Saulius Grigaliūnas

Post no. 4 - Status reporting in software development HOW TO BUILD A CULTURE WHERE HONESTY ISN'T PUNISHED Transparency is the hardest thing to sustain in software development. Not because people are malicious, but because they’re human. I’ve seen it happen many times: Teams polish reports to look more positive than reality, risks are left unspoken and vague optimism replaces direct answers. Most of this doesn’t come from bad intent. It comes from fear and pressure. If raising a problem feels dangerous, people will delay it until the last possible moment. By then, options are limited and costs are higher. What you need to do is build a culture of transparency. Here are the practices I’ve seen work: ➡️ Normalize problems. A Yellow (at risk) or Red (off track) project status should not be seen as failure. It should be seen as a signal that the system is working: the problem surfaced early enough to be addressed. If marking something red gets you punished, no one will ever do it again. ➡️ Add structure to reduce spin. Honesty thrives when there’s less room for subjective interpretation. A traffic light system (Green / Yellow / Red) helps, but only if it’s tied to facts: % of work completed % of time elapsed % of budgeted hours consumed If 80% of the time and hours are gone but only 40% of the work is finished, the project can’t be green. No amount of confidence changes that. ➡️ Support the messenger. How leadership reacts when someone reports risk is critical. If the reaction is blame, honesty disappears. If the reaction is: “Thanks for flagging this, let’s work on it together,” honesty gets reinforced. ➡️ Make predictions explicit. When asked “Will this deadline be met?”, project leads should be expected to give a clear yes or no. Not vague optimism, but a prediction. It might contradict earlier sales estimates, but it’s far better than letting silence hide the truth. ➡️ Treat mistakes as learning. After setbacks, postmortems should focus on what happened, what contributed, and what needs to change. Not who to blame. When the outcome is learning, people are more willing to be honest about what really went wrong. The key lesson 👇 Transparency doesn’t grow on its own. It needs a culture where people can raise problems without fear. P.S. I’ll detail the traffic light system in one of my future posts.

25

4.6/5 rating at Devoxx Poland for my databases talk. The feedback wasn't about perfect slides or flawless delivery - it was about staying authentic to my presentation style. My slides feature unconventional layouts and animations that help me explain complex concepts related to distributed systems. Before the talk, I’ve got suggestions to make them more "professional"... 😉 However, here's what I've learned ➡️ when you present with slides that feel natural to your thought process, the whole talk flows better. When you force yourself into someone else's format, you lose that connection to your material. Your presentation style is part of how you communicate technical concepts. 💡 Those personal touches aren't just aesthetic choices - they're how you think through problems. #Devoxx #TechTalks #PublicSpeaking

47

3 Comments

Funny thing is that even after those years and market changes, for many developers talking to customers in "non-techie" is a challenge they are unable to overcome. Moreover, so many of them focus too much on "how", instead of "why" and miss the big picture. So many issues from our industry simple come from these two facts - wrong things build - deadlines missed - PM's who are simply translators from "business" to "tech talk" I could go on...

14

We often talk about "instant" performance, but the numbers tell a different story. I spent yesterday evening at the Node.js Warsaw XXIII meetup. Great vibes (and snacks!) 🍕 , but the standout was the talk by Mattias Hansson on system latency. He used a brilliant visualization: if a CPU cycle is as fast as a blink of an eye, the difference between reading data from a local SSD versus fetching it from the Cloud is staggering. It’s easy to forget the sheer scale of that gap when you're just looking at lines of code. The key takeaway? During work with high level IT technologies, we’re used to everything feeling "instant." But true optimization happens when you respect the physical reality of hardware. Sending a single bit to the Cloud involves a relatively astronomical delay when compared to a local database read—to say nothing of the CPU-Memory cycle. It was a solid reality check on why smart caching and minimizing external round-trips are still the kings of performance. Great talk, and see you on next MeetUps! 👏 #NodeJS #WarsawNodeJS

20

2 Comments

If you are today in Brain Embassy Poland on Młynarska street in Warsaw and your AI usage is like: - using Chat GPT like Google Search (for search, asking question) - asking to reply to emails Come talk to me, I wan't to know what other use cases you have and if what we are trying to build can potentially help you.

