In Production Podcast

Nick Melnychuk
  • 0.0 (0)
  • Business

Most AI conversations happen in boardrooms. This one happens where AI actually runs. In Production is the podcast for CTOs, CIOs, CISOs, CDOs, and tech founders who have moved beyond the theory. The executives and builders making consequential decisions about AI inside real organizations, under real pressure, with real accountability. Every episode, host Nick sits down with a practitioner who has actually shipped AI into production, fought the organizational battles, navigated the resistance, and earned the right to an opinion. Not consultants. Not analysts. Not keynote speakers. The engineers, executives, and leaders who carry the scars and the wisdom to prove it. ______________________________ What we cover: - The gap between the AI demo and the AI deployment, and why most enterprises never cross it - What experienced leaders actually discovered when they tried to automate decisions that lived entirely in someone's head - Why data readiness, legacy infrastructure, and organizational alignment kill more AI programs than bad technology ever will - What genuine AI transformation looks like across financial services, manufacturing, healthcare, legal intelligence, and enterprise software - The honest conversation every technology executive needs to have internally before a single vendor is engaged or a single model is touched ______________________________ Who you will hear from: Guests on In Production have shipped AI at organizations across financial services, banking, healthcare, manufacturing, entertainment, and enterprise technology. They have led teams of hundreds, managed P&Ls built on data products, launched agentic systems inside regulated industries, and navigated the full arc from whiteboard to incident report to scaled deployment. These are not people with opinions about AI. These are people with earned experience in AI. The distinction is everything, and you will feel it in every conversation. ______________________________ Who this is for: - CTOs and CIOs navigating enterprise AI adoption with zero tolerance for hype and full accountability for outcomes - CISOs who need to understand how AI reshapes their risk landscape before the landscape reshapes itself - CDOs building the data foundations that determine whether AI programs succeed or quietly collapse - Tech Founders building in the enterprise AI space who want to understand precisely how sophisticated buyers think and decide - Engineering Leaders who have been asked to deliver on AI and want the honest, unfiltered roadmap from the people who have already done it - Investors who want clear signal on where enterprise AI is actually landing versus where the pitch decks say it will ______________________________ Why In Production? In software, being in production means one thing. It is real. It is live. It has to work. No more pilots running in isolation. No more proof of concepts that never ship. No more AI strategies that exist only inside a presentation. This show is named after the only moment that matters. The moment AI stops being a promise and becomes a system that real organizations depend on, every single day. That is the conversation we are here to have. ______________________________ New episodes weekly. Hosted by Nick, Enterprise AI professional with deep experience across LLMs, blockchain, and large-scale technology, now building the most rigorous and honest conversation in enterprise AI. Subscribe wherever you listen to podcasts. ______________________________ The views and opinions expressed in this podcast are those of the individual guests and do not represent the positions of their employers or affiliated organizations.

  1. Aug 12

    The AI Risk Nobody Budgets For with Rafaeel Chaudhry

    Rafaeel Chaudhry (https://www.linkedin.com/in/rafaeel/) has spent 20 years in FinTech, banking, and enterprise IT across the Greater Toronto Area and Waterloo region, and now advises nonprofits and SMEs on AI adoption. His argument: the biggest risk in most AI rollouts isn't a bad model, it's a good tool that quietly becomes mission critical before anyone has planned the cost or the governance around it. The conversation covers how an FAQ bot turns into a team's default AI colleague without anyone deciding that on purpose, why smaller organizations without a dedicated risk or finance function are the most exposed to runaway usage, the three question test Rafaeel uses to decide what stays with a human, and why vendor demos never show you the years of messy, merged, mislabeled data your team actually has. This one is for CISOs, CFOs, and technology leaders at organizations without a large dedicated risk team, who need a real answer before the AI bill becomes the AI problem. 00:00 - Intro 00:11 - Rafaeel's 20 years in FinTech, banking, and enterprise IT 01:09 - Naming the real biggest AI risk: burning budget 02:56 - The real risk: a good tool nobody planned for 04:47 - How an FAQ bot quietly becomes the team's AI colleague 05:44 - Why donor and customer data need a governance model early 09:21 - Why small team AI rollouts never adopt evenly 12:01 - Where AI resistance actually comes from 14:11 - Automating a person's job makes their invisible judgment visible 16:43 - The three question test: reversibility, explainability, population fit 19:52 - Why vendor demos never show you your real, messy data 21:54 - Making the cost case: an ungoverned tool is a liability 23:37 - Why the fastest movers are the best governed 28:03 - Why bigger companies will get leaner and governance heavy 30:35 - How one nonprofit replaced its whole vendor stack 31:39 - Closing thoughts and where to find Rafaeel

