Sergio Avalos

I have been using A.I. to help code for quite some time and server setup etc. It's brain turns to mush after too many prompts. I have not been able to identify exactly when, but as soon as it starts creating garbage I have to start a new chat. I asked it why this happens, I was working on SQL and here is what it said. I can't believe it can actually write an entire application despite all the bull****t on social media. Does anyone have a better experience with Claud code? or the other coding A.I.'s? with all the bashing I am doing, it is a great tool and greatly improves efficiency if you assume it will trash your code if you give it a chance.

2

1 kommentar

Let's bring back the days of making custom portfolio websites. With AI-generated everything, it adds a healthy splash of human creativity that we all need 💯

10

QA is not my main area of expertise, that's why I went to #QonfX yesterday to meet Berlin's engineering and QA leaders and study what works. Three things I take with me: 1. Quality cannot be delegated or reduced to a single team. It's a responsibility that spans the SDLC, and everyone needs to understand their role. 2. Fast feedback beats test volume. Today's tools allow us to write extensive test suites - and still miss the one thing users care about. 3. AI is changing what software engineers do, not whether we need them. If anything, AI has increased the demand for software engineering expertise. Thanks to the organizers for a great event and so many insightful conversations. I'm off to think about how to make quality everyone's responsibility on my teams.

38

2 kommentarer

I don’t usually post on LinkedIn, but today I'd like to share something with you. I have just launched something I've been building in my free time. What started as a small experiment in early 2023 has now grown into a fully fledged product. Flipswitch is a developer-first feature flag platform built for teams who want powerful tooling without the enterprise pricing games. No per-seat charges, no opaque tiers, no sales process. Just feature flags that work. If you've ever dealt with LaunchDarkly, Flagsmith, or Unleash, you'll know exactly what I mean. The platform supports full targeting rules, gradual rollouts, segments, role based access control, and real-time updates. Native OpenFeature support out of the box. And the best part? It's completely free through end of April. Full access. And if you sign up during the launch period, you'll get founding-member discount when paid plans launch. I'd really appreciate if people in my network would try it. Even more, I'd love to hear what you think. See you there, Fredrik (Link in the first comment)

95

3 kommentarer

In the early 2010s, the "Frontend Developer" appeared; a bit later, the "Mobile Developer", a career path I had been following for a long time. But AI has rewritten the rulebook. While we will always need developers, the craft is shifting rapidly. Let's talk about how we got here and what's next at DevTalksRomania

67

6 kommentarer

As a Software Engineering Manager at Medius and an active mentor at CareerFoundry, I recently had the chance to explore the new course "AI for Beginners: From Concept to Application" as part of the early access program. While I don’t fit the target beginner profile, I still found the experience insightful, especially the hands-on focus on LLM pipelines and prompt engineering using Vertex AI. The course is well-structured, starting with foundational concepts like AI vs. ML vs. NLP, and gradually moving toward real-world applications with LLMs. I appreciated the clear breakdown of complex ideas for new learners, and I believe it’s a strong entry point for those with little or no AI background.  Personally, the most interesting part for me was the LLM pipeline setup and hallucination testing. Even though the course is designed for beginners, this section still gave me ideas I can bring back to my work, particularly for building more reliable GenAI-powered tools. Students are paired with a mentor who is an expert in the AI field to support them throughout the learning journey. Early in the course, I had a 1:1 video call with my mentor, Max Sop, which was incredibly helpful in setting expectations around the course structure and submission formats. That said, I believe it would be a great addition if CareerFoundry offered a more advanced course in the future, tailored to students or professionals with prior technical experience who want to dive deeper into the subject. I definitely recommend this course, once it becomes publicly available, to anyone curious about how AI works in practice, from junior developers to product managers. It’s an accessible and practical introduction to one of today’s most exciting tech fields. #CareerFoundry #AIforBeginners #LLMs #VertexAI #Mentorship

72

1 kommentar

Two years ago I improved checkout conversions at Bolt by 20%... simply by moving an if statement! 😅 It was probably my most impactful work that year and literally took less than a day!!! I was digging into the checkout funnel data trying to find where users are dropping off. Trying to find spots to reduce friction. Bolt powers ecomerce checkouts and we prompt the user to login if we recognize them. I noticed there was a BIG delay between typing my email and receiving that login prompt. It lined up with the data as well... lots of recognized people who chose not to login. Weird!? What I changed was to have the email check happen BEFORE they finish typing. For example, as you are writing the ".com" portion of your email, the code is already checking for accounts that match. If we found an account and it matched with what you finally wrote, it would save you about 3 seconds and 1 less click. This happy path happend about 80% of the time. That got users to get recognized faster and resulted in 20% more checkout conversions. Bolt now has a shopper network of +100M users who all benefit from this change. There are dozens of little wins like this that make the overall product "innovative." It's all about iteration and testing.

