Yakisno LLC

How do we avoid burnout? Honestly, no idea :) We learned that productivity without recovery doesn’t work long-term. ⚠️ Burnout rarely happens in one day ㅤ 🧩 Not everything should feel urgent ㅤ 🧠 Focus matters more than hours Burnout often starts with simple stuff: • too many meetings • too many parallel tasks • constant interruptions ㅤ 💬 We also try to speak openly about pressure Ignoring, as usual, is not the way out. To name the problem and face it head-on is the only way to start solving it. ㅤ What helps your team prevent burnout? ㅤ Full on our LinkedIn 👇 linkedin.com/feed/update/ur… ㅤ #InsideYakisno #startup #companyculture #techteams #burnout #buildinpublic

Mindstone Lisbon May AI Meetup is already tomorrow 👁️👁️ See you there! 👋 If you’re in Lisbon - there’s still time to register community.mindstone.com/events/mindsto… #AI #LLM #GenerativeAI #MindstoneLisbon

Memorial Day Remembering those who served and sacrificed. #MemorialDay

Mindstone Lisbon May AI Meetup is happening soon and we´ll attend 🙃🙂 📍 Lisbon 🗓 27 May 2026 The meetup is focused on: ⠀• building with LLMs ⠀• AI workflows & productivity ⠀• real-world AI products ⠀• the future impact of AI on work and society Speakers include: Sergey Ross Artem Orlov Luis Espirito-Santo 🍕 Agenda also includes pizza, drinks, and networking after the talks. Looks like a good place to meet people building and experimenting with AI. Maybe see you 👇👇 community.mindstone.com/events/mindsto… #AI #LLM #ArtificialIntelligence #TechEvents #GenerativeAI #MachineLearning #MindstoneLisbon #LisbonTech #MachineLearning #Networking

AI agents already know how to use tools. The weird part is thatg we still don’t really know how they decide to use them ⚠️ This research looks inside the model itself using: ⠀• Sparse Autoencoders ⠀• hidden activations ⠀• latent execution features ⠀• linear probes The goal is to understand the internal representations behind autonomous behavior. Feels more like debugging cognition itself 🧠 Quick breakdown 👇 linkedin.com/feed/update/ur… Full paper: arxiv.org/abs/2605.06890 #AI #LLM #AgenticAI #Interpretability #MachineLearning #AIAgents #DeepLearning

Google just turned the mouse cursor into something straight out of the future 😭 Their new thing called Magic Pointer lets Gemini understand both your cursor movements and whatever you type at the same time. So instead of explaining stuff forever, you can literally point at something on the screen and tell AI what to do with it.ㅤ You can drag objects around inside images, highlight specific parts instantly, or even ask it to build a shopping list from a recipe page in seconds And this isn’t just a concept demo either — Google already started testing it with real users and some Chrome users are apparently getting access right now 👀👀👀👀ㅤ deepmind.google/blog/ai-pointe… #AI #Google #Gemini #MagicPointer #GeminiIntegration #AIPointer

LLM agents are becoming autonomous. They use tools, memory, reasoning loops, and multi-agent workflows 🤖 But most systems still can’t properly explain: ⠀• why an agent acted ⠀• what influenced the decision ⠀• how risk propagated ⠀• whether the behavior was actually safe ⚠️ This paper proposes Agent-BOM — a graph-based auditing layer for AI agents. Think runtime provenance + observability + security telemetry for agent systems. Quick breakdown 👇 linkedin.com/feed/update/ur… Full paper: arxiv.org/abs/2605.06812 #AI #AgenticAI #CyberSecurity #LLM #AIAgents #SoftwareEngineering

Google I/O 2026 is happening in just 4 days 🚀🚀 ⠀ May 19–20 Registration is still open :)) ⠀ Some of the topics this year: ⠀ ⠀• What’s new in Google AI ⠀• Agent-first workflows from prompt to production ⠀• Google AI Studio & Google Antigravity ⠀• AI coding workflows ⠀• What’s new in Android, Chrome & Firebase ⠀ Event 👇 io.google/2026/ #GoogleIO #GoogleAI #Antigravity #AIagents #WebDev

After the shutdown of Sora AI, we found an interesting archive — AI Graveyard:⚰️ ⚰️153 AI tools that shut down, got acquired and sunset, or got acquired and kept shipping. Worth exploring it 📖 tooldirectory.ai/ai-graveyard #Sora #OpenAI #AIGraveyard #OpenAISora #AICommunity #Startup #AIShutdown

Stay tuned for updates on our linkedin page 👇🏽👇🏽👇🏽 linkedin.com/company/yakisn…

How we try not to get too corporate?? As teams grow, stuff starts to appear: more meetings, more approvals, more "this is just how we do it now." Some structure is fine. Too much, and suddenly you're spending more time on process than on actual work. ⚙️ At Yakisno, we try to keep it simple: We add processes only when something actually hurts We kill the ones that slow people down We stay away from unnecessary layers and approvals Processes should help you work ^^ 💬 What about your team? Which processes make you faster and which busier? Full post on our LinkedIn 👇 linkedin.com/feed/update/ur… #InsideYakisno #companyculture #techteams

