The Tech Trek

Agentic AI is not just a model problem. It is exposing gaps in how teams store, share, retrieve, and coordinate context across applications, agents, and people.

In this episode, Amir talks with Karthik Ranganathan, cofounder and co CEO at Yugabyte, about why databases are under new pressure as AI moves from model serving into agentic workflows. They discuss Yugabyte’s evolution, the limits of today’s data infrastructure, and why memory, knowledge, and shared context may become central to how agentic systems actually work.

Practical takeaways

• Agentic workloads push databases beyond simple relational access because agents may need relational, vector, graph, NoSQL, scale, and multi tenant support in the same workflow.
• A query can be optimized inside each data store and still be slow, expensive, or wasteful when the work spans multiple systems.
• Context sounds simple to humans, but it becomes messy when it includes private memory, shared project knowledge, conversation history, team collaboration, and agent actions.
• Human handoffs can erase much of the speed promised by agents when teams have to copy outputs, re explain reasoning, and manually reconcile conflicts.
• Yugabyte is working on Meko as a data infrastructure layer for agents, with a focus on memory, knowledge, context quality, and shared learnings.

Timestamped highlights

00:43
What Yugabyte does and why critical data needs to survive infrastructure change
04:08
How databases evolved from mainframes to internet apps, mobile, cloud native systems, and now AI
09:42
Why agentic workloads create new demands across relational, vector, graph, NoSQL, and multi tenant data
12:29
Why the current agentic data stack is still in the messy middle
15:24
Why context becomes hard when agents, people, teams, and permissions collide
21:50
How agent collaboration can fall back to human speed
24:34
How eeko aims to capture memory, knowledge, learnings, and reasoning across agent workflows

One Line That Stuck

“We have killed the velocity of agentic development and brought it back to human speed.”

Practical signals for teams building with agents

• Do not treat context as one generic blob.
• Decide what should stay private, what should be shared, and what should become reusable project knowledge.
• Watch for hidden cost when agents query across separate systems.
• Pay attention to agent collaboration, not just single agent output.
• Build for memory and knowledge flow before team size makes the gaps harder to fix.

Follow The Tech Trek for more conversations with technical leaders building modern teams, products, and infrastructure around AI, data, and engineering execution.

Agentic coding is not just making engineers faster. It is changing how teams triage bugs, prototype features, involve product, and think about hiring.

Scott Weller, CTO and founder at EnFi, joins The Tech Trek to talk about how his team is building around agentic software development while operating in financial services, where trust, accuracy, and human judgment still matter. EnFi uses AI agents to work through complex financial data rooms, extract knowledge, and support faster analysis in commercial lending.

In this episode, Scott breaks down how EnFi moved from simple coding assistance to a broader development harness, why Slack became a central interface for agents, how product and business leaders can now participate earlier in feature creation, and why engineering interviews need to change when AI is part of the actual job.

Practical Takeaways

• Start with specific productivity goals before trying to rebuild the whole development process.
• Agentic tools work better when they connect to the team’s real workflow, shared context, and software lifecycle data.
• Faster code generation changes the cost model, but it also creates new problems around review, testing, prioritization, and decision fatigue.
• Product, sales, and executive teams may be able to prototype ideas faster, but engineering still has to make the work production ready.
• Hiring needs to test how people solve problems with AI, not whether they can perform the old interview format without help.

Timestamped Highlights

00:38, What EnFi is building around financial data, AI agents, and commercial lending
02:13, Why software teams may need to forget part of their old development process
04:45, How EnFi started with productivity gains before building a broader development harness
09:53, Why merge requests went up, and why that alone is not the same as better outcomes
10:30, How Slack became the entry point for an agentic development harness
14:10, What happens to agile ceremonies when teams can create discovery builds much faster
25:08, Scott’s view on whether AI reduces engineering headcount or changes the work engineers do
31:00, How EnFi is changing technical interviews for an AI assisted engineering environment

One Line That Stuck

“We do not care if you use AI to solve the problems, we just want to know you can solve the problem.”

Practical Takeaways For Technical Teams

Put agents close to where work already happens.

Keep humans in the loop for review, testing, and production judgment.

Treat AI generated code as cheaper to create, not free to maintain.

Build stronger test harnesses instead of slowing everything down with excessive process.

Update interviews to reflect how engineering work is actually getting done.

Subscribe to The Tech Trek for more conversations with technical leaders building, hiring, and operating through the next stage of AI, data, product, and engineering execution.

AI adoption looks very different when mistakes can create legal, financial, and reputational risk.

Vijay Gandra, Global CDO at Acrisure, joins The Tech Trek to talk about AI transformation inside a regulated industry, where explainability, data quality, governance, cost, and team readiness matter just as much as model capability.

The conversation covers the trust gap in AI, how data teams are shifting from dashboard production to conversational data access, when to buy versus build, and why AI proof of concepts need to be judged by business value, operational efficiency, and customer impact.

Practical Takeaways

• Regulated industries cannot treat AI as a black box. Decisions need traceability, consistency, and often a human review layer.
• Data quality has to be addressed from the start. AI can amplify bad data as easily as it can create value.
• Data teams are moving beyond dashboard factories toward conversational data access and generative interfaces.
• Most companies can likely use existing AI tools for many needs, but sensitive IP and core business logic may require internal capabilities.
• AI cost will become a bigger production question as companies move from experimentation to scaled deployment.

