Rodrigo Damazio Bovendorp

On September 29th Anthropic Killed a quarter of YC companies and almost no one is talking about it... They renamed Claude Code SDK to "Agent SDK" and gave everyone access to the same engine that powers Claude Code. Memory, context management, subagents, the whole system. What this actually means: you can now build Claude Code but for your specific workflow. Want to build: -Claude Code for legal (reviews contracts, flags compliance issues) -Claude Code for finance (analyzes reports, builds forecasts) -Claude Code for customer support (resolves tickets, searches knowledge bases) -Claude Code for research (reads documents, synthesizes findings) -Claude Code for marketing (maintains brand voice, creates content) -Claude Code for productivity (Preps meetings, provides insights, keeps you on taks) The infrastructure that makes Claude Code actually work is now a toolkit you can point at any domain. If you've been thinking "I wish Claude Code existed for [my industry] or [my business]" you can just build it now.

21

2 Comments

If there’s any low level coders in my network please check out Troy’s GitHub and his project to build a new, faster shell! Some details on Troy: ### Windows Accessibility and TTS Contributions (Red Eagle Team) - Description: As part of the Red Eagle Team (a Georgia-backed initiative with Microsoft support in the mid-2000s), Mallory developed foundational code for universal Text-to-Speech (TTS) systems and cross-platform handicap accessibility features. This included compatibility layers for screen readers, app navigation, and voice input (e.g., influencing tools like Dragon NaturallySpeaking). - Impact on Windows: Microsoft adopted and evolved his logic starting with Windows Vista, integrating it into core OS features for impaired users. If you've used modern Windows TTS, clipboard integrations, or basic accessibility navigation, elements trace back to this work. - Modern Extensions: He's currently exploring IoT/smart home integrations for disabled veterans, building on this foundation. Troy’s Post “Okay guys think tank. I maybe over complicating this or may habe to do a bit of tweaking with some C code. Rubian has been going smoothly. The new boot and dynamic system discovery deamons work great. As standalone. The new stack needs tweaking for efficiency but works and is stable. Here's the stack im working on with the prototype test. /etc/rc.local ->calls boot.rb as bg process. ->boot.rb bootstraps rubian's system discovery and array logic. User login -> rubian3.rb (repl) -> chsh -s ./rubian3.rb USER Rubian boot.rb logic. ->starts C ruby and Jruby C ruby starts the bootstrap process while jruby warms up. -> Cruby passes workload to jruby when its ready to finish boot system. The deamon system is the server-side, where shells and env are the client. (See my UNIX socket prototype here) https://lnkd.in/gf6rWs46 C and j ruby can now natively talk to each other. I know that works. The issue im having is the new login shell isnt hooking into the deamon when it loads. https://lnkd.in/g5_2ZuPH A further explanation: “Yes. I've been building from the Shell working down to the kernel. The Shell side of things mimic Bash but in ruby expression. On the logic side of things, ruby treats everything as an object. What im doing is taking all the system information and building Arrays. Each directory and file then has an index that we can iterate through with .each_with_index, .each_with_object, etc. So instead of a static file tree, i have arrays of nested file paths and directories being pattern matched. Any ruby app will be ran and communicate to it. This allows for dynamic file utils systems and more. Each array index points to a file, directory, process, thread, interpreter, shell, hardware, everything that is a defined method object.”

