Assembly Line Podcast

Pygio
  • 0.0 (0)
  • Technology

LLMs are stateless. RAG is only as good as your data quality. And "buying" AI isn't a strategy, it’s a procurement error. Welcome to The Assembly Line, where we strip away the marketing ambiguity of the AI era. Join Nic and Dimitri as they bridge the "GenAI Divide." This isn't a podcast about chatbots, it's a deep dive into the engineering layer required to connect models to proprietary data.

Episodes

  1. 5d ago

    There is no AI Strategy Without a Data Strategy | Assembly Line EP7

    Abu-Bakr Khan on LinkedIn: https://www.linkedin.com/in/abukhanaplan/Snowflake: https://www.snowflake.com/en/Pygio: https://pygio.com/Pygio on LinkedIn: https://www.linkedin.com/company/pygio/Every enterprise has an AI pilot that looked brilliant in the demo. Then it rolled out to a thousand use-cases and the token bill arrived.Abu-Bakr Khan leads AI go-to-market across the Middle East, Turkey and Africa at Snowflake. He joined us in Johannesburg to talk about what happens beyond the pilot: who actually owns AI in the enterprise, why routing every prompt to the biggest model is burning budgets, and why the model you pick is not your AI strategy. One client in Saudi Arabia moved from Claude to Kimi in two weeks. The models will keep changing. The problem of cost control and data governance between your source systems of action has to be addressed for a successful AI Strategy to hold.We also get into revenue assurance and the office of the CFO, data sovereignty across Africa and the GCC, how for one client, their data posture solve the problem of having warehouse full of frozen chickens nobody could store, and a job that doesn't exist yet: the virtual employee.There is no AI strategy without a data strategy.This is Episode 7 of The Assembly Line Podcast by Pygio.Guest: Abu-Bakr Khan, GTM AI Leader, Middle East, Turkey and Africa, SnowflakeHosts: Dimitri Gonatas and Nic GonatasCHAPTERS00:00 Intro00:48 Welcome to The Assembly Line, Episode 702:25 Five pounds and a note: the story behind Abu-Bakr07:07 From Sybase to SAP: following his father into tech09:46 Beyond the pilot: the economics of the agentic enterprise13:55 Who owns AI in the enterprise?16:33 Control planes and model routing18:27 The retailer that tried to rebuild it in open source20:21 Are enterprises really scaling open source models?24:04 The biggest bottleneck: accountability28:32 Partnerships, zero copy and living with SAP30:20 Will frontier models make data platforms obsolete?37:15 The additive cost problem on the P&L38:48 What revenue assurance actually means41:27 The CFO as custodian of the data42:20 Frozen chickens: when procurement wins and the business loses43:49 Why waiting days for a report now costs money47:51 Context, models and harnesses: the design patterns49:04 What is Snowflake CoWork?52:29 Data sovereignty across Africa and the GCC54:56 Hybrid, on-prem and pointing Snowflake at local GPUs56:42 Why he joined Snowflake1:01:04 Rapid fire: the most under-hyped thing about the UAE1:02:21 The job that doesn't exist yet: the virtual employee1:04:16 Would you let

