The Learning Curve

The Learning Curve

Download on the App Store

All The Learning Curve episodes

  • What AI native looks like: Agentic loops with human review

    What does AI native look like in practice? 

    Tom Willis, VP of Software Engineering and Product at Flexion, has been working it out one project at a time. He uses agentic loops to do the heavy lifting and keeps a human in the loop for the decisions that matter.

    His biggest test so far is Opportunity Capture, a pipeline built to support Flexion’s business development by finding government and commercial work the team was missing. The volume it now handles is a long way from where it started, but the first version didn’t go to plan, and what he learned from that early mess changed how he thinks about where AI belongs in a workflow.

    In this episode, Tom walks Matt Sharp and Holly Fake through the Claude Code loop behind the build and the test-driven development habits that keep it honest. He also shares why he won’t make big productivity claims, and what running LLMs locally in a home lab has taught him about AI cost optimization. Along the way, AI turns up in a few less expected places, from the kitchen to a tabletop RPG.

    00:00 Introduction
    01:18 What AI native means right now
    05:04 The Opportunity Capture project
    10:35 Test-driven development with Claude Code
    12:13 How agentic loops work
    18:03 AI as an amplifier of good practice
    22:55 Cooking with AI
    29:56 Running LLMs locally in a home lab
    33:37 Why constraints build better AI habits

    Subscribe to the Learning Curve for real conversations about AI at work, AI implementation, and what it actually looks like to build with AI inside a technology company.

    Learn more about Flexion: https://flexion.us/
    Follow Flexion on LinkedIn: https://www.linkedin.com/company/flexion-inc/

    40 min
  • Self-organizing AI agents: four personas, one team


    Multi-agent AI systems rarely work the way most teams expect.


    In this episode, self-organizing AI agents build an entire product with no orchestrator in charge, deciding together when the work is done. David Puglielli, senior advisor and software engineer at Flexion, walks hosts Matt Sharp and Holly Fake through an experiment in ensemble coding with AI most teams haven't tried yet. Rather than one agent handling everything, David built four distinct AI agent personas that debate, self-correct, and vote on their own progress, a structure he chose specifically to avoid the risks of a single point of failure.


    David explains what happened when he asked multiple agents to each pick a random word, and why they kept landing on the same answer. He also describes a simple AI code verification workflow anyone can try, and the AI test coverage risks worth watching for when agents are left to optimize for a single metric on their own.


    This is what AI agents are capable of building when the only thing checking their work is each other, without any single agent having the final say.


    00:00 Introduction

    06:28 Why David moved away from a single orchestrator

    08:38 The random word test that agents kept failing

    17:26 Fixing it with seeded randomness

    20:00 Turning agents loose on a real build

    21:25 Four personas, one team

    25:03 The risk of optimizing for one thing

    26:34 The driver and verifier pairing


    Subscribe to the Learning Curve for real conversations about AI at work, AI implementation, and what it actually looks like to build with AI inside a technology company.

    Learn more about Flexion: https://flexion.us/
    Follow Flexion on LinkedIn: https://www.linkedin.com/company/flexion-inc
    Subscribe to our channel: https://www.youtube.com/@FlexionInc

    37 min
  • AI in product management: When saying no is the right call

    Every product manager eventually learns to catch the AI project that isn't worth building before it gets expensive.

    In this episode, hosts Matt Sharp and Holly Fake are joined by Technical Strategist and product leader Victoria Ragulina to talk about what happens when a product manager decides to learn AI properly, rather than pick it up as they go. Victoria walks through the formal AI courses she's taken, including one where she built her own agentic AI product from scratch, a tool that prices original artwork by scraping market data and asking a few questions about medium and size.

    Victoria explains how that same instinct shows up in her day-to-day work, including why "should we even use AI for this" has become her first question, how the hidden backend costs of an AI feature are easy to miss until it's too late, and how she builds guardrails into chatbots so end users know when an answer might be wrong. She also shares the personal AI agent she built to solve a very unglamorous problem.

    The real lesson isn't what AI can do for you. It's how quickly you learn to catch the moment it shouldn't be the one doing the work.