66

3 kommentarer

💡 New Blog Post Alert: Understanding Goroutines and Channels in Go I’ve written a breakdown of how Go handles concurrency, diving into goroutines, channels, and how they enable safe communication between concurrent tasks. If you’re exploring Go or want to understand how its concurrency model differs from traditional threading, this guide gives you practical examples and mental models to help it click. 🔗 Read here: https://lnkd.in/d3Fzfjm6 Would love to hear your thoughts or experiences with Go’s concurrency model! #golang #softwareengineering #backend #concurrency #developers

14

Let’s talk about stress in tech. I often see posts about burnout. Personally, I’ve never experienced it, and I think that’s because I draw a very clear line: if I ever have to choose between my mental health and my job, I’ll always choose my health. That doesn’t mean I never feel stress. I’m a responsible person. When I see responsibilities piling up beyond my capacity or control, it weighs on me. And in tech, that happens more often than we’d like to admit. ✨ The biggest factor? Staffing. You simply can’t do more than you can do. When there aren’t enough people, you’re forced to prioritize, dropping tasks, delaying features, or shifting responsibilities elsewhere. But if you’re working on something fundamental that other teams rely on, things get tricky. New requests keep coming, not everyone understands your constraints, and sometimes… you just can’t say “no.” ✨ Another factor is authority. Stress builds when you’re accountable for outcomes without having the decision-making power or resources to achieve them. Either responsibilities need to be reduced, or authority needs to be expanded. Without that balance, stress compounds quickly. ✨ And then there’s support from management. Good leadership doesn’t just hand out tasks. It shields teams from unrealistic demands, sets clear priorities, and steps in when workloads become unsustainable. When managers advocate for their people, stress becomes manageable. When they don’t, it multiplies. I’ve been fortunate to work in environments with healthy staffing, supportive management, and realistic expectations. Still, I face stress at times, perhaps even now as I work through it, and I know many others experience it even more intensely. 👉 So I’m curious: What causes you the most stress at work? And on the flip side, what has helped you avoid stress? Because at the end of the day, burnout is often the result of prolonged stress. The key is recognizing stress early and making adjustments before it compounds into something harder to recover from. #wendysThoughts #stress

12

4 kommentarer

This is a perspective I strongly agree with 👇 The idea of engineers spending time in a Forward Deployed Engineer (FDE) role is underrated. At its core, it forces you to move beyond just writing code → into understanding why the code exists in the first place. This shift is becoming even more important in the AI era. Because writing code is getting easier. Understanding the problem is not.

1

WaitGroup is leaving us 😞😞😞 With the new Go release, we might not require WaitGroup.Add and WaitGroup.Done. Earlier, to manage goroutines, we had to manually call wg.Add(1) before starting a goroutine and then signal completion with wg.Done(). Now, with the new wg.Go() feature, Go will handle all of that internally. Interestingly, this pattern has existed in the Uber Fx framework since day one 🫡🫡 Curious to hear your thoughts on this shift… Old method var wg sync.WaitGroup wg.Add(1) go func() { defer wg.Done() fmt.Println("go is awesome") }() wg.Add(1) go func() { defer wg.Done() fmt.Println("cats are cute") }() wg.Wait() fmt.Println("done") #golang #waitgroups

71

16 kommentarer

After a year of working closely with AI tools, one thing is clear: the real value comes when we know how to work with them. Good software design—especially around maintainability and long-term growth—is more important than ever. When AI takes care of repetitive tasks, we’re free to focus on building smarter, more scalable solutions. And this evolution isn’t limited to engineering. Across industries, the best outcomes come from using new tools is to amplify existing skills—not replace them. Curious how others are navigating this transformation.

6

Ever since I became a Manager, my calendar stopped letting me code - so I built a full-stack Rust SaaS anyway. Obviously just kidding, nobody in tech can get around playing and testing with the most modern AI tools, if they want to stay in the game and I'm curious for new stuff basically every evening. See the first comment for the complete journey but here's a short summary :) **Zero lines typed. Zero diffs read.** (I would have been frightened to death two years ago) The compiler + tests were the reviewers. What I built: KitaCrustacean (multi-tenant kindergarten management SaaS) - Rust backend + WASM frontend + shared DTO crate - RBAC, per-tenant SQLite DBs - ~24k LOC generated by AI - >200 BDD scenarios + `clippy` + `cargo test` as the quality gate ** 6 things that made “vibecoding” actually work in Rust:** - Shared DTO crate → no API drift (compiler enforces contract) - Pick frameworks by LLM training data (reliability > HN opinions) BDD everywhere → “make scenario X pass”, not “please fix login” - “Fix clippy.” = two-word code review - Living docs to survive AI amnesia - Multi-AI dev (Claude vs Codex) + compiler picks the winner Would I ship this at SPREAD.ai? Probably not. But for getting a product idea off the ground: surprisingly effective. Question: If you had to “vibecode” a real app today - what constraints/tests would you put in place first? #rust #actixweb #wasm #bdd #softwareengineering #ai