Timestamped Highlights

00:47, Acrisure’s shift from insurance brokerage toward fintech and financial tools.
01:44, Why regulated industries face a trust gap with AI and need explainable decisions.
04:41, How data teams are evolving from dashboards to conversational data enablement.
08:28, The build versus buy question and where internal AI tools may still make sense.
10:52, Why AI experimentation can get expensive before companies know what works.
16:15, How to evaluate AI proof of concepts based on customer value, efficiency, and business impact.
18:14, Why data governance and data quality need to be treated as day one requirements.

One Line That Stuck

“In an industry like this, a 5 percent deviation is not just a simple glitch. It is actually a legal liability.”

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

Leonid Belkind, co founder and CTO at Torq, joins The Tech Trek to talk about what changes when an engineering organization does more than experiment with AI tools. Torq builds agentic security operations, and Leonid shares how his team is using AI across engineering, product, hiring, customer success, and go to market work.

This conversation gets past the shallow version of “AI makes coding faster.” Leonid makes a clear distinction between coding and software engineering, and explains why the best teams are using AI to shift cognitive load, not remove judgment.

Practical takeaways

• AI does not erase software engineering. It changes where engineering judgment shows up.
• Strong engineers still produce better AI generated work because they know what to ask, what to test, and what tradeoffs matter.
• Hiring processes need to reflect how engineers actually work now, including how they use AI to build, explain, and defend technical decisions.
• Productivity should not only be measured by speed. Leonid talks about throughput, maturity of delivery, and whether teams can produce more without lowering quality.
• AI adoption becomes more powerful when it moves beyond engineering into product, customer success, revenue operations, and talent.

Key moments

00:32
What Torq means by agentic security operations and why different tasks need different AI approaches.
01:49
Why building AI native products with AI native methods creates a useful feedback loop for engineering teams.
05:28
How AI shifts cognitive load so engineers can spend more attention on user experience, architecture, and product value.
10:34
The difference between software engineering and coding, and why that distinction matters more now.
15:13
How Torq has changed technical interviews to evaluate AI assisted engineering instead of pretending AI does not exist.
21:51
How one R&D group measured meaningful delivery gains after adopting AI more deeply.
24:25
Why AI adoption is moving into product, customer success, revenue operations, and talent teams.

One Line That Stuck

“Software engineering as a discipline is not going away. It just changes a phase a bit.”

Practical moves to steal

For hiring, Leonid suggests giving candidates more complex take home work because AI is now part of the real engineering workflow. The evaluation then shifts to the candidate’s ability to explain the architecture, defend decisions, describe how AI was used, and show how they tested and constrained the output.

That is a much better signal than asking someone to work as if the tools do not exist.

Subscribe or follow The Tech Trek for more conversations with technical leaders building, hiring, and operating through the next shift in software, data, AI, and engineering execution.

AI coding tools are not just changing how software gets written. They are changing how teams work, how engineers are evaluated, and where bottlenecks show up.

Scott Breitenother, CEO and cofounder of Kilo, joins The Tech Trek to talk about what engineering looks like when developers are managing multiple agents, work continues overnight, and the real constraint is no longer typing code, but judgment, ownership, and process design.

Scott shares how Kilo uses Kilo to build its own product, why AI only creates speed when companies rethink their workflows, and how teams can build trust in agent generated code without creating a new layer of busywork.

Practical Takeaways

• AI does not automatically make teams faster. If approvals, meetings, and handoffs stay the same, the bottlenecks simply move.
• Engineers using coding agents still own the outcome. AI can assist with the work, but accountability for quality does not disappear.
• The strongest teams will find a middle ground between blindly accepting AI output and reviewing every line as if nothing changed.
• Agentic engineering may feel novel now, but Scott believes it will eventually just be called engineering.
• Always on agents are already useful for monitoring, triage, and preparing recommended fixes, even if full autonomy is still selective.

Episode Highlights

00:38 Scott explains what Kilo is building across AI coding, open source infrastructure, and always on agents.
01:16 How Kilo uses its own tools internally, and why developers are shifting from working with one agent to managing many at once.
05:34 Why companies often fail to see AI speed gains when they layer new tools onto old processes.
08:51 The trust curve with coding agents, from early experimentation to accountability, review, and better judgment.
12:39 Why Scott sees agentic coding as a transition phase, not a permanent category.
15:32 Two habits he thinks matter most right now, staying curious and trying a wide range of models and tools.
18:03 What always on agents can already do today, and how that could expand over the next year.

One Line That Stuck

“Bringing in AI does not remove accountability from whoever creates the PR.”

Pro Tips

• Start small with AI assisted workflows, then expand into single agents, multiple agents, and automated review as trust grows.
• Match review depth to risk. A mission critical system deserves more scrutiny than a simple cosmetic change.
• Use automated review to guide human reviewers toward the areas that deserve the most attention.
• Keep experimenting. A tool that fails on Monday may be materially better by Wednesday.

Stay Connected

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