Good talks/podcasts (Jul) / AI & AI - Augmented Coding Edition!

These are the best podcasts/talks I've seen/listened to recently — AI & AI-Augmented Coding Edition! All of them explore how AI and Large Language Models (LLMs) are reshaping software development, product design, and engineering culture.

• How To Get The Most Out Of Vibe Coding | Startup School 🔗 talk notes (Tom Blomfield) [AI, Software Design, testing] [Duration: 00:16] A talk exploring best practices and practical tips for leveraging AI tools and Large Language Models (LLMs) to achieve great results in software development through the "vibe coding" approach.
• Stop Writing Code – That’s what LLMs are for 🔗 talk notes (Steve Yegge) [AI, Engineering Career, testing] [Duration: 00:29] (⭐⭐⭐⭐⭐) This talk by Steve Yegge exploring the inevitable transformation of software engineering due to AI and LLMs, detailing changing developer roles and the increasing importance of testing and validation
• Andrew Ng: Building Faster with AI 🔗 talk notes (Andrew Ng) [AI, Product, startup] [Duration: 00:43] (⭐⭐⭐⭐⭐) Andrew Ng's talk provides best practices for startups to achieve unprecedented execution speed by leveraging new AI technology for rapid engineering, product iteration, and strategic decision-making
• Vibe Coding For Grownups with Gene Kim 🔗 talk notes (Gene Kim) [AI, Architecture, Devops] [Duration: 00:37] Gene Kim discusses the transformative power and inherent dangers of AI-assisted "Vibe Coding", emphasizing the critical role of architecture and sound practices in achieving high performance and avoiding "vibe coding disasters"
• How custom GPTs can make you a better manager | Hilary Gridley (Head of Core Product at Whoop) 🔗 talk notes (Hilary Gridley) [AI, Management, leadership] [Duration: 00:36] This talk demonstrates how custom GPTs significantly leverage managers' time by scaling their expertise and providing consistent, automated feedback to their teams.
• AI prompt engineering in 2025: What works and what doesn’t 🔗 talk notes (Sander Schulhoff, Lenny Rachitsky) [AI, Generative AI, Product, Security] [Duration: 01:37] Sander Schulhoff, discusses tangible prompt engineering techniques (including few-shot prompting, decomposition, self-criticism, and additional information), distinguishes between conversational and product-focused prompt engineering, and delves into the critical and unsolvable problem of AI prompt injection and red teaming.
• How AI is changing software engineering at Shopify with Farhan Thawar 🔗 talk notes (Farhan Thawar, Gergely Orosz) [AI, Developer Productivity, Engineering Culture, Technology Strategy] [Duration: 00:47] (⭐⭐⭐⭐⭐) A deep dive into Shopify's AI-first transformation, emphasizing pervasive AI tool adoption, internal AI infrastructure, and cultural shifts empowering all employees to leverage AI.
• Gene Kim on developer experience and AI engineering 🔗 talk notes (Gene Kim) [AI, Developer Productivity, Devex, Generative AI] [Duration: 00:44] (⭐⭐⭐⭐⭐) Gene Kim explores how Developer Experience, Generative AI, and Platform Engineering serve as the next chapter in organizational transformation, fundamentally improving developer productivity and value creation.
• TDD, AI agents and coding with Kent Beck 🔗 talk notes (Kent Beck, Gergely Orosz) [AI, XP, tdd] [Duration: 01:15] Industry legend Kent Beck, creator of XP and TDD, shares insights on the evolution of Agile, Extreme Programming, and Test-Driven Development, alongside his current experience of "most fun ever" coding with AI agents.
• The Agent Native Company — Rick Blalock, Agentuity 🔗 talk notes (Rick Blalock) [AI, Company Culture, Management, Teams] [Duration: 00:20] This talk introduces the concept of agent native companies, where AI is fundamental to their operations, fundamentally reshaping culture, workflows, and team structures
• AI at Honeycomb: What’s Actually Working 🔗 talk notes (Charity Majors) [AI, Engineering Culture, Product] [Duration: 00:11] A talk with Honeycomb's CTO exploring AI's impact on engineering velocity and culture, the emergence of disposable software, and the future of product interfaces focused on production.

