Kacper Włodarczyk’s Post

I built an AI agent from scratch. No LangChain. No LangGraph. No CrewAI. Just Python, Gemini 2.5 Flash, and raw tool calling. Here's what I learned that no framework tutorial will teach you: 1. The agentic loop is embarrassingly simple Build messages → call LLM → if tool_call → execute → feed result back → repeat. That's it. Every framework is just a wrapper around this. Once you see it raw, you can never unsee it. 2. Frameworks hide your bugs from you When something breaks in LangChain, you're debugging the framework. When something breaks in raw Python, you're debugging your logic. Big difference. One makes you smarter. One makes you dependent. 3. Tool schema design is where agents actually fail The LLM doesn't call the wrong tool because it's dumb. It calls the wrong tool because your schema description was ambiguous. Write your tool descriptions like you're explaining them to a junior dev on their first day. Precise. No assumptions. 4. 50 lines of Python is enough to go to production My personal concierge agent — the one that lives on my portfolio, captures leads, and pings my phone instantly — is ~50 lines. No overhead. No magic. Just code I fully understand and can debug at 2am. 5. You should build one without a framework at least once Not because frameworks are bad. LangGraph is excellent. I'm using it next. But if you've never written the raw loop yourself, you're flying blind. You're trusting abstractions you don't understand. Build it raw first. Then use the framework. You'll use it 10x better. --- Full source code in the comments — ~50 lines, no magic, just the loop. Follow along if you're into agentic AI and building real things, not just demos. #AgenticAI #Python #BuildingInPublic #LLM #SoftwareEngineering

🚀 Just built my first LangGraph project and I'm genuinely excited about it! LangGraph lets you design AI workflows as graphs — where each step is just a plain Python function connected by edges. No magic, no black boxes. What I built: 🔹 A chatbot powered by GPT-4o-mini wired up through a LangGraph flow 🔹 A second bot with zero AI — just Python — to prove LangGraph isn't about LLMs, it's about structure 🔹 A visual diagram of the graph that renders automatically on every run The biggest insight? You don't need an LLM to use LangGraph. Once you understand the pattern — State → Node → Edge — swapping in a real model is trivial. Small project, big learning. If you're exploring AI agent frameworks, LangGraph is worth your time. 🔗 https://lnkd.in/gawNJMWA #LangGraph #LLM #AIAgents #Python

Today's topic is a tool combo breakdown focusing on three exciting combinations that can revolutionize your workflow and save you time. Whether it’s integrating Claude Code with Obsidian for a seamless knowledge management system or harnessing n8n combined with the Claude API to automate complex tasks, these tools offer specific benefits. Let's dive into one of our options: using Python along with the Claude API. This combo allows developers to leverage AI capabilities directly within their existing workflows. Here’s how you can set it up: 1. **Setup**: First, ensure you have Python installed on your machine. You'll also need n8n and Claude Code plugins for n8n. 2. **Write Your Script**: Start by writing a simple Python script that uses the Claude API to process text inputs. For example: ```python import n8n from claude import ClaudeAPI # Initialize Claude API cl = ClaudeAPI() # Function to get AI generated response def get_response(prompt): response = cl.get_completion(prompt) return response # Example usage of the function result = get_response("What's the weather like in New York today?") print(result) 3. **Integrate with Obsidian**: Next, you can integrate this script with Obsidian using n8n to automate tasks. This setup can save significant time and effort, reducing manual processing and allowing for more efficient workflows. Would you be interested in exploring further AI integration opportunities like this one? Let us know your thoughts or challenges in the comments below. #ClaudeCode #AIAutomation #AITools #BuildWithAI #loopfeedai

If your CI pipeline is slow, start by looking at how your test suite is being split. 𝗽𝘆𝘁𝗲𝘀𝘁-𝘀𝗽𝗹𝗶𝘁 is a pytest plugin that does one thing well: it splits your test suite into equally timed sub-suites, not equally sized ones. Most naive approaches split by test count. 𝗽𝘆𝘁𝗲𝘀𝘁-𝘀𝗽𝗹𝗶𝘁 stores actual execution times in a .test_durations file and uses that data to balance wall-clock time across groups. Run --store-durations once, commit the file, and your CI groups will finish at roughly the same time. New or renamed tests are handled gracefully by falling back to average durations. No need to re-run --store-durations after every change. 🔗 Link to repo: github(.)com/jerry-git/pytest-split --- ♻️ Found this useful? Share it with another builder. ➕ For daily practical AI and Python posts, follow Banias Baabe.

🚀 Why uv is replacing pip in modern Python workflows For years, pip has been the default tool for installing Python packages. It works—but it was never designed to handle today’s complexity around environments, reproducibility, and speed. That’s where uv comes in. --- 🔹 1. Speed that actually matters uv is written in Rust and is insanely fast—often 10–100x faster than pip. 👉 Example: Installing a heavy stack like pandas + numpy + scikit-learn - pip → noticeable wait time - uv → installs in seconds For data scientists and ML engineers, this alone is a game changer. --- 🔹 2. One tool instead of many With pip, you usually combine: - venv (for environments) - pip (for install) - pip-tools/poetry (for dependency management) 👉 uv replaces all of these in a single unified tool No more juggling multiple commands and tools. --- 🔹 3. Better dependency resolution pip can sometimes: - install conflicting versions - behave inconsistently across machines uv provides more reliable and deterministic installs, reducing “works on my machine” issues. --- 🔹 4. Built-in lockfiles (Reproducibility) uv generates lockfiles to ensure: - same versions - same environment - same results This is critical in: - ML experiments - production pipelines - team collaboration --- 🔹 5. Easy migration (Drop-in replacement) You don’t need to relearn everything. 👉 Same workflow: uv pip install numpy uv pip install -r requirements.txt So you get better performance without changing habits much. --- 🔹 6. Real-world workflow comparison 👉 Using pip: python -m venv env source env/bin/activate pip install -r requirements.txt 👉 Using uv: uv venv uv pip install -r requirements.txt Cleaner. Faster. Simpler. --- 💡 Final Thoughts pip isn’t “bad”—it’s just outdated for modern workflows. If you’re working in: - Data Science - AI/ML - Backend Python Switching to uv can save time, reduce friction, and improve reliability. --- ⚡ Bottom line: uv is not just an alternative—it’s an upgrade. #Python #DataScience #AI #MLOps #SoftwareEngineering #Developers #Productivity