    00:00 Introduction

    01:03 From product manager to AI power user

    03:37 Taking AI courses through Maven

    05:36 Pricing art with AI

    07:13 Picking the right tools

    08:42 The new first question from product managers

    12:12 Why AI can't replace a trained expert

    16:13 Unpredictable AI questions

    21:11 Guardrails for AI answers

    24:18 The summer camp agent

    28:05 Key takeaways

    Subscribe to the Learning Curve for real conversations about AI at work, AI implementation, and what it actually looks like to build with AI inside a technology company.

    Learn more about Flexion: https://flexion.us/
    Follow Flexion on LinkedIn: https://www.linkedin.com/company/flexion-inc
    Subscribe to our channel: https://www.youtube.com/@FlexionInc

    31 min
  • Core Web Vitals: What AI optimizes for instead

    AI is changing how we write code and how fast our websites load. But it won't measure, prioritize, or optimize any of it for you.

    In this episode, hosts Matt Sharp and Holly Fake sit down with full-stack engineer Ethan Gardner to talk Core Web Vitals, why AI-generated code is quietly getting heavier and slower even as it ships faster, and what it takes to get cited by AI search tools rather than skipped over. Ethan also walks through a live demo using AI-assisted debugging in Chrome DevTools to diagnose a real performance issue in seconds.

    Alongside the technical work, Ethan shares how he's paired an LLM with a traditional career coach, one of many AI productivity tools for work now reshaping how people think about growth, to figure out what parts of his work are truly bringing him fulfillment and where he'd been spending energy that wasn't serving him.

    Two very different problems with the same lesson: AI can move fast, but it still takes a person to decide what's worth optimizing.

    00:00 Introduction
    01:06 Using AI as a career coach
    07:15 Core Web Vitals, explained
    10:54 Performance as a proxy for the bottom line
    12:59 Why AI-generated code is getting heavier
    21:26 Testing in CI/CD vs. the real world
    28:53 Live demo: AI-assisted debugging
    34:52 Key takeaways

    Subscribe to the Learning Curve for real conversations about AI at work, AI implementation, and what it actually looks like to build with AI inside a technology company.

    Learn more about Flexion: https://flexion.us/
    Follow Flexion on LinkedIn: https://www.linkedin.com/company/flexion-inc
    Subscribe to our channel: https://www.youtube.com/@FlexionInc

    36 min
  • More minds, better code: Teamwork in the age of AI

    Ensemble coding means James Herr and Arthur Morrow spend most of their workday on the same task at the same time.


    In this episode, the two engineers join Matt Sharp and Holly Fake to unpack what AI pair programming looks like on their team and how AI agents have changed the practice.


    Their conversation covers how their team splits navigator and driver roles, why the hardest part of AI implementation has shifted from writing syntax to getting the architecture right, and what happens when multiple people run the same prompt through Claude Code and get different results. AI lets junior engineers produce far more code than senior engineers can review, and ensembling might be part of the fix.


    It's not unique to Flexion, and how this team is adapting could be worth a look for any organization feeling the same pressure.


    Subscribe to the Learning Curve for real conversations about AI at work, AI implementation, and what it actually looks like to build with AI inside a technology company.

    32 min
  • AI invoice automation: What we learned on a pig farm

    A bar conversation. A handshake. A three-week AI proof of concept on a pig farm. Not every AI project starts with a formal brief, and some of the best ones don't.


    In this episode, Matt Sharp and Holly Fake sit down with Flexion software engineer Senai Mesfin to walk through the AI invoice automation proof of concept the team built for Hanor, a pork producer operating across six US states. Using AWS Bedrock and machine learning, the team built an AI invoice processing solution from scratch, and by the time the demo was done, the accuracy was sitting at 97%. 


    Hanor was impressed, but what they really wanted was someone to rebuild their entire data landscape.


    Senai speaks about why that pivot made sense, what it actually takes to get your data foundation into a shape where AI can do anything useful with it, and the personal AI tools he builds outside of work – a side of the story that says a lot about where practical AI is heading.


    What began as a three-week brief became a five-month data engineering engagement and lasting client relationship. For organizations planning their first AI project, the Hanor story is a reminder that the data problem almost always comes before the AI solution.


    Subscribe to the Learning Curve for real conversations about AI at work, AI implementation, and what it actually looks like to build with AI inside a technology company.