5

A great summary of the DevOps Lifecycle from my wonderful colleague Agnieszka Peryie 🌟 What is the DevOps Lifecycle? 🤔 The DevOps lifecycle is central to how we approach DevOps at Gearset. As Aga says, the DevOps lifecycle shows how development and operations are part of one continuous process. The infinity loop represents continuous delivery with a tight feedback cycle ♾️ The lifecycle also promotes a ‘shift-left’ mindset: fixing problems as early as possible, when they’re smaller, faster, and cheaper to resolve 😍 Teams that wait risk larger problems down the line 😭 Give the video a listen and let me know your thoughts!! More information on the DevOps Lifecycle can be found here - https://lnkd.in/g83HUy7z

25

2 Comments

The GDD might map the game, but it’s the tech that shapes how (and if) it actually works. ➡️ When building a casino game, you have to think beyond implementing features. The focus lies on constraints: regulatory, architectural, and operational. Yana Tsekhanska, Lead Delivery Manager, takes you through the lifecycle of a casino game from an engineering-first perspective. It’s simply where architecture, not just ideas, drives success. Swipe through for a look beneath the roadmap. 🗺️ #InnovecsPlay #igaming #delivery

24

I had a great time presenting at Callstack's AI Meetup in Wroclaw, Poland, right before React Universe Conf kicks off. According to surveys, 60% of engineering leaders report no significant productivity gains from AI tools. Why? Developers only spend only 16% of their time actually writing code. Experienced developers will tell you, coding isn't the bottleneck, the issue is the other 84% of the job. In this talk I presented some stats about the current state of AI agents at work. I also introduced some practical tips for optimizing the non-coding parts of the job, like building presentations as React apps, automating technical documentation, and using MCP and APIs to connect AI agents directly to the tools we have to work with, like JIRA, Notion, or Github. The presentation itself was vibe-coded, which allowed me to build it quickly, demonstrating the advantages we can give ourselves as software developers. My slides: https://lnkd.in/exdVhyeY Blog post: https://lnkd.in/eUqHbBRT

35

2 Comments

The Compliance Judge: Meet Themis Are Appstore rejections slowing you down? Been there. It's horrible. Hi, I'm Wisdom, a fullstack dev and sometimes I build really silly things because i have the superpower of coding. I meannnnn... why not? But this is not one of those silly projects that i'll dump after 2 weeks. During the last four (hectic and sleepless 🫩) weeks of the #Codematic_io X #GoogleCloud hackathon—on top of that Cortex 3.0 drilling, which is a story for another day 😭—I built something focused entirely on speed and foundational reliability: Themis, the Hybrid Compliance Checker! Themis is my answer to the misery of rejections from @GooglePlay and @AppStore. It automates pre-submission validation aiming to reduce back and forth, and wait times by up to 85%. That manual auditing—checking complex, frequently updated policies line by line—is a roadblock that costs developers time-to-market. The power of Themis isn't just checking boxes; it's in the intelligence and the precision:  It uses a fast Deterministic Rules Engine for non-negotiable checks (like verifying SDK targets) and leverages Google Cloud AI/ML (Gemini/VertexAI) for the complex policy compliance checking. It pinpoints fixes to the exact line numbers and suggests code snippets. No more guessing games. I built this using foundational Google Cloud Technologies like CloudSQL for data reliability. Our job is to deliver reliable systems; Themis is my tool for delivering reliable deployment. If you are a builder who has lost two weeks of your life waiting for a re-review because of a single, missed manifest field, Themis is coming to save you! #Themis #DevTools #buildwithcodematic #buildwithgoogle #FullStackDeveloper #AI

19

5 Comments

Phone scammers have gotten scary good! I was reading Reddit today and came across an interesting post on the Estonian subreddit. Here’s the post: https://lnkd.in/duvzW__V For the non-Estonians: The post is about an AI chatbot calling an Estonian person and speaking fluent Estonian in an attempt to scam them into entering their Smart-ID PIN to log into a secure government or bank website. To give more context: Estonian is a language spoken by only about 1.2 million people and is notoriously difficult to learn, as it’s not closely related to most other languages—except Finnish. I’ve been working on PolyglotPal, an AI speech-to-speech chatbot, so I’m honestly not too surprised by this. In the last year and a half, AI speech-to-speech agents have improved at lightspeed. You can now use tools like LiveKit or Vapi.ai to build AI agents that combine different STT (speech-to-text), TTS (text-to-speech), and LLMs to sound indistinguishable from a human—even in languages that aren't widely spoken. Many businesses are already building AI chatbots for customer support across platforms like WhatsApp, phone, and email. Some text-to-speech providers can even clone human voices, complete with filler words and subtle imperfections, to make them even more convincing. It’s probably not far off—or maybe already happening—that scammers clone a loved one’s voice to make fake emergency calls. Imagine getting a call from your child, begging for help and money, when it’s actually an AI chatbot pretending to be them. One of my favorite examples of what’s possible with AI speech-to-speech systems: https://www.sesame.com/

2

Is thirty years a lot for a mainstream programming language? Check out our take on how Java has evolved over such time.

