Real engineers. Real deployments. Zero hype. We interview the top engineers who actually put AI in production. Learn what the best engineers have figured out through years of experience. Hosted by Nicolay Gerold, CEO of Aisbach and CTO at Proxdeal and Multiply Content.
How AI Is Built
Claim This Podcastby Nicolay Gerold
4.9(21 reviews)
83 episodes
Updated Daily
Accepts GuestsHas Sponsors
64
Podcast Authority
BetaGoodBased on show quality, social media presence, reviews, charts, and more
Quality97
Social0
YouTube0
Engagement77
Podcast Overview
Real engineers. Real deployments. Zero hype. We interview the top engineers who actually put AI in production. Learn what the best engineers have figured out through years of experience. Hosted by Nicolay Gerold, CEO of Aisbach and CTO at Proxdeal and Multiply Content.
Language
🇺🇲
Publishing Since
4/5/2024
Unlock The Full Podcast Authority Score Report
See how your podcast performs across key metrics
64
Podcast Authority
BetaGoodBased on show quality, social media presence, reviews, charts, and more
Quality97
Social0
YouTube0
Engagement77
8
Excellent Areas
1
Good Performance
10
Growth Opportunities
excellent
Publishing Consistency
Every 6 days
Performing excellently!
good
Show Experience
83 episodes over 1.4 years
Recommendations available
Unlock the full report to see detailed tips
poor
Episode Thumbnails
Recommendations available
Unlock the full report to see detailed tips
+16 More Metrics
Unlock comprehensive insights including:
- • YouTube presence analysis
- • Social media reach metrics
- • RSS compliance scoring
- • Podcast 2.0 features
- • Technical standards
What's Included in Your Full Report
Detailed Analytics
- Complete breakdown of all 19 authority metrics
- Personalized recommendations for each metric
- Industry benchmarks and comparisons
- Technical RSS feed analysis and compliance scoring
Growth Strategies
- Step-by-step action plans for improvement
- Quick wins to boost your score immediately
- Pro tips from successful podcasters
Get your free podcast insights report
See how your show performs across every key metric
Instant delivery
No spam
Attract Better Guests
High authority scores make your podcast more attractive to industry leaders and influencers who want to appear on credible shows.
Secure Sponsorships
Sponsors look for podcasts with proven authority and engagement. Your score demonstrates your podcast's value to potential partners.
Grow Your Audience
Understanding your strengths and weaknesses helps you make data-driven decisions to expand your listener base effectively.
1 verified contact email on file for How AI Is Built
Pitch yourself as a guest, propose sponsorships, or reach out directly to the host.
Recent Episodes
September 11, 2025
Building Solo: How One Engineer Uses AI Agents to Ship Production Code
<p>Nicolay here,</p><p>Most AI coding conversations focus on which model to use. This one focuses on workflow - the specific commands, git strategies, and review processes that let one engineer ship production code with AI agents doing 80% of the work.</p><p>Today I have the chance to talk to Kieran Klaassen, who built Cora (an AI email management tool) almost entirely solo using AI agents.</p><p>His approach: treat AI agents like junior developers you manage, not tools you operate.</p><p>The key insight centers on "compound engineering" - extracting reusable systems from every code review and interaction. Instead of just reviewing pull requests, Kieran records his review sessions with his colleague, transcribes them, and feeds the transcriptions to Claude to extract coding patterns and philosophical approaches into custom slash commands.</p><p>In the podcast, we also touch on:</p><ul><li>Git worktrees for running multiple AI agents simultaneously</li><li>The evolution from Cursor Composer to Claude Code and Friday</li><li>Why pull request review is the real bottleneck, not code generation</li><li>How to structure research phases to avoid AI going off the rails</li><li>and more</li></ul><p><strong>💡 Core Concepts</strong></p><ul><li><strong>Compound Engineering</strong>: Extracting reusable systems, SOPs, and taste from every AI interaction - treating each code review or feature build as an opportunity to teach the AI your standards and decision-making patterns.</li><li><strong>Git Worktrees for AI Agents</strong>: Running multiple AI coding agents simultaneously by checking out different branches in separate file system directories, allowing parallel feature development without conflicts.