Carlos Serrano López

What else would you add to this cheat sheet on how to build AI Agents for data management? Here are 3 examples of DM Agents to start with: 1. DQ Triage Agent Inputs: table profile + Great Expectations results Actions: explain failures, propose SQL fixes, open Jira, assign steward, add dbt test 2. Glossary Curator Agent Inputs: column names, dashboards, docs Actions: suggest standardized terms, link synonyms, route to steward for approval, publish to catalog 3. Lineage Impact Agent Inputs: planned schema change Actions: trace downstream objects, estimate blast radius, notify owners, generate migration checklist Let's keep putting the Lights On Data and AI. Please feel free to share if you find it useful. -George Firican #data #datamanagement #AI

346

23 comentarios

AI Agents are the future 🤯 But no one is telling how to build for FREE in 2025. This cheatsheet will completely show you how to build Agents that will make you millions What is an Agent? → OpenAI: An LLM that manages workflows, decisions, and tools dynamically. → Anthropic: Systems where LLMs direct their own processes instead of just following code paths. When to Build an Agent? → Complex decision-making tasks → Hard-to-maintain rules → Heavy reliance on unstructured data Agent Design Foundations → Model: The LLM powering reasoning → Tools: APIs and external functions → Instructions: Rules guiding the agent’s behavior → Single-agent vs Multi-agent systems: Expand gradually or distribute tasks across specialized agents → Patterns: Prompt chaining, routing, parallelization, evaluators, and orchestrator-workers This cheat sheet is basically a blueprint for anyone building AI agents in 2025. 👉 Save this. You’ll need it. ♻️ Repost to share it with your network. #AIAgents #ArtificialIntelligence

275

93 comentarios

damnn… OpenAI, Google & Anthropic just dropped their internal AI playbooks for free • how to build AI agents • how to scale infra • how to prompt better than 99% of devs Get all the guides below with links for free (no login required) ↓ • Prompting 101 – Google’s no-nonsense intro → https://lnkd.in/gU7pfTEN • Build Agents That Work – Practical, not theoretical → https://lnkd.in/gH-xMd8t • Agent Coding Tips – Shortcuts from Anthropic’s engineers → https://lnkd.in/gNM9JRYd • Finding and Scaling Use Cases – A lens from OpenAI → https://lnkd.in/gCUNPsQ9 • Trust in AI – Deploy responsibly, especially in enterprise → https://lnkd.in/gqvcab7Y • AI at Scale – OpenAI’s take on enterprise integration → https://lnkd.in/gpxdZSkU • Agents Companion – Google’s toolkit for advanced builders → https://lnkd.in/gJ59YrH4 • Prompt Engineering – More signal, less guesswork → https://lnkd.in/gcVbA6S8 Save this. Send it to someone just getting started. Enjoy.

1990

154 comentarios

AI Agents and Agentic AI are not the same. Your top competitors already know this:  There are 3 types of AI workflows worth knowing,  And each one serves a different purpose. If you don't understand them, you're falling behind...  (While your competitors are building autonomous systems that run 24/7.) Here's how to tell them apart: 🤖 Non-Agentic AI ↳ Basic prompt-response AI with no memory/reasoning. ⚙️ How it works: - User types a single prompt, model processes  - Generates a direct response immediately  - Interaction ends with no context retention ✅ Pros: ↳ Fast, cheap, and universally accessible ↳ Requires no technical build or integration ↳ Great for clear, one-off tasks 🚫 Cons: ↳ No reasoning or context awareness ↳ Quality depends heavily on prompts ↳ Struggles with multi-step work 📈 How to build fast: ↳ Open ChatGPT, Claude, or Gemini ↳ Write clear, specific prompts with context ↳ Copy output, edit as needed, and repeat 🧠 Agentic AI ↳ Self-managing AI system that can plan and execute. ⚙️ How it works: - The system breaks your goal into sub-tasks - Connects to tools, APIs, data sources  - Evaluates results with feedback loops ✅ Pros: ↳ Handles complex, changing projects  ↳ Can integrate with tools and databases ↳ Produces more reliable outcomes 🚫 Cons: ↳ Slower and more expensive than agents ↳ Requires human oversight and guardrails ↳ Risk of overbuilding for simple problems 📈 How to build fast: ↳ Use LangChain, CrewAI, or AutoGen frameworks ↳ Assign roles: Planner, Executor, Critic  ↳ Add memory and feedback layers for consistency ⚡ AI Agent ↳ Single-task AI worker designed to automate one task. ⚙️ How it works: - User defines one clear responsibility - Agent receives inputs, accesses connected tools - Executes and outputs without guidance ✅ Pros: ↳ Automates repetitive, time-consuming tasks ↳ Quick setup and cost-efficient ↳ Easy to test and refine within roles 🚫 Cons: ↳ It can't handle broader projects  ↳ Breaks if inputs are unclear or data is missing ↳ Needs orchestration to work alongside other agents 📈 How to build fast: ↳ Define one specific, repetitive task for automation ↳ Use Zapier, LangChain, or OpenAI Assistants API ↳ Connect required inputs/outputs, test before deploying TLDR; AI Agents = Single-task automation Agentic AI = Multi-step problem solving Most businesses need both, but they're building neither. The dominant businesses will master all 3, and build accordingly. How are you using AI for business?  Drop your answer in the comments. I'm building an Agentic AI called Searchable.  It analyses and optimises your site automatically for SEO/AEO. Learn more and join the waitlist:   https://lnkd.in/epgXyFmi -- 📌 If you want a high-res PDF of this sheet: 1. Follow Chris Donnelly 2. Like the post 3. Repost to your network 4. Subscribe to: https://lnkd.in/eUTCQTWb