27

3 kommentarer

A few months ago I decided to build something in Java in the AI agent space. So I did. So I started Javaclaw on weekends. A personal AI assistant framework for Java developers, built on Spring Boot and Spring AI. Slowly, piece by piece, it came together: ✅ A runtime engine that runs agent tasks in a loop ✅ Tools for reading files, searching code, and running shell commands ✅ A policy engine that decides what the agent can do on its own, and what needs your approval ✅ A Slack bot so you can talk to your agent and approve actions right in a thread ✅ GitHub integration: read issues, read files, create PRs The thing that matters most to me: the agent doesn't just act. It asks first. Before creating a PR, it pauses, shows you what it's about to do, and waits for your go-ahead in Slack. That approval gate is the whole point. AI that fits into how teams actually work. I recorded a short demo showing the full flow: Slack → GitHub issue → code fix → approval gate → PR created. It's a side project. It's not perfect. But it works, and I'm genuinely enjoying building it. If this sounds interesting, I'd love a ⭐ on GitHub and would really appreciate your thoughts, what would you find useful in a tool like this? GitHub link in the comment section. #Java #SpringBoot #AIAgents #OpenSource #BuildingInPublic

17

3 kommentarer

Why high-performing tech founders struggle with feedback more than juniors… You’d expect the opposite. More experience. More success. More openness. But it’s not how it works. As a founder, your identity is built on: Being right. Moving fast. Having answers. So when feedback comes… You don’t reject it. You reinterpret it. “They don’t have the full context.” “This wouldn’t work here.” And sometimes that’s true. But sometimes… you’re just protecting how you see yourself. Juniors expect to be wrong. Founders don’t. The better you get… the easier it is to filter out what would actually help you grow. Next time… Don’t just listen to feedback. 💡 Watch how you internally react to it. That’s where the real signal is.

19

18 kommentarer

Amazing article by Shaun Bent what he describes is basically Stoicism applied to teamwork. "Sometimes decisions are made for you".

2

2 kommentarer

POV: How will AI change product teams? 🤔💭 Looking at the current job market for junior devs and QA engineers it is clear that a change is already upon us. AI is becoming more and more capable and integrated in the dev flow. With this we see less need for junior devs and more need for forward leaning senior devs that have a broad T shape as a skillset. AI is rapidly changing how we build and is becoming a more solid part in all parts of the SDLC. From user research and design to development, testing and monitoring. Sam Altman recently said that coding will more resemble playing StarCraft. I think this stands for product development in general and many of the roles we know today will blend together. Regardless if you are new to Software Development or a seasoned pro I think you need to reflect on what this means and adapt to stay relevant. Here is what I think will be important: - Versatility: Understanding various aspects of product development will enable us to contribute more effectively and innovate faster. - Embracing AI: Leveraging AI as a collaborator allows us to focus on creative and strategic tasks, adding more value to our projects. - Passion-Driven Work: With AI handling routine tasks, our creativity and drive become our greatest assets. As we move forward, adaptability and a willingness to learn will be crucial. As well as a deep passion for building. Hating on vibe coding and being stuck in your role will not. #AI #SoftwareDevelopment #Product

28

1 kommentar

Are you able to release software updates as frequently as you would like? Why? I was re-reading the DORA 2024 report (https://lnkd.in/dfdz8kHG) yesterday and wondered how people's use of build and release tooling lined up with the frequency and success rates of their releases. I'm used to the release management tooling described by Dinah McNutt in her chapter in Google's SRE Book (Chapter 8 - Release Engineering, https://lnkd.in/dEuhQ4hZ). The deployment tooling we use is also interesting but not described there. Across the DORA respondents there was a wide range of practices in deployment frequency (from multiple times a day down to perhaps as few as a couple a year), and I see an interesting anti-correlation with release failure rate. Many won't be surprised that infrequent, larger releases fail more often and take longer to recover from, while smaller frequent releases are more likely to succeed and take less time to roll back if they don't succeed. What factors determine your release frequency? Are you limited by tooling, testing throughput, policy, operational workload, or feature velocity? Do you feel that you would be able to release faster if you used different tools? Or is your release cadence primarily determined by something else? It's tempting to suggest industry sector as a factor, but fascinatingly the DORA research refutes this (except in the case of the Retail sector, which they find to be especially nimble).

4

1 kommentar