🔒 Myth “Users do not want to talk. Voice interfaces do not work.” 🔓 The truth Users want to talk/click less and finish tasks faster Voice agents work when they reduce friction and when speed + availability matter It fits best when: ⠀• hands/eyes are busy ⠀• the task is time-sensitive ⠀• back-and-forth is needed (support, booking, qualification) Also: ⠀• it’s not limited by human availability ⠀• it can handle requests 24/7 ⠀• it often reduces cost vs human support at scale Voice is about removing waiting time. 🔑 What to do 1️⃣ Use voice where waiting is a problem ⠀(support, inbound calls, high-volume flows) 2️⃣ Keep interactions short and goal-driven 3️⃣ Design handoff properly ⠀(AI + human when needed) Voice works when it saves time — for both users and the team. 💬 What about you? Where would the voice be faster than your current flow? #VoiceAgent #Automation #CustomerExperience #Tech #Yakisno

Software engineering is moving past the Copilot era 🚀 Devs define the outcome → agents explore, test, and ship 🤖 The hard part now is controlling how agents work. Quick breakdown in our latest post 👇 linkedin.com/feed/update/ur… #AgenticAI #AgenticSoftware #AI #SoftwareEngineering #LLM #DevTools

OpenAI just dropped a GPT-5.5 prompting guide ⠀ If you want something to be done right you need to ask for it in a way Feels like you are giving a task to someone you work with rather than just talking to a machine ⠀ A few takeaways: ⠀ 🎯Short, outcome-focused prompts usually work better than complex prompt stacks. 🧩Re-evaluate low/medium effort before escalating reasoning. 📌Preambles, phase handling, and item replay still matter in tool-heavy workflows. 🔁Explicit personality, retrieval budgets, and validation rules shape UX. ⠀ Check it out developers.openai.com/api/docs/guide…

🔒 Myth “We’ll fix security later.” 🔓 The truth You can “add” security later, but you pay for it later. What starts as a shortcut turns into: ⠀• risky workarounds ⠀• fragile access control ⠀• expensive rewrites And the worst part - you often discover it under pressure. Security should be the part of the system from day one. 🔑 What to do 1️⃣ Define basic security rules early ⠀(auth, access, data handling) 2️⃣ Don’t overcomplicate, better to start simple but intentional ⠀(roles, permissions, logging) 3️⃣ Schedule security checks as you grow ⠀(before scaling, not after incidents) Fixing it early is cheap, later - painful. 💬 What about you? Have you ever had to fix security under pressure? #Security #Engineering #Architecture #Startups #Tech #Product #Yakisno

Inside Yakisno #3 🧩 How We Distribute Ownership In small teams, ownership is everything. When there is no owner, tasks get stuck. On the other hand when too many people are in charge it gets really messy. At one point we figured out that ownership is actually about being responsible and taking care of things. 👤 We try to avoid shared responsibility. Instead: • one person owns the feature • one person owns the system • one person owns the outcome Others can help, but ownership stays clear. ⚙️ Being a boss doesn’t mean doing everything yourself. For us it means: • making decisions • asking for help when needed • making sure things move Taking responsibility is something you cannot share, but you can share the work. 🧠 Ownership only works with context. So we try to provide: • clear goals • constraints • business impact People move faster when they understand why. 💬 In your team — Do people really own things? Or is responsibility still blurred? #InsideYakisno #techteams #companyculture

Your AI assistant is out here typing essays when you asked a yes/no question… "I'd be happy to help you with that blabla" //8 tokens, 0 value// So why use many token when few do trick 🪨 @juliusbrussee Caveman - a one-line Claude Code skill that grunts the filler out of AI responses. 65–87% fewer tokens. Same answer :)))) github.com/juliusbrussee/…

🔒 Myth “Adding AI = building an AI product.” 🔓 The truth Adding a chatbot or an LLM API doesn’t make your product AI-driven. If AI doesn’t improve a core workflow, it’s just a feature - and often one user tries once (if tries) and never comes back to. AI products: ⠀• solve a clear problem ⠀• fit into the main flow ⠀• save time in a measurable way Everything else is just a demo. 🔑 What to do 1️⃣ Start from the user problem ⠀(what actually gets easier with AI?) 2️⃣ Put AI inside the main workflow ⠀(not as a separate “extra feature”) 3️⃣ Track real usage ⠀(do people rely on it, or just try it once?) If users don’t depend on it - it’s not a good product. 💬 What about you? Have you seen AI features people abandon right away? #AI #Startups #Tech #UserExperience #TechTrends

Had a great Easter. Now back to emails. #postholidaymood #backtowork