</li><li><strong>Research-First AI Development</strong>: Starting every feature with a dedicated research phase where AI gathers context, explores multiple approaches, and creates detailed GitHub issues before any code is written.</li><li><strong>Tiered Code Review Systems</strong>: Implementing different review checklists and standards based on risk level (payments, migrations, etc.) with AI assistants handling initial passes before human review.</li></ul><p>📶 <strong>Connect with Kieran:</strong></p><ul><li><a href="https://x.com/kieranklaassen">X / Twitter</a></li><li><a href="https://cora.computer/">Cora</a></li></ul><p>📶 <strong>Connect with Nicolay:</strong></p><ul><li><a href="https://nicolaygerold.substack.com/subscribe">Newsletter</a></li><li><a href="https://www.linkedin.com/in/nicolay-gerold/">LinkedIn</a></li><li><a href="https://x.com/nicolaygerold">X / Twitter</a></li><li><a href="https://bsky.app/profile/nicolaygerold.com">Bluesky</a></li><li><a href="https://www.nicolaygerold.com/">Website</a></li><li><a href="https://www.aisbach.com/">My Agency Aisbach</a> (for ai implementations / strategy)</li></ul><p><strong>⏱️ Important Moments</strong></p><ul><li><strong>The Sonnet 3.5 Breakthrough Moment</strong>: [09:30] Kieran describes vibe-coding a Swift app in one evening, realizing AI could support solo entrepreneurship for the first time.</li><li><strong>Building Cora's First Prototype</strong>: [12:45] One night to build a prototype that drafts email responses - the moment they knew there was something special about AI handling email.</li><li><strong>The Nice, France Experiment</strong>: [13:40] Testing automatic email archiving while walking around town, discovering the "calm feeling" that became Cora's core value proposition.</li><li><strong>Git Worktrees Discovery</strong>: [50:50] How Kieran discovered worktrees by asking AI for a solution to run multiple agents simultaneously, leading to his current parallel development workflow.</li><li><strong>Cursor 3.7 Breaking Point</strong>: [19:57] The moment Cursor became unusable after shipping too many changes at once, forcing the search for better agentic tools.</li><li><strong>Friday vs Claude Code Comparison</strong>: [22:23] Why Friday's "YOLO mode" and end-to-end pull request creation felt more like having a colleague than using a tool.</li><li><strong>Compound Engineering Philosophy</strong>: [33:18] Recording code review sessions and extracting engineering taste into reusable Claude commands for future development.</li><li><strong>The Research Phase Strategy</strong>: [04:48] Why starting with comprehensive GitHub issue research prevents AI agents from going off-rails during implementation.</li><li><strong>Pull Request Review Bottleneck</strong>: [28:44] How reviewing AI-generated code, not writing it, becomes the main constraint when scaling with agents.</li><li><strong>Multiple Agent Management</strong>: [48:14] Running Claude Code work trees in parallel terminals, treating each agent as a separate team member with distinct tasks.</li></ul><p><strong>🛠️ Tools & Tech Mentioned</strong></p><ul><li><a href="https://docs.anthropic.com/en/docs/claude-code">Claude Code</a></li><li><a href="https://www.cursor.com/">Cursor</a></li><li><a href="https://friday.app/">Friday AI</a></li><li><a href="https://git-scm.com/docs/git-worktree">Git Worktrees</a></li><li><a href="https://www.warp.dev/">Warp Terminal</a></li><li><a href="https://cli.github.com/">GitHub CLI</a></li><li><a href="https://github.com/charlieai/charlie">Charlie AI</a> (PR review bot)</li><li><a href="https://github.com/ktrnka/context7">Context7 MCP</a></li><li><a href="https://console.anthropic.com/">Anthropic Prompt Improver</a></li></ul>
August 13, 2025
Embedding Intelligence: AI's Move to the Edge