  2. Aug 7

    Conditions for success with AI pilots with Karen Chan

    Karen Chan (https://www.linkedin.com/in/karenfongchan/) is a mechanical engineer by training who has run three full transformation cycles: automotive and lean manufacturing, enterprise agile, and now AI. She advises executives and boards on governance and emerging issues like AI, sits on the Board of Governors for Western University, and is a public appointee to the Council of Professional Geoscientists Ontario. Her core argument, earned across every cycle, is that transformation is a people problem disguised as a technology problem. A system can be objectively better, and people will still resist it, until they understand what is in it for them. The conversation gets specific: a Rogers sales ordering rollout that cut monthly costs 60% and still met resistance from the one team whose workload did not change, and how translating the why into their commission won them over; why the middle management layer, not the technology, is where most rollouts stall; and why not every problem is a Gen AI problem, using Jira-as-overkill and a buy-versus-solve checklist as the cautionary cases. Worth the listen for any CTO, CIO, or CDO feeling pressure to move faster on AI than their organization is ready for. Timestamps: 00:00 | Intro00:12 | Karen's path from mechanical engineering to transformation01:42 | Continuous improvement as the thread through every cycle02:35 | AI is old, Gen AI is what exploded03:02 | Move with it, but at a responsible pace05:15 | Why the process of building, not the product, drew her in06:19 | Change is hard, so people need to understand the why08:58 | The middle management translation layer10:51 | Double work kills momentum and enthusiasm12:36 | Getting people off a system that worked for 10 years13:00 | The Rogers rollout that cut costs 60%13:22 | Every team's life easier, except sales13:36 | The sales team asks: why should I14:09 | Translating the change into what is in it for you15:19 | Most people are not willfully noncompliant15:50 | One change per work order and why shortcuts lose data16:55 | The Uber Eats analogy for broken process inputs18:28 | The FOMO reframe: step back to the problem18:48 | Not every problem is a Gen AI problem20:16 | Jira as overkill and choosing the right tool21:27 | Cheap skill versus expensive agent to read a PDF23:21 | How Karen actually uses AI on executive work27:40 | Where to reach Karen

  3. Aug 5

    The AI Override Nobody Writes Down with Hirak Chatterjee

    Hirak Chatterjee(https://www.linkedin.com/in/hirakchatterjee/) has spent twenty years in technology, half of it inside financial services and insurance, and now owns the general insurance core platforms and services at TD Insurance. He has also been part of major transformation programs across North America. His argument: every time a senior claims adjuster overrides an AI recommendation and turns out to be right, the company is watching its most valuable asset disappear. Not the model. The reasoning behind the override, which almost never gets written down anywhere. The conversation covers why free text boxes fail to capture human judgment, why "human in the loop" doesn't work the way most companies have built it, why AI committees are good at stopping bad decisions but terrible at making sharp ones, and why insurance regulation's decision provenance requirement changes who should own an AI program's limits. This one is for CIOs, CISOs, and claims or underwriting leaders trying to figure out what happens to their best people's judgment once an AI model enters the workflow. 00:00 - Intro 00:13 - Hirak's 20 years across insurance and financial platforms 01:09 - The senior adjuster override that started this whole idea 02:14 - Why most AI projects skip understanding today's business first 03:34 - Naming the judgment layer 06:50 - Why platform cost is the smallest transformation risk 09:19 - The RPA parallel: a team just to keep it running 09:57 - Mapping the exception scenarios where AI breaks 10:49 - Building the context layer AI agents actually need 12:15 - Insurance as an exception heavy, judgment heavy business 14:04 - Why an AI committee kills sharp decisions 15:47 - Decision provenance and why regulation demands a name attached 16:19 - Why human in the loop fails: bias and blind spots 18:44 - Automating a bad process just fails faster, at scale 19:33 - Rethinking the underwriting questionnaire from scratch 20:20 - Killing the yearly renewal batch for real time re-rating 23:37 - The three step fix for capturing judgment 25:40 - Closing thoughts and where to find Hirak

  4. Jul 28

    Everyone is racing into AI. The bill arrives in production with Peter Hondrogiannis