Reminder: All of these talks are interesting, even just listening to them.

You can now explore all recommended talks and podcasts interactively on our new site:

• 🌐 Interactive Site: https://eferro.github.io/eferro-picks-site/

The new site allows you to:

• 🏷️ Browse talks by topic
• 👤 Filter by speaker
• 🎤 Search by conference
• 📅 Navigate by year

Feedback Welcome!
Your feedback and suggestions are highly appreciated to help improve the site and content. Feel free to contribute or share your thoughts!
Related:

• Recommended Talk/Podcast Database
• Recommended resources tag

Refined and Expanded: Custom GPTs for Lean Discovery and Delivery

A while back, I shared how I built a Custom GPT to help teams ship smaller, safer, faster. The goal was clear: challenge teams to focus on learning and delivering value in small, safe steps.

Since then, I've refined that original GPT and expanded the approach.

From broad to focused: eferro Lean Delivery

The original Custom GPT has evolved into eferro Lean Delivery. This version zeroes in on one thing: helping teams deliver value in the smallest, safest, most reversible steps possible — always with zero downtime and minimal risk.

Whether you’re planning a migration, thinking about a tricky rollout, or just trying to keep changes deployable at all times, eferro Lean Delivery is your partner in challenging assumptions and slicing work even thinner.

Check it out here: eferro Lean Delivery

New: eferro Lean Discovery

Alongside Delivery, I’ve created a new GPT: eferro Lean Discovery. This one is all about the upstream part of product work: ideation, problem framing, hypothesis creation, and finding the smallest meaningful experiments.

Where Delivery helps you move safely and incrementally once you start building, Discovery helps you figure out what’s really worth building in the first place. It asks you uncomfortable (but necessary) questions like:

• What problem are we really solving?
• What’s the smallest experiment to validate this idea?
• What signals are we looking for before investing further?

Try it here: eferro Lean Discovery

Where to find them

You can find both GPTs — eferro Lean Delivery and eferro Lean Discovery — in the GPT marketplace. Just search for eferro Lean, and you’ll see them there.

---

Both tools are built on the same spirit:
Software is about learning. The faster (and safer) we learn, the more impact we create.

No tracking, no hidden agendas — just simple tools to help teams focus on what matters: delivering value continuously, with confidence and calm.

Charla: El coste oculto de la complejidad: reconstruyendo una Data Platform

Ayer tuve el placer de dar esta charla en la Pamplona Software Crafters 2025, organizada por 540deg. Fue una experiencia increíble, en dos días (26 y 27 de junio) llenos de energía, aprendizaje y buen ambiente en el Castillo de Gorraiz.

Me lo pasé genial. Vuelvo a casa con la cabeza llena de ideas, ganas de aplicar nuevos enfoques… y, sobre todo, con la alegría de haber charlado con viejos amigos y conocer a gente nueva de nuestra comunidad.

Aunque por temas logísticos no pude disfrutar de toda la conferencia, lo que viví fue 100 % espectacular.

🎥 La charla en vídeo

Ya está disponible el vídeo completo gracias a @sirviendocodigo, que hicieron un trabajo espectacular grabando y editando todas las charlas. ¡Mil gracias por hacerlo posible! 🙌

Si quieres, puedes echar un ojo también a la lista completa de vídeos de la conferencia, hay charlas muy interesantes.

🔗 Si prefieres verla directamente en Google Slides, aquí tienes el enlace: El coste oculto de la complejidad

¿De qué va la charla?

En las plataformas de datos, una complejidad invisible puede ser un gran lastre: equipos atrapados apagando fuegos, operaciones costosas y poca capacidad de innovar. En esta charla cuento cómo, a través de uno caso real, conseguimos hacer visible ese “coste oculto” y reducirlo:

• Equipos atrapados en decisiones técnicas heredadas.
• Falta de feedback rápido.
• Cultura enfocada al control más que al aprendizaje.

Aplicamos principios Lean y XP al mundo de los datos, incluso con sus limitaciones. El resultado: una plataforma más sencilla, resistente y alineada con lo que realmente importa al negocio. Todo ello explicado con sus retos y aciertos.