3

Good communication can change everything. From the quality of recruitment to the team atmosphere. In a time when the job market is changing faster than ever, how people communicate determines whether organizations attract or lose talent. What do teams gain when organizations pay attention to developing communication skills? 🧐 🟣 Employees resolve conflicts faster and negotiate more effectively. 🟣 Candidates better understand expectations and are more likely to stay longer. 🟣 Leaders build trust and loyalty through openness and cohesion. 🟣 Teams create a culture where everyone feels heard. Recruitment is the first test of communication. As many as 48% of people leave a new job because the reality differed from what they heard in the interview. Clear communication, open-ended questions, and honesty from the outset help prevent turnover and build good relationships from the start. In difficult times, such as downsizing, communication becomes a shield protecting morale and the company's reputation. Transparency and empathy reduce tension and demonstrate that the organization treats people with respect. When companies prioritize open dialogue and the development of communication skills, they create an environment where collaboration becomes natural and people want to stay. It's an investment that truly pays off. #communicationstrategy #leadershipdevelopment #hrtrends

4

We just shipped Custom Metrics in TraceKit. Here is why this matters: Traces are great for debugging individual requests. But when your CEO asks "how many orders did we process today?" or your ops team needs to know if the job queue is backing up - traces alone do not cut it. Custom metrics let you track the numbers that actually matter to your business: - Revenue per minute - User signups by source - Queue depth across services - Cache hit rates - Payment success rates - Any KPI you can think of The best part? You add them with 3 lines of code. Same simple API across Go, Python, Node.js, PHP, Laravel, and Java. No new infrastructure. No Prometheus. No Grafana. Just metrics that show up next to your traces. We built this because we kept hearing the same thing: "I love TraceKit for debugging, but I still need Datadog for metrics." Not anymore. Check it out: https://lnkd.in/ewrXVS2d

3

1 Comment

Mateusz Olejarka has just started his talk on current threats in software development - with real examples and practical recommendations. It’s hard to find a better person to speak about this topic: ✅ delivered 90+ trainings for engineering teams across Europe ✅ performed countless penetration tests and code reviews ✅ understands how real vulnerabilities are introduced into products If you build any kind of software - you should hear what Mateusz has to say. The Hack Summit 🤟

22

2 Comments

Dream Broker tried to use employee NDAs so that their toxic leadership practices would not get out. Surprise! They were still widely known. Most NDAs I've seen in my career have not protected innovations, instead: - ticked the box for foreign investors, who thought "every startup company should have an NDA", - covered up quality issues in a product, or - hid the fact that the company did not have any "secrets", i.e., innovations.

13

I've been following Piotr Karwatka's posts about Open Mercato and how he approaches AI development and it motivated me to share my own AI-assisted workflow around Cursor IDE. I've been experimenting recently a lot with Cursor rules combined with MCP, and to be honest, I think I found a pretty nice solution that works for me. What are Cursor rules? They're markdown files in .cursor/rules/ that tell Cursor how to write code for your project. It's like coding standards that the AI follows automatically. The part that really surprised me was that I created a rule that watches for patterns and suggests new rules. If a pattern appears in three or more files, it suggests creating a rule. The same happens when an agent detects that I prompted something that could be a standard. Here's an example from my current project. I'm using Supabase, and I wanted all database operations to go through Edge Functions instead of direct queries. So I created a rule that shows the wrong way (direct supabase.from() calls) and the right way (using supabase.functions.invoke()). It includes examples and explains why. It applies automatically to all frontend TypeScript and JavaScript files. The AI now enforces this pattern. If I forget and try to query directly, it catches it and suggests the Edge Functions approach. The rule also guides how to structure Edge Functions: proper error handling, input validation, consistent response formats, and security considerations. It's become a guide for building good API endpoints. This gets even more powerful when combined with MCP. MCP gives the AI access to tools like browser automation for testing, file system operations, etc. I can ask the AI to test the frontend, verify API endpoints, and check edge functions - all through natural conversation. It's like having a technical assistant that can actually interact with your system. One important note on security. I use a dedicated browser only for AI development. I don't use it for daily browsing or personal accounts. This keeps my development environment isolated from my personal data. It's a small step, but it matters when working with AI tools that have broad access. Cursor rules plus MCP equals a more consistent, maintainable codebase. The AI learns your patterns, enforces your standards, and can actually test and verify your work (or its work). If you're using Cursor, I'd recommend starting with one rule for your most common pattern, then adding a self-improvement rule that suggests new ones, and letting the system evolve with your codebase. Another good place to start: https://lnkd.in/dp4qfWBD.

