions without complete information, learning from Big Tech without copying it, and staying focused when AI moves faster than your team can absorb.

Practical Takeaways

• Big Tech can teach you strong operating patterns, but startups force you to build your own style.
• Founder decisions rarely come with complete data. Moving creates the next data point.
• In AI startups, speed can become a distraction if every new tool or feature changes the plan.
• Clear vision helps teams make decisions without waiting on the founder.
• Knowing when to share an idea matters as much as having the idea.

Timestamped Highlights

00:38, Mike explains Koah and why AI products need new monetization models.
02:25, Mike shares how his father’s Korean Air Force service brought him to the United States as a child.
05:01, Mandatory military service, pausing college, and learning to code around strong engineers.
07:29, The long term goal of working at Apple and the unexpected feeling after getting there.
10:57, Why Mike chose to build from scratch instead of staying on the Big Tech path.
14:05, What Big Tech did and did not prepare him for as a founder.
17:03, The founder lesson of making decisions before the full picture is clear.
19:35, Why AI startups move so fast and how shiny object syndrome drains energy, time, and attention.

One Line That Stuck

“Just make the decision, produce data points that way through actions, and make a better decision tomorrow.”

Subscribe to The Tech Trek for more conversations on how modern technical teams are building, hiring, operating, and adapting around AI, data, platform, product, and engineering execution.

AI is moving fast, but the bigger question for companies and governments may be control. Who owns the data, the workflow, the output, and the risk?

In this episode, Amir talks with Shaun Modi, cofounder and CEO of Capitol AI, about sovereign AI, shadow AI, model dependency, government use cases, and why organizations need repeatable, governed, auditable workflows before AI becomes part of core operations.

Shaun also brings a design lens to the conversation, connecting AI adoption to user experience, voice interfaces, and the next wave of AI in the physical world.

Practical Takeaways

• Sovereign AI means having control over your data, outcomes, upside, and risk.
• Shadow AI creates short term productivity, but can also create silos, governance gaps, and data exposure.
• Organizations need repeatable, governed, auditable AI workflows, especially in regulated environments.
• Model independence matters because model costs, performance, and capabilities keep changing.
• Design will come back into focus as AI systems become more powerful and more embedded in work.

Timestamped Highlights

00:40, What Capitol AI does and why decision ready artifacts matter
01:50, Shaun defines sovereign AI in plain language
02:36, The intelligence paradox, more data, less control
04:15, Why shadow AI can become a governance and accuracy problem
10:29, Zero data retention, model independence, and evaluation criteria
17:53, Why AI user experience may be entering a new design cycle
25:56, Where AI may create major impact in the physical world

One Line That Stuck

“Companies and governments have more data than ever, but they are losing control over the outcomes.”

Practical Takeaways For Teams

If AI is moving into real business processes, start by asking what needs to be controlled. Data rights, model choice, accuracy standards, workflow governance, and auditability all matter more once AI is producing work that affects customers, citizens, or critical operations.

Follow The Tech Trek for more conversations on how technical teams are building, operating, and adapting around AI, data, product, and engineering execution.

Most healthcare AI stories start with diagnosis. Edmund Jackson thinks that misses the real bottleneck.

In this episode of The Tech Trek, Edmund Jackson, CEO and founder of Unity AI, joins Amir to talk about AI for healthcare operations. The conversation gets into why scheduling, staffing, follow up, payer coordination, and interoperability are often where healthcare breaks down, and why solving those operational problems may matter more than chasing the flashiest use cases.

Edmund brings a healthcare first view to AI. His argument is simple: healthcare is not slow because people are ignoring technology. It is slow because the real workflows are complex, regulated, high context, and hard to capture cleanly in software.

What You’ll Take Away

• Why healthcare experience matters when choosing which AI problems are actually worth solving
• Why diagnosis is not always the best starting point for healthcare AI
• How scheduling becomes much more complex when patients, payers, clinics, staff, protocols, and follow up all have to line up
• Why AI can help clinics save time while moving human staff toward higher value patient interactions
• Why interoperability is still hard, even with standards like FHIR gaining momentum

Timestamped Highlights

00:29, What Unity AI does and why healthcare operations is the focus
01:11, Why healthcare AI needs people who understand the domain, not just the technology
02:26, The danger of solving the hardest or flashiest problem instead of the most pragmatic one
05:15, Why AI may finally help healthcare handle personalization and operational complexity at scale
07:47, Why scheduling a healthcare visit is nothing like scheduling a delivery or restaurant order
10:42, How operational AI can save time and reduce downstream chaos in clinics
24:33, Why healthcare data is much harder to structure than financial data

One Line That Stuck

“Software is like children. Making it is all fun and games. Maintaining it is a whole other question.”

Practical Takeaways

• Start with the workflow that actually blocks progress, not the one that sounds most impressive
• In healthcare, operational context is often the product
• AI should create more room for humans to handle the interactions that require judgment, care, and clinical responsibility
• More software is not always the answer, especially in regulated environments where maintenance, compliance, and security matter

Subscribe to The Tech Trek for more conversations with founders, operators, and technical leaders building through AI, data, product, platform, and engineering execution.