Stop spending days reading code alone. 🚀 I built CodeBoard—an AI-Powered Repository Intelligence Platform that acts as your personal technical consultant. 🧠 Whether you are onboarding to a new project or auditing complex business logic, CodeBoard leverages Python AST parsing and Llama 3.3 (via Groq API) to give you instant insights. Key Highlights in this Demo: ✅ Real-time Logic Auditing: Identifying API handlers and function flows in seconds. ✅ Contextual Understanding: No more "hallucinations"—the AI understands the actual code structure. ✅ Onboarding Efficiency: Perfect for developers who want to skip the manual line-by-line review. Built with: Python, FastAPI, React, and Groq. 🛠️ Check out the demo below! 👇 #FullStack #AI #GenerativeAI #Python #WebDevelopment #OpenSource

I’ve been building an AI agent system from scratch (Jan 2026 – Present). Not relying on pre-built frameworks — designing and implementing the internals myself. Here’s what I’ve built so far: 🧠 ReAct agent loop (Reason → Act → Observe → Repeat) 💬 Conversation memory (sliding window + LLM summarization) ⚙️ Sandboxed Python execution inside the agent 🤖 Multi-agent pipeline (Planner → Executor → Synthesizer) 🔁 Self-healing agents (retry logic, critique agents, graceful degradation) 📊 Structured JSON outputs with LLM-as-a-Judge evaluation The most interesting part: When Python code fails, a “Critique Agent” reads the traceback, fixes the issue, and re-runs it automatically. When context grows too large, the system dynamically reroutes to a local Ollama instance — enabling load balancing and protecting API limits. All of this runs on the free tier using Groq. Still in progress: → Long-term memory with ChromaDB → Parallel tool execution → Human-in-the-loop approval workflows → End-to-end pipeline: research → write → review GitHub: https://lnkd.in/ev8e9kGQ #AI #LLM #AgentAI #Python #MachineLearning #SystemDesign #BuildInPublic

🚀 New open-source project: playwright-ai-pilot I'm building an AI-powered test automation framework using Playwright, Python, and Claude. Three AI pillars so far: 🔧 Self-healing locators — when a selector breaks, Claude finds an alternative automatically 🤖 AI test generation — point it at a URL and get a full pytest test file back 📋 AI test planning — give it a user story and get a structured test plan with risks and automation notes Work in progress — Windows/NTLM and MFA authentication mocking coming next. GitHub: https://lnkd.in/g7hiN6Fq Would love feedback from other automation engineers. What would you add? #TestAutomation #Playwright #Python #AI #SDET #OpenSource

🚀 Day 14 of My Generative & Agentic AI Journey! Today’s focus was on Functions in Python — one of the most important concepts for writing clean and reusable code. Here’s what I learned: 🔁 Avoiding Code Duplication: • Functions help reuse the same logic multiple times • Instead of writing the same code again and again, we can define it once and call it whenever needed 👉 Example use case: Creating a function to handle multiple chai orders by just passing different names and chai types. ⚙️ Handling Complex Tasks: • Broke down a big task into smaller functions • Combined multiple functions to complete a full workflow 👉 Example use case: Building a report system by separating steps like fetching data, filtering it, and summarizing it. 📖 Improving Readability: • Functions make code easier to read and understand • Logic becomes more structured and organized 👉 Example use case: Creating a function to calculate bills instead of writing calculations multiple times. 💰 Real-world Application: • Applied functions to process multiple values using loops 👉 Example use case: Adding VAT to a list of orders using a reusable function. 💡 Key takeaway: Functions help write cleaner, reusable, and scalable code — a must-have skill for real-world development. Moving one step closer to writing production-level Python code 🚀 #Day14 #Python #GenerativeAI #AgenticAI #LearningJourney #BuildInPublic

I’ve been using the chore/agent-friendly-repo-foundation branch(https://lnkd.in/gtypVfA9) in ExcelAlchemy as an experiment in agent-friendly engineering. The interesting part is not just asking an agent to write code. It’s shaping the repo so an agent can actually navigate it: clear docs, explicit architecture, stable public boundaries, linked plans, tests, invariants, and examples. That idea maps closely to OpenAI’s recent post on [harness engineering](https://lnkd.in/gtHiDZbU): humans define intent and constraints; agents execute inside a well-structured environment. ExcelAlchemy itself is a schema-driven Python library for typed Excel workflows with Pydantic: template generation, workbook validation, error mapping, locale-aware result workbooks, and storage-backed integration. This branch is our small testbed for a bigger question: not “can agents code?” but “can we build repos they can reliably understand?” #AgentEngineering #HarnessEngineering #DeveloperTools #RepositoryDesign #Python #OpenSource #Pydantic #ExcelAutomation