15

I want to be honest about something. Our industry is scared. And honestly? We have every reason to be. AI, English bias, semantic leakage, tokenization imbalances. Linguistic doom spirals. Layer after layer of issues. I actively write about most of this (and will continue to)! And when you’re deep in the weeds of all that, it is so easy, so easy, to slide into a place where the problems are so big and so tangled that the only logical response seems to brace for the worst, to feel disillusioned with our industry, or worse... to throw your arms up and give up. But fear is absolutely TERRIBLE for imagination. Fear closes us off. It narrows our thinking. It makes us reactive. It makes us second-guess everything before it even has a chance to take shape. And I see this happening to us right now. Many of us are in a creative doom spiral! One where we get so caught up in everything that’s broken, everything that’s uncertain, everything that could go wrong, that we forget to ask much more exciting questions: ✨ What could be? ✨ What if we get to shape what’s next? ✨ What could go right? ✨ What if, instead of spending ALL our energy cataloguing the risks (still important), we spent SOME of that same energy imagining the future we actually want (also very important)? We have so many brilliant minds. I know I say that a lot lately, but I mean it sincerely. The people in localization, in language technology, language research, in the intersection of culture and tech... we are full of incredibly smart, creative, deeply thoughtful people. People who have spent years, decades even, understanding how language works, how culture shapes meaning, how a single word choice can completely change how a user feels about a product. That is not a small thing. That is an extraordinary thing. So I wrote about dreaming. I want us to dream so big and so boldly and so LOUDLY. I want YOU to ask yourself, what do I wish the world of language and technology looked like? What would make me genuinely excited to come to work? What would make me feel like what we’re doing actually matters in the way I always hoped it could? What wild crazy thing do I wish existed? We are living through a period where the tools and the technology and the attention are all converging in a way that has never happened before for our industry. For the first time, the problems that loc professionals have been tackling, iterating on + solving for years (like accuracy, consistency, cultural appropriateness, bias, tone, trust, context/memory) are problems that the entire tech world is now talking about. You guys, THE ENTIRE TECH WORLD is talking about this stuff! That is not a small thing. This is an open window! And it won’t stay open forever! I so mean it when I say THIS IS OUR MOMENT. Linking to my full piece "We Have Been Afraid for Long Enough. It's Time to Dream!" in the comments, can't wait to hear your thoughts! Image credit: Two Windows on the Same View, 2024 by Diango Hernandez

49

9 Comments

Sometimes the cleanest cost optimizations break production in the most unexpected ways. Last week was… intense. We were doing a cost optimization on an ECS-based production system. Nothing exotic. A very reasonable goal. Our EFS costs had grown to ~$30–40/day — mostly due to read/write operations, not storage itself. After digging in, we found a background process that: • downloaded files • zipped them • uploaded the archive to object storage • cleaned up temporary files EFS was effectively being used as transient storage. So we made a change: • moved the process to /tmp • increased ephemeral storage for containers • no long-term data, no shared filesystem Result? 📉 ~70% cost reduction on EFS. Looked perfect. Until alerts started coming in. The process was executed asynchronously via an event-driven flow, but the actual work was happening inside the same container that was serving live web traffic. No dedicated workers. No resource isolation. Just “one more async job.” It worked — until it didn’t. Question to the community: Where do you see the architectural risk here, and how would you redesign this to keep cost optimizations from turning into production incidents? I’ll share our solution in a follow-up. #aws #cloudengineering #architecture #production #costoptimization

5

1 Comment

🚀 Reimagining Spring Boot Deployments with GraalVM As Java developers, we’ve all accepted that running a Spring Boot service means having a JDK installed, large Docker images, and longer startup times. But with GraalVM, that assumption no longer holds true. GraalVM enables Ahead-of-Time (AOT) compilation, allowing us to transform a Spring Boot application into a standalone native executable — no JVM required at runtime. This executable contains both the application logic and the necessary JDK components, resulting in: ⚡ Near-instant startup times 📦 Significantly smaller container images 🧩 Simplified deployment pipelines 💰 Reduced infrastructure costs The result is a self-contained binary (.exe on Windows or ./app on Linux) that you can run directly. This shift from JIT to AOT isn’t just a performance tweak — it’s a paradigm change for how we think about deploying Java in cloud-native environments. If you’re using Spring Boot 3+, exploring GraalVM isn’t optional anymore — it’s the next logical step in evolving your architecture. #java #springboot #developer #java17 #spring #graalvm

17