    Peter Hondrogiannis(https://www.linkedin.com/in/peter-hondrogiannis-3045ab58/) has spent close to 26 years in banking and financial services, with nearly two decades in fraud and risk, across TD Bank, NTT (formerly Millennium One Solutions), and now Equifax, where he leads AI enablement. He runs AI into production for a living, which is exactly why he is telling senior leaders to slow down. The industry is moving at what he calls the AI gold rush, and leaders are getting burned in post production because the architecture, the tokenization costs, the agentic resources, and the platform fees were never mapped before the first model was touched. One leader found his AI implementation now costs more than the human-in-the-loop it replaced. The conversation gets specific: the due diligence questions to answer before you engage a vendor, why a fraud and risk mindset catches security and privacy gaps a pure technologist misses, how orchestrated hybridity and an LLM-as-judge keep a human as the final arbiter on complex cases, and why AI theater (high adoption, shallow utility) quietly stalls organizations while the metrics still look green. Worth the listen for any CTO, CIO, or CISO being pushed to move faster on AI than the economics can justify. Timestamps: 00:00 | Intro 00:13 | Peter's path from banking into fraud and risk 03:46 | The AI gold rush and why speed burns leaders 04:16 | The costs teams skip: architecture, tokens, agentic resources 05:30 | When AI ends up costing more than the humans 06:01 | Tokenization economics you have to model first 07:11 | The fraud mindset edge over a pure technologist 08:40 | White hat your own AI before you ship it 10:53 | Logic flattening when you force a veteran's judgment 11:31 | Orchestrated hybridity keeps the human as arbiter 12:10 | LLM as a judge and institutional memory 12:30 | Human overrides are training signal, not failure 14:29 | The translation layer between C-suite and frontline 15:38 | Stealth failure mode and the 60% correction tax 17:33 | The resistance nobody expects: intellectual atrophy 17:48 | Reframing AI from skip the work to force multiplier 19:48 | Beating FOMO: learn to crawl, then walk, then run 21:35 | Why swinging for the fences backfires 24:05 | AI theater versus real force multiplication 27:19 | Voice agents handling 95% of call types 28:05 | Eating the elephant one bite at a time 29:05 | Where to reach Peter

  5. Jul 14

    Fraudsters drive sports cars. Healthcare drives a school bus with Akhil Pandit Pautra

    Akhil Pandit Pautra(https://www.linkedin.com/in/akhil-panditpautra/) is a pharmacist with an MBA who has seen healthcare from three sides: the front line as a pharmacist and pharmacy manager, the regulatory side through committees of the Ontario College of Pharmacists, and now the payer side, running healthcare benefits, fraud prevention, and AI-enabled risk management at one of the largest PBMs in Canada. He opens with the asymmetry that keeps risk leaders up at night. Fraudsters adopt AI faster than the organizations trying to catch them, not because their technology is better, but because they answer to no one on privacy, fairness, or the cost of being wrong. His reframe is that this is an accountability problem, not a technology problem, and that a detected signal is not the same thing as evidence to act. The conversation covers the gap between detection and action and the real damage that happens inside it, why more information often creates less clarity and how to decide which signals actually deserve attention, why every audit carries a hidden cost to providers and patients, and a concrete 90-day starting point that puts curiosity and data readiness ahead of buying any model. Akhil closes on why judgment, not intelligence, becomes the scarce resource, and why the coming fight is over data and model sovereignty. Worth the listen for any CISO, CDO, or head of risk in a regulated industry deciding how to move on AI without breaking the trust they are there to protect. 00:00 | Intro and guest introduction 00:44 | Fraudsters adopt AI faster than the organizations detecting it 01:18 | Why this is an accountability problem, not technology 01:59 | The industry shifts from can we to should we 08:49 | The damage that happens between detection and action 09:03 | A signal is not evidence, the check engine light 10:12 | Governance as what lets organizations move confidently 12:11 | Deciding which signals actually deserve attention 12:37 | The hidden cost of every audit and intervention 14:52 | People do not ask about algorithms, they ask can I trust this 15:15 | AI as decision support, not decision replacement 16:10 | Moving from needle in a haystack to guided search 17:24 | Why e-pharmacies change where the risk lives 20:44 | The 90 day starting point, curiosity before AI 21:11 | A three step framework for responsible adoption 26:06 | AI as a GPS, why judgment outlasts intelligence 26:56 | The coming fight over data and model sovereignty 28:03 | AI changes how we work, not why we do it 29:16 | Closing exchange and where to reach Akhil

  6. Jul 7

    The model is the easy part. Owning it for six months is the job with Tamojit Saha