<p>Nicolay here,</p><p>while everyone races to cloud-scale LLMs, Pete Warden is solving AI problems by going completely offline. No network connectivity required.</p><p>Today I have the chance to talk to Pete Warden, CEO of Useful Sensors and author of the TinyML book.</p><p>His philosophy: if you can't explain to users exactly what happens to their data, your privacy model is broken.</p><p>Key Insight: The Real World Action Gap</p><p>LLMs excel at text-to-text transformations but fail catastrophically at connecting language to physical actions. There's nothing in the web corpus that teaches a model how "turn on the light" maps to sending a pin high on a microcontroller.</p><p>This explains why every AI agent demo focuses on booking flights and API calls - those actions are documented in text. The moment you step off the web into real-world device control, even simple commands become impossible without custom training on action-to-outcome data.</p><p>Pete's company builds speech-to-intent systems that skip text entirely, going directly from audio to device actions using embeddings trained on limited action sets.</p><p>💡 Core Concepts</p><p> Speech-to-Intent: Direct audio-to-action mapping that bypasses text conversion, preserving ambiguity until final classification</p><p> ML Sensors: Self-contained circuit boards processing sensitive data locally, outputting only simple signals without exposing raw video/audio</p><p> Embedding-Based Action Matching: Vector representations mapping natural language variations to canonical device actions within constrained domains</p><p>⏱ Important Moments</p><p> Real World Action Problem: [06:27] LLMs discuss turning on lights but lack training data connecting text commands to device control</p><p>Apple Intelligence Challenges: [04:07] Design-led culture clashes with AI accuracy limitations</p><p>Speech-to-Intent vs Speech-to-Text: [12:01] Breaking audio into text loses critical ambiguity information</p><p> Limited Action Set Strategy: [15:30] Smart speakers succeed by constraining to ~3 functions rather than infinite commands</p><p> 8-Bit Quantization: [33:12] Remains deployment sweet spot - processor instruction support matters more than compression</p><p> On-Device Privacy: [47:00] Complete local processing provides explainable guarantees vs confusing hybrid systems</p><p>🛠 Tools & Tech</p><p>Whisper: github.com/openai/whisper</p><p>Moonshine: github.com/usefulsensors/moonshine</p><p>TinyML Book: oreilly.com/library/view/tinyml/9781492052036</p><p>Stanford Edge ML: github.com/petewarden/stanford-edge-ml</p><p>📚 Resources</p><p>Looking to Listen Paper: looking-to-listen.github.io</p><p>Lottery Ticket Hypothesis: arxiv.org/abs/1803.03635</p><p>Connect: pete@usefulsensors.com | petewarden.com | usefulsensors.com</p><p>Beta Opportunity: Moonshine browser implementation for client-side speech processing in</p><p>JavaScript</p><p><br /></p>
July 29, 2025
Maxime Labonne on Model Merging, AI Trends, and Beyond
<p>Nicolay here,most AI conversations focus on training bigger models with more compute. This one explores the counterintuitive world where averaging weights from different models creates better performance than expensive post-training.</p><p>Today I have the chance to talk to Maxime Labonne, who's a researcher at Liquid AI and the architect of some of the most popular open source models on Hugging Face.</p><p>He went from researching neural networks for cybersecurity to building "Frankenstein models" through techniques that shouldn't work but consistently do.</p><p><strong>Key Insight: Model Merging as a Free Lunch</strong>The core breakthrough is deceptively simple: take two fine-tuned models, average their weights layer by layer, and often get better performance than either individual model. Maxime initially started writing an article to explain why this couldn't work, but his own experiments convinced him otherwise.</p><p>The magic lies in knowledge compression and regularization. When you train a model multiple times on similar data, each run creates slightly different weight configurations due to training noise. Averaging these weights creates a smoother optimization path that avoids local minima. You can literally run model merging on a CPU - no GPUs required.</p><p>In the podcast, we also touch on:</p><ul><li>Obliteration: removing safety refusal mechanisms without retraining</li><li>Why synthetic data now comprises 90%+ of fine-tuning datasets</li><li>The evaluation crisis and automated benchmarks missing real-world performance</li><li>Chain of thought compression techniques for reasoning models</li></ul><p>💡 <strong>Core Concepts</strong></p><ul><li><strong>Model Merging</strong>: Averaging weights across layers from multiple fine-tuned models to create improved performance without additional training</li><li><strong>Obliteration</strong>: Training-free method to remove refusal directions from models by computing activation differences</li><li><strong>Linear Merging</strong>: The least opinionated