    34 min
  • Building an AI policy: Practical AI governance insights from a tech founder

    Everyone's talking about AI. Fewer organizations have actually decided what they stand for when it comes to using it.

    In this episode, Matt Sharp and Holly Fake sit down with Scott Hasse, one of Flexion's founding partners, to talk through how Flexion built its AI policy from the ground up. What triggered it, who was in the room, and what they were trying to get right.

    Scott talks honestly about the internal moment that made a policy impossible to ignore, why they deliberately brought skeptics into the process, and how that made the end result something people could actually get behind. He also gets into why having clear AI governance guardrails is what lets teams move faster with AI adoption, not slower, and how the policy continues to hold up in practice today.

    The approach Flexion took, from assembling diverse perspectives to defining clear guardrails, is one any company can learn from. If your organization is still figuring out its position on AI, this conversation is a practical place to start.

    Subscribe to the Learning Curve for real conversations about AI at work, AI implementation, and what it actually looks like to build with AI inside a technology company.

    31 min
  • How AI fits naturally into agile work: A scrum master’s real workflow


    In this episode of Learning Curve, we explore how AI has quietly become a natural part of agile workflows without feeling disruptive or forced. Matt Sharp and Holly Fake are joined by Shawn Cleary, a scrum master at Flexion, to discuss how AI tools support real, everyday work across planning, sprints, retrospectives, and team communication.


    Shawn shares how AI shows up throughout a sprint, from summarizing discussions and pulling out themes to supporting retros in tools like EasyRetro and improving meeting workflows with Zoom and Otter.ai. We discuss where AI saves time, where it still feels clunky, and why the fact that AI feels “not weird” anymore might be the clearest sign of its value.


    The conversation also digs into small, embedded AI features that make a big impact, the importance of seamless integration for adoption, and how scrum masters can use ai to support team health, navigate difficult conversations, and reduce cognitive load. Plus, we hear a fun example of creative ai use outside of work, including trivia playlists that adapt to the flow of a game.


    If you’re a scrum master, agile practitioner, or team lead curious about practical ai tools, large language models, and low-stakes ways to start integrating ai into your workflow, this episode offers grounded, real-world insights you can actually use. Subscribe to Learning Curve for honest conversations about ai, agile teams, and building technology that works for people.

    21 min
  • How AI agents are actually used at work: Real workflows, real lessons

    In this episode of Learning Curve, we cut through the AI hype to explore how AI agents and large language models are actually being used in day-to-day workflows at Flexion. Matt Sharp and Holly Fake sit down with Kevin Varga to talk about the real transition from limited AI use to multi-agent systems that support productivity, experimentation, and creative problem solving.

    We cover practical AI use cases, what AI agents are genuinely good at, where they still struggle, and what remains weird or surprisingly delightful. Kevin shares how he builds and evaluates AI agents, how different tools perform in different contexts, and how cognitive offloading can change the way teams work. From professional workflows to supercharging tabletop gaming as a dungeon master, this conversation explores real world ai beyond the buzzwords.

    If you’re curious about integrating AI into your daily workflow, understanding multi-agent approaches, or learning how teams are using large language models in practice, this episode offers grounded insights you can actually apply. Subscribe to Learning Curve for honest conversations about practical AI, human-centered design, and building technology people actually want to use.

    21 min
  • Applying AI in the real world through innovation and experimentation

    In episode 4 of Learning Curve, we explore what it really means to apply AI and emerging technology outside of labs and big tech companies.

    Host Matt Sharp and Holly Fake sit down with innovators working hands-on in regenerative systems, experimentation, and real-world problem solving. This episode moves beyond hype to unpack how applied artificial intelligence, learning through iteration, and human-centered design come together in complex environments like food systems and sustainability.

    In this episode, you’ll hear insights on:
    - Using AI in real-world applications
    - Learning through experimentation and iteration
    - Innovation outside traditional technology ecosystems
    - Regenerative thinking and sustainable systems
    - Why building with technology requires trust, curiosity, and adaptability

    If you’re interested in AI, innovation, digital transformation, or how learning really happens in practice, this conversation offers an honest look at what it takes to build, test, and evolve ideas where the stakes are real.

    Subscribe to Learning Curve for thoughtful conversations at the intersection of technology, learning, and innovation.

    32 min