Lo que noté en el público

Tras terminar me vinieron un montón de felicitaciones. Me dijeron que apreciaron lo honesto de la charla: conté las dificultades reales y los momentos complejos, sin edulcorarlo. Fue genial sentir esa conexión y saber que esa transparencia caló hondo.

Un abrazo enorme para 540deg y al equipo de Pamplona Software Crafters por invitarme y montar una experiencia tan chula. Y también a los organizadores del open‑space: el ambiente de colaboración y comunidad fue de 10.

Si estuviste en la charla, ¡escríbeme! Me encantará seguir hablando, responder dudas o compartir ideas.

Si te lo perdiste, date una vuelta por los slides y compártelos con quien creas que le puede venir bien 😉

Hasta la próxima edición… ¡nos vemos!

YAGNI and the Value of Learning: An Additional Premise

For years, I’ve been applying—almost without realizing it—an extension of the YAGNI principle that I want to share. It’s become part of how we work as a team, a “gut feeling” we’ve refined through experience, and I believe it’s worth making explicit.

Beyond Traditional YAGNI

YAGNI (You Aren't Gonna Need It) is a fundamental principle reminding us not to implement features just because we think we might need them in the future. It's a powerful defense against overengineering and unnecessary complexity.

But there are situations where the premise shifts. Sometimes we know we’re going to need something. It’s not speculation—it’s a reasonable certainty based on product context, business needs, or the natural evolution of the system.

In those cases, our response is not to implement the full solution just because “we know we’ll need it.” Instead, we ask ourselves:

Is there a smaller version of this that lets us learn earlier?

The Value of Learning as a Decision Criterion

The key is to evaluate the learning value of each intermediate step. Not every small step is worth taking—only those that provide meaningful insight into:

• Actual user behavior
• The technical feasibility of our approach
• The validity of our assumptions about the problem
• The real impact on the metrics we care about

When the cost of that small step is lower than the value of the learning it brings, it’s almost always worth it. This is a practical application of Lean Startup principles to technical development.

Nonlinear Risk: Why Small Steps Matter

There’s another factor reinforcing this approach: risk doesn’t grow linearly with the size of the change. A change that’s twice as big doesn’t carry twice the risk—it carries exponentially more risk.

Small steps allow us to:

• Catch issues while they’re still manageable and easy to fix
• Validate assumptions before investing more resources
• Maintain the ability to pivot without major cost (optionality)
• Generate more frequent and higher-quality feedback

How We Apply This in Practice

We’re quite radical about this approach. We aim to get product changes to users within 1–1.5 days, and within that cycle, we ship even smaller technical changes to production. These micro-changes give us valuable information about the “how” while we continue refining the “what.”

Our mental process is almost instinctive: whenever a need arises, we consider multiple options—some that others might call “hacky”—and always choose the smallest possible step, no matter how strange it may seem.

We use techniques like Gojko Adzic’s hamburger method to slice functionality, but we go even further. We constantly ask ourselves:

• “Can we start with a hardcoded version to validate the UX?”
• “What if we begin with a manually uploaded CSV before building an automated integration?”
• “Can we simulate this feature with manual config while we learn the real flow?”
• “What if we do it just for one user or a specific case first?”

This isn’t about being naive about future needs. It’s about being smart about how we get there. Each micro-step gives us signals about whether we’re going in the right direction, both technically and functionally. And when something doesn’t work as expected, the cost to pivot is minimal.

This obsession with the smallest possible step not only reduces risk, it also accelerates real learning about the problem we’re solving and the behavior of the solution we’re implementing.

Connection with Other Premises

This way of working naturally aligns with other guiding principles in our approach:

• Postpone decisions: Small steps allow us to delay irreversible choices until we have more information
• Small safe steps: We work incrementally to reduce risk and increase learning
• Software as a means: We focus on impact, not on building the most complete solution upfront
• Optimize for feedback: We prioritize fast learning over perfect implementation, because we know we don’t have all the answers—we need to discover them

A Premise in Evolution

Like all the premises we use, this isn’t universal or applicable in every context. But in software product development, where uncertainty is high and the cost of mistakes can be significant, it has proven extremely valuable.