    There is no AI Strategy Without a Data Strategy | Assembly Line EP7

  2. Aug 19

    What Cloud Taught Us About AI Economics | Assembly Line EP6

    Learn more about Oktant: https://www.oktant.app/Download Oktant: https://apps.apple.com/za/app/oktant/id6760536966 Darryl Govender spent eleven years building cloud practices for some of Africa's biggest banks and enterprises. He watched an industry promise transformation, deliver something more complicated, and then watched the same story start again with AI. Then he left all of it to build a consumer product about how people spend their lives. We go from a matric exam written out on paper to token costs moving seven hundred times in six months, from sovereignty and IP leakage to the question that made him walk away: would your eight year old self and your eighty year old self be proud of what you are doing right now? CHAPTERS 00:00 Intro00:47 Meet Darryl: eleven years building cloud across Africa02:34 From a scripted plan to writing code at thirteen11:57 Thirty million developers, and the road into consulting13:09 Inside FNB, then Synthesis at twenty five people19:38 What was the promise of cloud?26:41 Cloud as the precursor to the AI wave34:00 Where the economics disappointed38:29 Ninety percent of a business, not one percent of it40:38 Chips, margins and a cost curve that moved seven hundred times43:43 Sovereignty, IP leakage and what the labs keep47:41 Scaling Synthesis from twenty five to three hundred and fifty56:12 Why he walked away, and the eight and eighty question62:59 The eight slices69:09 Rubber balls, glass balls and one degree off course73:11 You fall to the level of your systems79:47 Rapid fire, and where to find Oktant MENTIONED IN THIS EPISODE Companies and products: Oktant, Synthesis, FNB, FirstRand, AWS, Microsoft Azure, Nvidia, Samsung, Anthropic, Claude Code, Cursor, Windsurf, Lovable, Amazon Bedrock, Google Vertex AI, Microsoft Foundry, Gemini, Terraform, Java, Turbo Pascal, HP 12CConcepts: sovereign AI, IP leakage, capex to opex, high bandwidth memory, cloud repatriation, MCP, vibe coding, build operate transferReferences: Atomic Habits by James Clear. The line discussed at 73:11 is "You do not rise to the level of your goals. You fall to the level of your systems." Abraham Maslow's hierarchy of needs at 60:32. Dan Buettner's Blue Zones research on longevity at 66:08, noting this research has been subject to methodological criticism regarding age record reliability. Alex Hormozi on education versus entertainment at 66:26.

    What Cloud Taught Us About AI Economics | Assembly Line EP6

  3. Jul 19

    Selling Sovereignty: Will Snow on GovTech, Trust, and Building Across Africa and the Gulf

    Most founders build a moat to lock customers in. Will Snow does the opposite. He hands governments everything he builds, and argues that giving away the IP is the strongest form of trust there is. In episode 5 of Assembly Line, Dimitri and Nicolas sit down with Will Snow, founder of the Dubai based AI product studio Launch Foundry, for a conversation that runs from an ostrich farm in the Karoo to robots cleaning solar panels on satellites and GovTech. Will traces an unlikely path. A science degree in entomology and geography. A first job mapping a sugarcane to ethanol project in Sierra Leone. A drone company where he learned as much from the crashes as the flights. Following on with years running digital and delivery for one of the world's biggest chicken restaurant brands across Africa and the Gulf, right through the middle of a pandemic. From there we get into the ideas that drive Launch Foundry today. Why roughly 0.3 percent of startups that pitch the big funds ever get backed, and what that inefficiency opens up. Why partnerships, not headcount, are the real engine of scale. How trust actually gets built in the UAE. He builds the case for GovTech and sovereignty, where owning nothing and handing everything over turns out to be the most durable position of all. Chapters:00:00 Cold open00:45 Meet Will Snow01:50 From an ostrich farm to studying insects03:35 Mapping Sierra Leone straight out of university05:30 Building and crashing drones07:30 Elevation models and discovering GPUs early09:30 Mapping at scale across Africa10:30 Running a chicken QSR brand through a pandemic13:00 Owning delivery instead of renting it15:05 Founding Launch Foundry15:35 Why only 0.3 percent of startups get funded17:05 The AI product studio model20:30 Moats, and going to market fast22:00 Trust as the real scaling engine27:30 Building a business through regional turbulence32:10 Diversification as survival34:15 Dubai as a launchpad into Africa41:45 Capital allocation and the launch pads idea45:49 IP, GovTech and selling sovereignty52:05 Rapid fire: AGI, future jobs, advice to a younger self53:55 Wrap up Subscribe for new episodes of Assembly Line.