2454

556 comentarios

Forget expensive degrees.   Here are Top 10 YouTube Channels to Learn AI & Machine Learning in 2025:   1.     Andrej Karpathy Ideal for: Developers interested in building LLMs and understanding neural networks from scratch.  🔗https://lnkd.in/dWrRGRw8   2.     sentdex Ideal for: Python enthusiasts aiming to apply programming skills in AI, ML, and data analysis.  🔗https://lnkd.in/ddMx-Xuq   3.     Sebastian Raschka Ideal for: Learners seeking in-depth tutorials on deep learning and PyTorch implementations.  🔗https://lnkd.in/dnPkGCD5 4.     Jeremy Howard Ideal for: Coders looking for practical deep learning applications and fast.ai course materials.  🔗https://lnkd.in/dbEZiXHE   5.     MIT OpenCourseWare Ideal for: Students desiring comprehensive academic courses on AI and machine learning topics.  🔗 https://lnkd.in/dCikzSHZ   6.     Stanford Online Ideal for: Professionals seeking structured learning from Stanford's AI and ML course offerings.  🔗https://lnkd.in/dUKAXVNW 7.     StatQuest with Josh Starmer Ideal for: Beginners wanting clear and engaging explanations of statistics and machine learning concepts.  🔗 https://lnkd.in/dyVUTx5c 8.     3Blue1Brown Ideal for: Visual learners interested in mathematical concepts underlying AI and ML algorithms.  🔗 https://lnkd.in/dhEf95aD 9.     Krish Naik Ideal for: Aspiring data scientists looking for practical insights into AI and ML industry applications.  🔗 https://lnkd.in/dFAPMyqW 10. CampusX Ideal for: Learners seeking structured tutorials and courses on AI, ML, and data science topics.  🔗 https://lnkd.in/dTgJqqZ7 More resources in the comments.   Repost this to help others in your network.

