LLMs for Devs: Model Selection, Hallucinations, Agents, AGI – Jodie Burchell | The Marco Show
The Marco ShowShow Notes
Jodie Burchell (Data Scientist) joins Marco to cut through the hype around large language models. From GPT and Gemini to Claude and beyond — how do you really choose the right one? Benchmarks, context length, hallucinations, ethics, and AGI predictions — all on the table. If you’ve ever asked yourself, “Which LLM should I use for my work?” — this conversation will help you see past the hype. 💡 Topics in this episode: How to compare LLMs for different developer tasks Enhanced vs adaptive thinking (is it just branding?) Corporate considerations: cost, privacy, and hosting The ethics and lawsuits shaping AI’s future Predictions for AGI
⏱️Timestamps: (00:00:00) Teaser (00:00:37) Intro (00:01:08) From PhD and academia to data science at JetBrains (00:02:31) Early NLP versus modern LLMs (00:04:46) From LSTMs to Transformers (BERT and GPT) (00:08:49) The DeepSeek surprise and model scaling limits (00:12:25) Benchmarks, assessments, and confusion for end users (00:17:18) Choosing models in practice (00:21:10) “Thinking” models and reasoning limits (00:23:48) Do you really need the newest model? (00:25:58) Hallucinations and how to handle them (00:28:30) Agents and RAG: real-world applications (00:32:55) Vibe coding: hype versus reality (00:37:46) What are AI agents? Tools, MCP, and multi-agent apps (00:43:01) Self-hosting versus proprietary models (00:45:20) Fine-tuning explained and Hugging Face (00:50:17) Building reliable AI apps (tests, A/B, traces) (00:55:33) Privacy, company data, and self-hosting concerns (00:58:37) Ethical issues: data sourcing and labor (01:04:23) Environmental costs and the push for smaller models (01:06:26) Juniors, skills, and the future of coding with AI (01:09:15) Learning fundamentals in the age of LLMs (01:13:24) AGI: definitions, timelines, and Jodie’s twenty euro bet (01:20:03) Rapid-fire questions: slang, food, and culture (01:25:47) Giveaway (01:26:26) Outro
New episodes every other Wednesday. Subscribe for more deep, developer-focused conversations. Watch the episodes on YouTube: https://jb.gg/themarco.show-youtube
Follow Jodie Burchell: LinkedIn: jodieburchell Bluesky: @t-redactyl.bsky.social Mastodon: @[email protected]
Follow The Marco Show: Instagram: @themarco.show TikTok: @themarco.show
More from The Marco Show
Michael Simons, a longtime Java developer and Neo4j engineer, joins Marco to talk about the engineering principles that have stayed relevant across decades of changing technologies. The conversation covers database-first thinking, Spring and the Java ecosystem, graph databases, developer tooling, and why pragmatic decisions often matter more than chasing the “best” framework. They also discuss how AI is changing software development, where AI tools genuinely help, and why human judgment and responsibility still matter. 💡 In this episode: Why Michael still starts with the data model when building applications
Venkat Subramaniam joins Marco to talk about what AI is really changing for developers. They discuss why AI often impresses novices more than experts, why “AI first” is the wrong mindset, and why critical thinking, fundamentals, and judgment matter more than ever. 💡 In this episode: Why AI is better at finding problems than writing code
Sébastien Deleuze, a longtime Spring Framework committer and member of the Spring AI team, joins Marco to talk about how 13 years on the Spring team shaped his pragmatic approach to software development. The conversation covers how Spring AI fits into the Spring ecosystem, why enterprise AI still needs deterministic workflows and human accountability, and why long-term software quality matters even as GenAI changes how code gets written. 💡 In this episode: How Sébastien joined the Spring team through open source contributions
Josh Long, Spring Developer Advocate at Broadcom, joins Marco to cover 25 years of Spring history and what the JVM's real shot at the AI era actually looks like. Josh has published a weekly blog since 2011 and a weekly podcast since 2018 without missing an episode. This conversation gets into why vibe coding concerns him as an engineer, what accountability for AI-generated code actually means, and why Python's dominance in AI is more a marketing gap than a technical one. 💡 In this episode: Spring's origin story: why the documentation came before the code
Martin Lippert, longtime maintainer of Spring Tools and one of the people who has witnessed developer tooling evolve from Eclipse to VS Code, joins Marco to discuss how AI is transforming Java development, and what that means for the future of software engineering. ⏱️ Timestamps: (00:00) Teaser(00:38) Meet Martin Lippert, Spring Tools & the evolution of Java IDEs(03:28) What happened to Eclipse? Lessons from the rise and fall of a Java giant(12:20) Why Spring Tools moved to VS Code and embraced the Language Server Protocol(20:03) How AI is changing Java development and everyday coding workflows(21:59) Martin's AI workflow: Cursor, IDEs, code review & testing(26:16) Building developer tools for AI agents instead of humans(31:10) Why AI is changing release cycles and developer tooling(36:03) AI review fatigue, cognitive depth & trusting generated code(39:06) Will AI replace senior developers? What happens to junior engineers?(42:16) The future of Java, Spring & software architecture in an AI-first world(48:42) AI's hidden energy problem and the sustainability challenge(54:39) Practical tips for greener software engineering and AI usage(1:00:06) Giveaway & rapid-fire questions