    Selling Sovereignty: Will Snow on GovTech, Trust, and Building Across Africa and the Gulf

  4. Jun 26

    From Farming to Frontier Models | Assembly Line EP4 | Lott Chidawaya

    Lott is back in the country and in the studio. Fresh off Snowflake Summit in San Francisco, he sits down with Dimitri in the studio and Nick (joining remotely) to unpack what he saw and learnt, why "sovereign AI" is becoming a national security issue, and what the Fable export restriction actually means for enterprises across Africa and the Middle East. We go from autonomous tractors and Africa as the breadbasket of the world, to frontier labs becoming kingmakers, to why optionality might be the most underrated idea in enterprise AI. CHAPTERS 00:00 Welcome back: meet Lott and his journey into AI 03:23 Farming, Africa's breadbasket and autonomous tractors 06:45 Will AI take jobs? Automation, UBI and hype vs the plumbing 11:37 The enterprise in five years: AI as the next internet 15:20 Scaling a team: why the unicorns don't exist19:02 Snowflake Summit: San Francisco, the founders and $14B in the room 25:58 From "if" to "how": the big theme and the AI ecosystem 30:28 Two worlds, and whether Africa has a seat at the table 36:11 The big news: the Fable restriction, sovereign AI and token exporters 41:43 Kingmakers, rate limiting and what it means for banks and telcos 45:19 Data sovereignty, Stargate, open source and who regulates AI49:43 Where the value accrues, regulation and an "AI accident fund" 55:07 Rapid fire, robot nannies and wrap up

    From Farming to Frontier Models | Assembly Line EP4 | Lott Chidawaya

  5. May 14

    Assembly Line EP3: Genesis of AI with Dr Marten Scheffer

    In this episode, Dimitri and Nic are joined by Dr. Marten Scheffer from Quintica to discuss the reality of enterprise AI and the path to full automation. A pioneer in the field, with a Phd in machine learning. Martin’s career has been defined by leading world-class infrastructure projects in over 74 countries, from managing global offices at Motorola to serving as a Managing Executive at Vodacom. Now, as the Head of Go-to-Market at Quintica, the Middle East and Africa’s largest independent ServiceNow systems integration and AI business, he brings a perspective on the evolution of enterprise software that no one else could. We know that today’s AI models are the worst they will ever be. In a world where benchmarks are blown away weekly, companies cannot afford to build for current constraints. Instead, we must build for longevity: focusing on where the technology is going not where it currently is. The conversation discusses the "Execution Gap": the space where data is lost between high-level systems and the physical reality of operations. Whether it is a bank or a manufacturing plant, the hurdle is rarely the algorithm itself; it is the quality and management of the underlying data. Key topics discussed in this episode: The Engineering Evolution: Why demand for software engineers is actually increasing, and how the role is shifting from manual execution to strategic steering. The Power of Context: Why LLMs fail without a unified map of your business, and how to bridge the gap between structured records and unstructured communication. The Future of "Dark NOCs": The reality of fully autonomous operations and the move towards systems that don't require human intervention for routine tasks. AGI and Beyond: Dr. Martin shares his predictions for General Intelligence and why we might be closer to superintelligence than most people realise. Real transformation requires an architecture that understands the "why" behind the work. rofessional Disclosures & References:Dr Marten Scheffer: Guest speaker and Head of Go-to-Market at Quintica.PhD Credentials: Obtained from the University of Johannesburg (formerly RAU).Quintica: The Middle East and Africa’s largest independent ServiceNow systems integration and AI business.External Studies: Discussion includes references to MIT studies concerning human-intensive systems and complex enterprise architecture.Digital References: Mention of the "MIT Monk" YouTube channel regarding insights on AGI and the one-person billion-dollar company.Trademark Disclosure: ServiceNow is a registered trademark and is referenced as a primary "System of Action".

