Should engineering managers become hands-on again? If even 20% of what we’re seeing is real, productivity, speed, and cost structures will change. Leadership blind spots at inflection points are expensive. I think eng leaders can’t afford to find it out.

🧵 Yesterday, after a discussion on Discord, and today with @sdepablos, I got inspired. AI is not just changing how software is built. It’s changing how the industry operated for years. The loop is: smaller teams → cheaper feature parity → margin pressure → labor shifts → capital reallocates. 👇

After reading the article on software industrialization, this is my (current) take on Software Engineering + AI. Where am I wrong?

“I already raised my concerns. If they chose to ignore them, that’s on them.” I heard it many times from Tech Leads. But this is not (tech) leadership at all. So... I wrote about it. 👇

As a Tech Lead, “we need more speed” always comes up. Without KPIs, SLIs, and SLOs, that turns into opinions: “I think it’s fine.” // “I think it’s risky.” This is why Operational Excellence matters. 👇

AI won’t break your company (yet?). It will expose slow org design, fragile delivery pipelines, and leadership habits that no longer scale. I wrote down what actually needs to change for CTOs running 100+ engineers. 👇

Si eres Tech Lead y tus métricas no cambian las decisiones, no te están ayudando. KPIs, SLIs y SLOs no van de medir más. Van de dejar de discutir opiniones y empezar a operar el sistema. Nuevo artículo 👇

I joined mytaxi as a Tech Lead in 2018 and quickly realized something uncomfortable: having good ideas and the title did not mean people would listen. What finally changed things was simpler. A small tool and sitting side by side with people, solving problems. Full article 👇

“Why do you blog?” I learned most of what I know by watching great teams up close and trying to apply their habits in the real world. Writing is my way of passing that learning forward. Happy to see a positive impact.

El otro día, un pequeño post de Félix López me inspiró. Los títulos no te dan influencia por arte de magia. Si eres TL y tu equipo no te hace caso, no suele ser porque tu estrategia sea mala. Suele ser porque aún no existe confianza. ¿Qué me gustaría haber sabido en 2018? 👇

The Tech Lead role is among the most misunderstood in engineering teams. It’s not “the best coder”, and it should never be a bottleneck. I wrote a post on what good Tech Leads actually do, the signals to look for, and how EMs can better support them.

¿Qué hace realmente un Tech Lead? Y, sobre todo, ¿cómo saber si lo estás haciendo bien? Hablo de arquitectura, calidad, alcance, principios operativos y del verdadero impacto de un TL: crear un equipo que toma mejores decisiones sin depender de él. 👇

1/2 Considering the success of my article in Spanish, I’ve translated it to English: “When software becomes fast food” AI commoditizes code. So the bottleneck shifts and intensifies elsewhere (CI/CD, architecture, taste, judgment, etc.).

Hoy os traigo un artículo sobre algo que a veces olvidamos: no debemos enamorarnos de la tecnología, sino de nuestros usuarios. Suena fácil y obvio, pero los equipos de ingeniería se olvidan a menudo de ello. open.substack.com/pub/esjoaoqa...

I failed two promotion attempts even when everything “looked right” on paper. In a post-ZIRP world, doing great work isn’t enough. You need level n+1 stories, visibility outside your team, and impact across boundaries. Full article below.

Nueva edición de “El doble clic” - Gestión de contexto en Claude Code, cómo Tinybird reforzó su infraestructura en entornos multi-tenant y mucho más. Aquí 👇 enespanol.joaoqalves.net/p/el-doble-...

Private Equity en tech: todo lo que siempre quisiste saber enespanol.joaoqalves.net/p/private-eq...

Doing active/active K8s setup is hard. In many cases — unless it’s a regional service with its own database, etc — it comes with a lot of hidden complexities. @tjamet writes about our experience at Adevinta doing it. Must read: medium.com/learnings-f...

Cloudflare down creating this huge impact is a nice reminder of why I’m a happy shareholder since a long time ago.

“Meta evaluating AI fluency” is making people mad about micromanagement. Using metrics as prompts, not targets, can solve issues and boost performance. Like DORA metrics at Adevinta, AI fluency needs clarity. Full article: world.hey.com/joaoqalves/...

Managing EN/ES accounts and blogs with diverse content is more challenging than I anticipated, but I’m really enjoying it. Tomorrow, I'll post about how Meta and others are measuring AI adoption and making it a requirement in performance reviews. Stay tuned!

Hoy lanzo “El doble clic” en mi blog. Cada semana compartiré lecturas, charlas y enlaces que me hayan gustado. Perspectivas de @paoramen.com, @eferro.net, @elwatto.bsky.social y más! Si te gustan la tecnología, los productos y las startups, ¡te interesará! 👇

Mi tesis sobre la IA: abarata picar código y desplaza el valor hacia el criterio. Habrá mucho “software fast-food”, pero también una mayor necesidad de expertos que sepan diseñar, integrar y tomar decisiones. open.substack.com/pub/esjoaoqa...

More updated than ever. 2020s engineering leaders must be more into the details. That’s a feature and not a bug. “Servant leadership” (misapplied) created a lot of peace time engineering leaders. Full post on world.hey.com/joaoqalves/...