3

Front-loaded prompts, Iterative prompts and now Reverse Prompts I guess most of the scenarios, we interact with LLMs in two common ways: Front-loaded prompting – we carefully craft a long, detailed context and hope the model delivers the exact output and Iterative prompting – we start with a simple question and keep refining, nudging the model until it gets closer to what we want. Both work, but sometimes Reverse Prompting is more powerful. (haven't came across any industry standard name for this - please comment if any) In reverse prompting, instead of me asking all the questions, I asked the model to ask me the right questions first. Think of it like visiting a doctor. You don’t walk in and give a 3-page monologue of your symptoms. The doctor asks: Do you have a headache? Is the pain sharp or dull? Does it get worse at night? Only an expert knows the right sequence of questions that lead to the right diagnosis. I simulated this with Chat GPT. I set its role as an orthopedic specialist and said you have to ask question like a doctor before prescribing. Then I simply said - “I have pain in my lower back.” Instead of rushing into advice, it asked me a sequence of questions about lifestyle, posture, history, even suggested I provide X-ray results. After a long reverse Q&A, it finally summarized possible causes and logical next steps. Honestly, it felt close to a real consultation (minus the waiting room), though in reality I am not replacing real doctor yet! What struck me is how powerful reverse prompting can be when we don’t have the expertise to ask the right questions ourselves. The AI guides the conversation, fills in the blanks, and leads us toward clarity. Maybe this is the future of working with AI: not just us prompting it, but it prompting us back — like a partner who knows what to ask, even when we don’t.

50

2 Comments

Funny enough, Silicon Valley, 11 years after the first episode aired, is still relevant today. The sarcastic jokes about the IT world now feels extremely relevant as we see the rise of code agents being released and adopted everyday. Today, many companies are cutting costs by laying off developers and only retaining a couple who rely heavily on these tools, accepting code changes without blinking twice, often at the expense of truly understanding what their code does. The result? We are likely heading toward a future where websites, mobile apps and their infrastructure are increasingly vulnerable. The overuse of code agents can lead to repeated, exploitable mistakes (like the "Vibe coded arcade" that leaked his own AWS credentials, costing thousands…), because when you don’t know what your code is actually doing, you can’t spot the flaws before someone more curious does.

6

Ever wondered how complex systems evolve over time? A must-read for fellow engineers from one of our Engineering Managers at Qantas Loyalty Ranjeeth (Ran)

19

Today I spoke with two frinds from Google AI and learned that the KPI for engineers at Google and many large tech companies has fundamentally changed. A key metric now is how many tokens you are able to use. The more tokens you use, the greater your potential contribution—one person can now accomplish the work that once required ten or even a hundred people. #HerPowerAI #Google #WorkModel #KPI

1

Mastering LLMs requires moving beyond prompt engineering into the mechanics of how models "think." Here is a technical breakdown from tokenization to local deployment. 1. Foundation: Tokenization Computers process numbers, not words. Tokenization converts text into numerical IDs. Modern models use sub-word units (e.g., "pre" or "ing") stored in a fixed vocab.json. A sentence like "I have a dream" becomes a sequence of unique IDs (e.g., [40, 464, 3317]) for computation. 2. Decoding Strategies LLMs predict the next token one-by-one based on probability. The strategy determines the selection: Greedy Search: Picks the highest-probability token at each step. It is fast but can become repetitive or miss better sequences. Beam Search: Tracks multiple potential paths (beams) simultaneously. It calculates cumulative probabilities across branches, allowing the model to "look ahead" for more coherent results. 3. Probabilities and Logits The model’s final layer outputs logits (raw scores), which are converted into probabilities via Softmax. Engineers use NetworkX or Matplotlib to graph these decision trees, visualizing why a model chose one specific token over others based on confidence scores. 4. Local Deployment with Ollama For privacy and cost efficiency, running models locally is essential. Ollama is the industry standard for managing local inference. Setup: Runs as a background engine. Hardware: Requires high VRAM GPUs (NVIDIA L4/A100) for matrix multiplications. CLI: Use ollama pull llama3.2 to download, run for chat, and show for architecture details. 5. Practical Workflow A standard pipeline follows four steps: Preprocessing: Tokenization via Hugging Face. Inference: Passing IDs to the model using Greedy or Beam search. Post-processing: Decoding IDs back into human-readable text. Deployment: Moving to a local API via Ollama for production. #GenerativeAI #LLM #MachineLearning #Python #Llama3 #Ollama #NLP