    Assembly Line EP3: Genesis of AI with Dr Marten Scheffer

  6. Apr 2

    Assembly, Unified Context for Agents

    Can AI finally understand the "Why" behind your business? In this episode of the Assembly Line, Dimitri and Nic debut our new flagship product: Assembly. Many enterprises are currently stuck in the "GenAI Divide", using AI for tactical, one-off tasks rather than strategic transformation. Assembly is designed to bridge that gap as a Unified Context Layer, acting as a "digital brain" that stitches together disparate data sources across your entire business fabric. In this episode, we dive into: The Problem: Why LLMs fail at scale without a structured memory layer. Structured vs. Unstructured Data: Integrating systems of record (Oracle, SAP, ServiceNow) with "kinetic" data like Slack, email Teams, and meeting transcripts. Ontologies & Context Graphs: How Assembly uses mapping to understand relationships and actions, not just keywords. Disrupting Professional Services: Why AI is coming for the "Big Four" consulting model and how it will augment human knowledge work. The Future of Outsourcing: How autonomous agents will change the way companies build software and manage legal/audit functions. Referenced in this EpisodeOrganisations & Enterprise SystemsPygio: pygio.comOracle: oracle.comSAP: sap.comServiceNow: servicenow.comMicrosoft Teams: microsoft.com/teamsSlack: slack.comGoogle Workspace (G Suite): workspace.google.comAI Models, Hardware & Point SolutionsOpenAI (ChatGPT): openai.com/chatgptAnthropic (Claude/Opus): anthropic.com/claudeGoogle DeepMind (Gemini): deepmind.google/geminiNVIDIA: nvidia.comGroq (LPU Inference): groq.comHarvey (Legal AI): harvey.aiMaxima (Agentic Accounting AI): maxima.aiThe "Big Four" Professional ServicesPwC (PricewaterhouseCoopers): pwc.comEY (Ernst & Young): ey.comKPMG: kpmg.comDeloitte: deloitte.comResearch & PhilosophyResearch Paper: Lee, J. et al. (2025) ‘On the Theoretical Limitations of Embedding-Based Retrieval’, https://arxiv.org/html/2508.21038v1#:~:text=Google%20DeepMind%20Jinhyuk%20Lee%20Google&text=This%20is%20true%20for%20metric%20spaces%2C%20however%2C,uses%20cosine%20similarity%20which%20operates%20in%20non%2DPhilosophy: Jeff Bezos on "Focusing on what makes the beer taste better" (2008 YC Keynote): ycombinator.com/blogRegional Continuity & SecurityDubai Police: dubaipolice.gov.aeUAE Ministry of Defence: mod.gov.ae

  7. Mar 5

    The GenAI Divide?

    The global technology landscape is currently split by the "GenAI Divide." While the marketing hype promises total transformation, recent research shows that approximately 95% of AI projects fail to deliver a measurable ROI. Why? Because most enterprises are treating AI like a SaaS procurement instead of an engineering challenge. In this inaugural episode, Nic and Dimitri diagnose the structural flaws in modern AI strategy. We move beyond the "chatbot" buzz to discuss the "Stateless" problem of LLMs and why the highest ROI often comes from the most "boring" back-office automations. Referenced in the show:MIT Study: https://mlq.ai/media/quarterly_decks/v0.1_State_of_AI_in_Business_2025_Report.pdf The Bitter lesson: https://www.cs.utexas.edu/~eunsol/courses/data/bitter_lesson.pdf Harvard Stats: https://x.com/saxena_puru/status/2026859505045352884IMF: imf.org PWC: https://www.pwc.com/ Avasant: https://avasant.com/report/it-spending-as-a-percentage-of-revenue-by-industry-company-size-and-region/ Snowflake: https://www.snowflake.com/en/ Google BigQuery: https://cloud.google.com/bigquery Openclaw: https://openclaw.ai/ Cornell: https://arxiv.org/abs/1706.03762

    The GenAI Divide?
  • 7 Episodes

About

LLMs are stateless. RAG is only as good as your data quality. And "buying" AI isn't a strategy, it’s a procurement error. Welcome to The Assembly Line, where we strip away the marketing ambiguity of the AI era. Join Nic and Dimitri as they bridge the "GenAI Divide." This isn't a podcast about chatbots, it's a deep dive into the engineering layer required to connect models to proprietary data.

Information