On the AI Agents vs workflows, my mental model is a sports team. * An agent is like a player knowing specific skills (speed, dribbling, …) * A workflow is like a tactic, using AI agents to achieve a goal (eg, score)

Cloud/FinOps Pro Tip: run Claud Code on AWS Bedrock, as they have the models available. Two benefits: 1. Built-in access control with AWS and temporary credentials. 2. It contributes to your AWS expenditure/commitment, allowing you to leverage this cost for volume discounts.

Un poco a contracorriente: no todo el esfuerzo es tóxico. open.substack.com/pub/esjoaoqa...

“How do you write so many articles in such a short time frame? You must be using AI” Nope. I use Grammarly to (re-)write as I’m not a native English guy. I use ChatGPT mainly to create an illustration, as otherwise HEY World (my blogging provider) puts my photo on it.

"538 microservices? What a mess." Four years ago, I wrote Disasters I've Seen in a Microservices World. I thought by now we'd have solved most of them. We didn't. We just learned to live with the chaos. Link below 👇

I've written a sequel to "Disasters I've Seen in a Microservices World," my most-read blog post. Ten disasters later, it's still chaotic, and with the rise of AI agents, it's bound to get worse. Planning to publish soon. Anyone want to read it before I do?

Sigo contando cosillas :) Hoy inauguro "Desde mi terminal". También en español, pero en formato largo. Explico las razones de crear una cuenta específica para contenido en español. Si queréis leerlo, paso el enlace en el post de abajo. enespanol.joaoqalves.net/p/la-desven...

Everyone talks about multi-cloud and reliability. Here’s what it looked like in practice when I joined Adevinta five years ago. The company had just agreed to acquire the eBay Classifieds Group, its secondhand marketplaces division. On the table: private cloud, GCP, and AWS.

Meta laying off a chunk of the FAIR (frontier research AI lab) led by @ylecun was not in my bingo card for 2025. Any clues why? aftermath after acquisition of @alexandr_wang and the team?

.@karpathy says the best way to learn is to build things. He’s right, but as engineers, we know that’s only half the story. Writing, teaching, or preparing a talk forces you to understand what you have built. That’s what this post is about. Link below 👇

Today in the incident review meeting: “We should prepare for AWS disasters. Multi-cloud.” Checks numbers... “Yeah, it’s cheaper to have the same downtime as AWS.” Every. Single. Time.

Do you know these guys who intern 1-2 months at a company to fix "that one thing" that annoys them about a product? I need to do that @37signals. Goal 1: add subdomains to Hey World. Goal 2: Add a simple "Plausible-alike" analytics for the posts. That's it. The End.

Many people don’t know, but AWS has a few services they run “globally” on us-east-1. So it’s not the same to have an outage there or in another big region (eg eu-west-1). Also likely the reason why Vercel and others are down

LLMs changed everything for me. I got my first money on @hirotahog, thanks to it. Before I had lots of ideas but with my limited time, I threw the towel before I went live. Now I go from idea to prod quickly and I try a lot of stuff. FAFO

At abistama.com, we’re cooking something new. 🧑‍🍳

2025 is my "back to blogging" year! After 1.5 years of hiatus, I've written eight blog posts this year. The fun part for me? Talking about SaaS, Private Equity, being a Head of Eng, and more.

I wrote a post, after @factorialapp's latest announcement: back in the office, five days a week 🔥 Remote is a tool, not dogma. Why? Juniors struggle, culture slips. Hybrid needs rules; remote edges, small teams. Tradeoffs. Full post below

I’ve spent three years "vibe coding" with LLMs. Shipping side projects I’d never have had time to build otherwise. In terms of implementation, I agree with @iagolast and I don't feel it's "magic" anymore, but I do feel a lot more productive. Full post below.

Does anyone else notice how the trend of responding to "Oh, layoffs at company X. I'm sad for you" with "At company Y, we're hiring" feels a bit off? Companies often discuss "empathy" and then...

I built a Slack-to-Jira support bot a while ago. It sat on the shelf: useful, but not perfect. Instead of letting it rot, I cleaned it up and open-sourced it. Inspired by @dhh, Omarchy, and the reminder that you can just share your work. Wanna know more about it? 👇

I’m open sourcing Abistama’s support hero! 🦸 ⚡ Emoji → Jira tickets 🤖 AI-powered summaries 📊 CSAT surveys 🔗 Slack & Jira integration The biz model failed, but maybe it helps your team. Repo in the post below 👇

The Zero Interest Rate Period (ZIRP) encouraged Engineering Managers to prioritize their team’s happiness and feelings almost like a union representative. That is not the job. These folks are having a hard-time the post-ZIRP era.

The two main blockers for AI use in sales, marketing, and non-tech departments are: * Data access: Lack of APIs or costly beyond standard reports/dashboards. * Credit card access: Limited by company systems. People need self-sufficiency to test niche tools.

The other day, I had the chance to have a conversation with @agilestateofM about DevOps, Platform Engineering, SLIs, SLOs and more. It’s meant for people that are start looking into all of this. >>>

🚩"Not a 9-5 job" in a job post: is it a red flag? Some roles, such as those at OpenAI or Anthropic, require extra hours but offer career growth, equity, or a mission-driven impact. Effort can be a choice, not a trap, if the upside’s real. Read more: bit.ly/3HIJKlw

"If yogurts have expiration dates, why don't your nodes?". Simple, to-the-point article by @danielillu on keeping an eye on old K8s nodes and leveraging things like "node TTL" from auto-scalers like Karpenter. medium.com/learnings-f...