    Tamojit Saha (https://www.linkedin.com/in/sahatamojit/) is a technology leader who started his career as a telecom developer and now leads engineering, software development, digital transformation, and AI initiatives across complex, highly regulated environments in healthcare. He sits on Northeastern's program advisory committee, and a significant part of his work is figuring out how AI will reshape engineering, leadership, and workforce capabilities over the next few years. His core argument in this conversation: most enterprises no longer have a technology gap. The cloud, the models, the data platforms are all accessible right now. The real constraint has moved to organizational capability, and the rare people who can own AI in production, not just build it, are the bottleneck almost nobody is budgeting for. The conversation gets specific about what that ownership actually looks like. Tamojit breaks down the interview tell that separates tool users from production owners (people who talk about monitoring, drift detection, and incident management rather than just model architecture), why owning a model means taking responsibility after 90 days and after six months, and why most teams budget to build AI but fail to budget to operate it, with governance as the real long-term investment. He also covers explainability in regulated healthcare, where compliance officers and lawyers, not data scientists, are the first to ask why a decision was made, and why training does not equal transformation unless people get to apply it to real work and fail safely. Worth the listen for any CTO, CIO, CDO, or engineering leader stuck between pilot and production who has realized the technology can be figured out, but the people to hand it to cannot. 00:00 | Intro 02:06 | Why the enterprise AI bottleneck moved from technology to people 03:23 | Technology scales fast, human adoption lags, speed learners win 06:01 | From project thinking to product thinking, the living asset 08:36 | Why protecting the status quo creates organizational debt 09:43 | Challenging your assumptions before the market forces it 11:20 | When a model fails the explainability test in healthcare 12:09 | Compliance and lawyers question the decision before data scientists 15:04 | Budgeting to build AI but never to operate it 17:19 | The interview tell separating tool users from production owners 18:04 | Owning a model means responsibility after 90 days 18:53 | Why training does not equal transformation 19:34 | Redesigning the work so people can fail safely 20:34 | Why practical experience beats certifications for new talent 22:05 | What universities miss: system thinking over technology thinking 24:28 | Closing advice: stop treating AI as only technology 25:42 | There is no shortcut around the fundamentals 27:32 | Where to reach Tamojit, and closing thoughts

  7. Jun 30

    The Demo Never Made It to Production. His Did. With Marc Rohde

    Most enterprise AI dies in the demo. Marc Rohde, Chief Business Integration Officer at Andis, runs it in production: 10% of order volume goes from email to warehouse with no human intervention, and a single support bot answers 25% of email volume perfectly. In this episode, Marc and Nick unpack what actually had to happen first. The unglamorous answer: 13 years of data cleansing before AI existed in its current form. They get into why his biggest wins were small and pointed (not big and ambitious), the difference between "AI for me" and "AI for we," and the problem he's still trying to solve, capturing institutional knowledge before it walks out the door with a 15-year employee. If you're a CTO, CIO, or operations leader trying to turn AI pilots into real production wins, this is the playbook. Marc's LinkedIn https://www.linkedin.com/in/marcrohde/ Takeaways Unifying business functions for AI deploymentStarting small and scaling AI solutions Chapters 00:00 — Intro: the Chief Business Integration Officer role explained03:39 — What had to happen internally for the transformation to work04:09 — The original bet: data-driven decisions, 13 years ago05:07 — Migrating off legacy ERP to a cloud platform07:09 — How the Microsoft Copilot decision actually got made09:16 — "I don't care if it's the best LLM" — outcomes over technology10:30 — AI for me vs. AI for we13:36 — The institutional knowledge problem16:48 — Where to push on resistance, and where to step aside19:31 — AI as an intern with PhD knowledge and no street smarts20:17 — The question almost nobody asks: "What did I do wrong?"24:23 — From proof-of-concept to production: pick a reasonable goal24:55 — The one-question bot that handled 25% of email25:41 — AI order entry: 10% of orders, zero human touch26:32 — The ROI reframe: don't think headcount, think capacity28:34 — Closing: ground every decision in business outcomes