merging technique that simply averages weights with optional scaling factors</li><li><strong>Refusal Direction</strong>: The activation pattern that indicates when a model will output a safety refusal</li></ul><p>📶 <strong>Connect with Maxime:</strong></p><ul><li>X / Twitter: https://x.com/maximelabonne</li><li>LinkedIn: https://www.linkedin.com/in/maxime-labonne/</li><li>Company: https://www.liquid.ai/</li></ul><p>📶 <strong>Connect with Nicolay:</strong></p><ul><li>LinkedIn: https://www.linkedin.com/in/nicolay-gerold/</li><li>X / Twitter: https://x.com/nicolaygerold</li><li>Website: https://www.nicolaygerold.com/</li></ul><p>⏱ <strong>Important Moments</strong></p><ul><li>Model Merging Discovery Process: [00:00:30] Maxime explains how he started writing an article to debunk model merging</li><li>Two Main Merging Use Cases: [11:04] Clear distinction between merging checkpoints versus combining different task-specific capabilities</li><li>Linear Merging as Best Practice: [21:00] Why simple weight averaging consistently outperforms more complex techniques</li><li>Layer Importance Hierarchy: [21:18] First and last layers have the most influence on model behavior</li><li>Obliteration Technique Explained: [36:07] How to compute and subtract refusal directions from model activations</li><li>Synthetic Data Dominance: [50:00] Modern fine-tuning uses 90%+ synthetic data</li></ul><p>🛠 <strong>Tools & Tech Mentioned</strong></p><ul><li>MergeKit: https://github.com/cg123/mergekit</li><li>Transformer Lens: https://github.com/TransformerLensOrg/TransformerLens</li><li>Hugging Face Transformers: https://github.com/huggingface/transformers</li><li>PyTorch: https://pytorch.org/</li></ul><p>📚 <strong>Recommended Resources</strong></p><ul><li>Maxime's Model Merging Articles: https://huggingface.co/blog/merge</li><li>Model Soups Paper: https://arxiv.org/abs/2203.05482</li><li>Will Brown's Rubric Engineering: https://x.com/willccbb/status/1883611121577517092</li></ul><p><br></p>
83 total episodes available
Recent guests on How AI Is Built
Guests from recent episodes — sign up to see every guest that has ever appeared on this show.
Charles Irlbeck
Guest
Charity Majors
Guest
Paul Iusztin
Guest
Adrien Morisot
Guest
Jonathan Yarkoni
Guest
Abhishek Choudhary
Guest
Nicolay Gerold
Guest
Richmond Alake
Guest
Deep-dive analytics for How AI Is Built
Frequently asked questions
Have a different question and can't find the answer you're looking for? Reach out to our support team by sending us an email and we'll get back to you as soon as we can.
- What is How AI Is Built?
- How often does this podcast release new episodes?
This podcast updates daily.
- Where can I listen to this podcast?
This podcast is available on 7 platforms including Apple Podcasts, Spotify, and more. You can also use the RSS feed directly.
- Does this podcast accept guests?
Yes, this podcast regularly features guests.
Legal Disclaimer
Pod Engine is not affiliated with, endorsed by, or officially connected with any of the podcasts displayed on this platform. We operate independently as a podcast discovery and analytics service.
All podcast artwork, thumbnails, and content displayed on this page are the property of their respective owners and are protected by applicable copyright laws. This includes, but is not limited to, podcast cover art, episode artwork, show descriptions, episode titles, transcripts, audio snippets, and any other content originating from the podcast creators or their licensors.
We display this content under fair use principles and/or implied license for the purpose of podcast discovery, information, and commentary. We make no claim of ownership over any podcast content, artwork, or related materials shown on this platform. All trademarks, service marks, and trade names are the property of their respective owners.
While we strive to ensure all content usage is properly authorized, if you are a rights holder and believe your content is being used inappropriately or without proper authorization, please contact us immediately at hey@podengine.ai for prompt review and appropriate action, which may include content removal or proper attribution.
By accessing and using this platform, you acknowledge and agree to respect all applicable copyright laws and intellectual property rights of content owners. Any unauthorized reproduction, distribution, or commercial use of the content displayed on this platform is strictly prohibited.