12

Podonos just raised $2.4 million in pre-seed funding, and the signal is loud and clear: the voice AI industry finally has someone measuring the sound, not just making the noise. Based in Los Gatos, Podonos is building the infrastructure layer that decides whether voice AI feels real, human, and ready for primetime. The industry is racing toward a projected $47.5 billion by 2034, but growth without standards is chaos. Podonos is making sure the hype actually holds up when you press play. At the center is ✦Soohyun Bae, PhD in Computer Engineering from Georgia Institute of Technology and Y Combinator W22 alumnus. His track record stretches from engineering roles at Google Maps to leading AR mapping at Niantic, plus co-founding Bobidi and TickTock AI. He has seen what happens when technology scales without proper #guardrails. With Podonos, he's betting the future of AI isn't just about how smart the models get, but how believable and trustworthy they sound. Podonos evaluates models across naturalness, similarity, emotion, recognition accuracy, pronunciation, tone, and resilience. Add in #personaconsistency, and suddenly you're not just testing machines, you're testing their ability to pass as human. The twist? They deliver results in under twelve hours, when the legacy approach drags on for two months. In AI time, that's the difference between relevance and irrelevance. Investors caught the beat. Serac Ventures, led by Kevin Moore, took the lead, backed by NAVER D2SF, the venture arm of South Korea's NAVER, and KAIST Venture Investment Holdings. Together, they've now backed Podonos twice, with this $2.4 million round stacking on top of a $750,000 raise in March 2025. Total funding sits at $3.15 million, and every dollar is pointed toward scaling #engineering, refining #analysistools, and launching into the U.S., Europe, and Southeast Asia. This isn't theory. Podonos already serves Resemble AI, Play AI, and Sanas AI, hitting six-digit ARR by November 2024 with just four employees. Their secret weapon is reach: 150,000 evaluators spread across nine languages and thirteen locales, forming a human-in-the-loop platform that blends speed, rigor, and global coverage. That reach lets Podonos deliver evaluations at a scale and pace that no competitor has matched. The strategy is ambitious: expand into high-demand sectors like #healthcare, #finance, #gaming, and #advertising, then move beyond voice into #multimodal model evaluation, spanning video AI and large language models. The mission isn't just to check if AI works, it's to decide if AI works for people. #Startups #StartupFunding #EarlyStage #VentureCapital #PreSeed #AI #AIInfrastructure #Voice #VoiceAI #VoiceTech #Infrastructure #Technology #Innovation #TechEcosystem #StartupEcosystem #Hiring #TechHiring If software engineering peace of mind is what you crave, Vention is your zen.

6

1 Comment

This is absolutely true. I would also add that just because AI decreases cycle time does not mean you should operate without a plan. To use an analogy, pretend you are going on a trip from New York to Los Angeles and fuel/electricity are free (or only cost pennies). Now imagine that you don't have a map but are just driving based on instinct. Because you are moving and the cost is low or zero, you feel you are getting somewhere even when you are not. You need a map (app or paper), or at least need to stop for directions occasionally. Movement, no matter the speed, is not progress unless it's toward where you want to go.

3

2 Comments

Senior developers often interact with LLMs as if they were very junior teammates. We provide guidance, review their output, and explain the architecture, much like we would with a new hire. The key difference, however, is a source of frustration: a junior developer learns and improves over time, eventually contributing their own ideas. An LLM doesn't—at least not yet. This is why their true value lies in tasks that don't rely on long-term learning. Think of them as a powerful, on-demand assistant for: * Summarizing documentation * Generating code examples * Clarifying complex concepts * Building quick prototypes * Finding very hard to spot bugs They can dramatically speed up development and even serve as a "living rubber duck" to help you talk through a problem. It's also important to note that model quality varies widely. The best models aren't always affordable. For instance Claude's Opus 4 is practically unusable with a Pro subscription. #LLMs #SoftwareDevelopment #SeniorDeveloper #AI

6