477

31 comentarios

𝐆𝐨𝐨𝐠𝐥𝐞'𝐬 𝐍𝐞𝐰 𝐀𝐈 𝐒𝐞𝐚𝐫𝐜𝐡: 𝐖𝐞𝐛 𝐆𝐮𝐢𝐝𝐞. I looked under the hood of Google's new AI-powered "𝐖𝐞𝐛 𝐆𝐮𝐢𝐝𝐞" in Search Labs. By analyzing the network traffic (HAR file), I was able to map out how it's moving beyond a simple list of links into 𝐚 𝐩𝐞𝐫𝐬𝐨𝐧𝐚𝐥𝐢𝐳𝐞𝐝 𝐞𝐱𝐩𝐞𝐫𝐢𝐞𝐧𝐜𝐞. Here are a few findings: 1️⃣ 𝐓𝐡𝐞 "𝐐𝐮𝐞𝐫𝐲 𝐅𝐚𝐧-𝐎𝐮𝐭" 𝐢𝐬 𝐒𝐞𝐫𝐯𝐞𝐫-𝐒𝐢𝐝𝐞 (like for AI Mode): Your broad query (like "tips for getting started with photography") isn't treated as one search. Google's AI breaks it down into multiple, smarter sub-queries on the backend (i.e., "camera basics," "composition techniques," "lighting tips"). This happens before the page even loads. 2️⃣ 𝐏𝐞𝐫𝐬𝐨𝐧𝐚𝐥𝐢𝐳𝐚𝐭𝐢𝐨𝐧 𝐢𝐬 𝐭𝐡𝐞 𝐂𝐨𝐫𝐞 𝐄𝐧𝐠𝐢𝐧𝐞: This fan-out isn't generic. The data clearly shows it's personalized using multiple vectors: 👤 Your Account: Search history and interests directly influence the types of sub-queries generated. 📍 Your Location: It inferred my location from past activity to suggest local results like "photo classes in NYC." 📱 Your Device: It knows I'm on a Pixel phone and can tailor suggestions accordingly. 3️⃣ 𝐂𝐥𝐞𝐯𝐞𝐫 𝐀𝐫𝐜𝐡𝐢𝐭𝐞𝐜𝐭𝐮𝐫𝐞 𝐟𝐨𝐫 𝐚 𝐃𝐲𝐧𝐚𝐦𝐢𝐜 𝐔𝐗: The complex AI work and sub-query aggregation happens on Google's servers. The browser receives a lightweight page structure first, and then the detailed, categorized results are loaded in asynchronously. This keeps the initial experience fast while the AI-curated content populates. 𝐓𝐡𝐞 𝐓𝐚𝐤𝐞𝐚𝐰𝐚𝐲: Search is evolving from a list of blue links into a personalized experience. The change is profound for anyone doing SEO and marketing. 𝐓𝐡𝐢𝐬 𝐞𝐯𝐨𝐥𝐮𝐭𝐢𝐨𝐧 𝐟𝐫𝐨𝐦 𝐚 𝐥𝐢𝐬𝐭 𝐨𝐟 𝐛𝐥𝐮𝐞 𝐥𝐢𝐧𝐤𝐬 𝐭𝐨 𝐚𝐧 𝐀𝐈-𝐜𝐮𝐫𝐚𝐭𝐞𝐝 𝐤𝐧𝐨𝐰𝐥𝐞𝐝𝐠𝐞 𝐠𝐮𝐢𝐝𝐞 𝐬𝐢𝐠𝐧𝐚𝐥𝐬 𝐭𝐡𝐞 𝐞𝐧𝐝 𝐨𝐟 𝐭𝐡𝐞 𝐭𝐫𝐚𝐝𝐢𝐭𝐢𝐨𝐧𝐚𝐥 𝐫𝐚𝐧𝐤 𝐭𝐫𝐚𝐜𝐤𝐢𝐧𝐠 𝐦𝐢𝐧𝐝𝐬𝐞𝐭 (LLM tracking included). With hyper-personalized results that change based on user, location, and history, a single "rank" is becoming a phantom metric. Instead, the strategic focus must shift to topical clustering. The new goal isn't to be #1 for a keyword or a conversational query, but to be present across multiple, dynamically-generated topic clusters. This forces us to think in terms of occurrences and semantics: 𝐎𝐜𝐜𝐮𝐫𝐫𝐞𝐧𝐜𝐞𝐬: How often does our content appear across these different AI-generated topics? This is the new measure of visibility. 𝐒𝐞𝐦𝐚𝐧𝐭𝐢𝐜𝐬: Is our content semantically rich and comprehensive enough for the AI to recognize its authority and place it within these relevant clusters? This is a significant move from tracking static positions to measuring dynamic presence and proving true topical authority.

320

37 comentarios

Most people want to “use AI Agents.” Almost nobody knows where to start. Here’s the roadmap. ⬇️ 2025 isn’t the year of AI Agents. We’re entering the decade of AI Agents. Level 1: Foundations ↳ Learn transformers, tokens, embeddings. ↳ Know pre-training vs fine-tuning. Level 2 Prompting & Reasoning ↳ Use structured prompts, chain-of-thought. ↳ Tune parameters for reliability. Level 3: Retrieval (RAG) ↳ Store knowledge in vector DBs. ↳ Chunk info by meaning, not size. Level 4: Tools ↳ Connect LLMs to APIs/functions. ↳ Try LangChain, LangGraph, CrewAI. Level 5: Agents ↳ Start with task agents. ↳ Test loops (ReAct, Plan-Execute). Level 6: Memory ↳ Blend buffer + summary + entity memory. ↳ Carry context across sessions. Level 7: Multi-Agent Systems ↳ Assign roles: planner, executor, critic. ↳ Coordinate handoffs with guardrails. Level 8: Feedback & Eval ↳ Gather human + AI feedback. ↳ Use reward models to improve. Level 9: Safety & Protocols ↳ Apply guardrails + MCP standards. ↳ Keep logs + traceability. Level 10: Production ↳ Deploy with FastAPI/Gradio. ↳ Monitor cost, latency, drift. Nobody can learn all this overnight. But leaders don’t need details, they need the map. Save this roadmap. ♻️ Repost to help others stay ahead. ➕ Follow Gabriel Millien for AI + leadership insights. 📌Visual Roadmap credit: Aishwarya Srinivasan, give her a follow

170

66 comentarios