  8. Jun 26

    Everyone's Cutting Heads Before the AI Even Works with Conrad Bedard

    Conrad Bedard has rebuilt the procurement function from scratch at 5 companies across banking, insurance, and telecom. He's sat on both sides of the table, buying the technology and governing the risk, and he sees enterprises making the same mistake with AI that manufacturing made with robotics 20 years ago: cutting capacity before the tool can absorb it. In this conversation, Conrad and Nick get into why source-to-pay transformations took almost two years to get approved (and why that was the point), what goes wrong when companies bolt new technology onto broken processes, and why it's almost never procurement that kills an AI deal. They dig into the maturity-jump problem, organizations trying to leap from level 2 to level 5 in one move, and where a regulated buyer should actually start. Essential listening for procurement leaders, CFOs, and anyone in a regulated industry being told to "do more with less" with AI. Conrad's LinkedIn https://www.linkedin.com/in/conrad-bedard/ Takeaways Procurement plays a crucial role in providing total value to the business, beyond cost savings and contract administration.Implementing AI in procurement requires a focus on data quality, process remodeling, and change management. Chapters 00:00 — Intro: 16 years in procurement, buy side and sell side00:56 — Walking into orgs that treat procurement as a cost center01:23 — What the first 90 days actually look like03:40 — Falling into procurement by accident07:13 — Assessing the lay of the land in a new organization09:55 — The #1 implementation failure: new tech on old processes11:32 — "Dirty data in is dirty data out"13:51 — Why approval took almost two years (and why that mattered)16:47 — The robotics parallel: buying the shiny thing too early18:56 — Why governance has to make AI defendable and defensible20:10 — Why procurement rarely says no, but risk and IT do21:44 — The maturity-jump problem: level 2 to 5 in one leap22:20 — Will AI replace junior and mid-level roles?22:56 — Evolving roles vs. eliminating them25:36 — Where to actually start: process, intake, low-risk tasks25:52 — Start small, then go complex

See All (26)

About

Most AI conversations happen in boardrooms. This one happens where AI actually runs. In Production is the podcast for CTOs, CIOs, CISOs, CDOs, and tech founders who have moved beyond the theory. The executives and builders making consequential decisions about AI inside real organizations, under real pressure, with real accountability. Every episode, host Nick sits down with a practitioner who has actually shipped AI into production, fought the organizational battles, navigated the resistance, and earned the right to an opinion. Not consultants. Not analysts. Not keynote speakers. The engineers, executives, and leaders who carry the scars and the wisdom to prove it. ______________________________ What we cover: - The gap between the AI demo and the AI deployment, and why most enterprises never cross it - What experienced leaders actually discovered when they tried to automate decisions that lived entirely in someone's head - Why data readiness, legacy infrastructure, and organizational alignment kill more AI programs than bad technology ever will - What genuine AI transformation looks like across financial services, manufacturing, healthcare, legal intelligence, and enterprise software - The honest conversation every technology executive needs to have internally before a single vendor is engaged or a single model is touched ______________________________ Who you will hear from: Guests on In Production have shipped AI at organizations across financial services, banking, healthcare, manufacturing, entertainment, and enterprise technology. They have led teams of hundreds, managed P&Ls built on data products, launched agentic systems inside regulated industries, and navigated the full arc from whiteboard to incident report to scaled deployment. These are not people with opinions about AI. These are people with earned experience in AI. The distinction is everything, and you will feel it in every conversation. ______________________________ Who this is for: - CTOs and CIOs navigating enterprise AI adoption with zero tolerance for hype and full accountability for outcomes - CISOs who need to understand how AI reshapes their risk landscape before the landscape reshapes itself - CDOs building the data foundations that determine whether AI programs succeed or quietly collapse - Tech Founders building in the enterprise AI space who want to understand precisely how sophisticated buyers think and decide - Engineering Leaders who have been asked to deliver on AI and want the honest, unfiltered roadmap from the people who have already done it - Investors who want clear signal on where enterprise AI is actually landing versus where the pitch decks say it will ______________________________ Why In Production? In software, being in production means one thing. It is real. It is live. It has to work. No more pilots running in isolation. No more proof of concepts that never ship. No more AI strategies that exist only inside a presentation. This show is named after the only moment that matters. The moment AI stops being a promise and becomes a system that real organizations depend on, every single day. That is the conversation we are here to have. ______________________________ New episodes weekly. Hosted by Nick, Enterprise AI professional with deep experience across LLMs, blockchain, and large-scale technology, now building the most rigorous and honest conversation in enterprise AI. Subscribe wherever you listen to podcasts. ______________________________ The views and opinions expressed in this podcast are those of the individual guests and do not represent the positions of their employers or affiliated organizations.

Information

  • Creator
    Nick Melnychuk
  • Years Active
    2026
  • Episodes
    26
  • Rating
    Clean
  • Copyright
    © 2026 Nick Melnychuk
  • Show Website
    In Production Podcast