It’s part of our default way of working: we always look for the smallest step that lets us learn something useful before committing to the full step. And when that learning has value, it’s almost always worth the detour.

Have you experienced something similar in your work? How do you evaluate the trade-off between implementing something fully and taking intermediate steps to learn?

References

• My premises about software development
• Small Safe Steps 3s workshop
• API contract & Expand. (Small Safe Steps)
• Talk: (3s) Small Safe Steps: The secret to agility
• Continuous learning. What / How
• The Art of Small Steps in Software Development: A Lean Vision
• Developing Software: Postponing Decisions and Working in Small Steps
• Lean Software development: The art of postponing decisions
• Lean Software Development: Decide as Late as Possible
• Lean Software Development: Amplify Learning

Built a Custom GPT to Help Teams Ship Smaller, Safer, Faster

Most teams build too much, too early, with too much anxiety. They optimize for perfect architecture before users, comprehensive features before learning, elaborate processes before understanding real constraints.

The result? Endless discussions, delayed releases, building the wrong thing.

So I built: 👉 eferro Lean – your no-BS delivery copilot

A Custom GPT that works with any product development artifact—PRDs, tickets, code reviews, roadmaps, architecture docs—and asks the uncomfortable, helpful questions:

• "What's the smallest shippable version?"
• "Do we actually need this complexity right now?"
• "What if we postponed this decision?"
• "How can we make this change in smaller, safer steps?"

Perfect for anyone building products: developers, PMs, designers, architects, team leads.

eferro Lean – your no-BS delivery copilot

Use it to:

• Slice big ideas into vertical experiments and safe technical steps
• Plan parallel changes, expand-and-contract migrations, or branch-by-abstraction
• Challenge bloated PRDs or over-engineered solutions
• Turn risky releases into incremental, reversible deployments

It challenges assumptions, slices features into experiments, and guides you toward the last responsible moment for decisions. Questions that create momentum instead of paralysis.

Software is a learning exercise. Every feature is an experiment. The faster we test hypotheses safely, the faster we learn what creates value.

No tracking, no upsell, no agenda. My only intention is to share what I've learned and keep learning from the community.

If it helps one team ship better—with less stress, more learning—that's enough.

Good talks/podcasts (Jun)

These are the best podcasts/talks I've seen/listened to recently:

• Data - The Land DevOps Forgot 🔗 talk notes (Michael T. Nygard) [Architecture, Data Engineering, Devops, Platform] [Duration: 00:47] (⭐⭐⭐⭐⭐) This talk offers a critical look at why the analytical data world is "the land DevOps forgot," and presents Data Mesh as a paradigm shift to enable decentralized, autonomous data operations, emphasizing that successful adoption requires significant organizational and cultural change.
• Jeff Bezos explains one-way door decisions and two-way door decisions 🔗 talk notes (Jeff Bezos) [Management, Mental models] [Duration: 00:03] (⭐⭐⭐⭐⭐) Jeff Bezos explains his mental model of two-way door (reversible) and one-way door (irreversible) decisions, highlighting how to apply different decision-making processes for each in organizations.
• TDD, AI agents and coding with Kent Beck 🔗 talk notes (Kent Beck, Gergely Orosz) [AI, XP, tdd] [Duration: 01:15] Industry legend Kent Beck, creator of XP and TDD, shares insights on the evolution of Agile, Extreme Programming, and Test-Driven Development, alongside his current experience of "most fun ever" coding with AI agents.

Reminder: All of these talks are interesting, even just listening to them.

You can now explore all recommended talks and podcasts interactively on our new site:

• 🌐 Interactive Site: https://eferro.github.io/eferro-picks-site/

The new site allows you to:

• 🏷️ Browse talks by topic
• 👤 Filter by speaker
• 🎤 Search by conference
• 📅 Navigate by year

Feedback Welcome!
Your feedback and suggestions are highly appreciated to help improve the site and content. Feel free to contribute or share your thoughts!
Related:

• Recommended Talk/Podcast Database
• Recommended resources tag