Eric D. Brown, DSc

Draco Ventures LLC • 7K followers

Most "AI agents" are just scripts with better marketing. I keep seeing posts about using Claude Code, MCP, n8n and now Clawdbot/Molt (or whatever its called today) to automate workflows. The funny thing is that you could do the same thing that most are doing with all these tools with a 50-line Python script hitting the Claude API. Or OpenAI...or any LLM endpoint. We've been able to script these workflows for years. Of course, you have to know python to do this, but maybe learning python is too hard? I don't know. Yes, the capability is real and agents will matter. But right now, a lot of what people call "agentic" is just sequential API calls dressed up in new terminology. If your "agent" follows a fixed path like "do A, then B, then C" that's not an agent. That's a script. Real agentic behavior means the system decides what to do next based on what it finds. It adapts and it handles the unexpected. Most of what I'm seeing doesn't do that. This matters because terminology shapes investment decisions. When everything gets labeled "agent," you lose the ability to evaluate what's actually new versus what's repackaged automation. Before you get excited about an agent demo, ask: could I build this with a cron job and an API key? Often the answer is yes. #AI #AIAgents #Automation #TechStrategy #AIReality

Bunty Shah

MSCI Inc. • 4K followers

🚀 Agentic AI Stress-Test: MCPMark Benchmark—What AI Architects Need to Know 🔎 Why It Matters: LLMs are fast evolving, but are they truly ready for the real world? MCPMark sets a bold new standard—stress-testing model-tool agents with tasks that mimic real, business-grade workflows. ✨ What's Unique in MCPMark? - 🛠 127 expert-designed tasks: Notion, GitHub, Filesystem, PostgreSQL & Playwright - 🔄 Intense CRUD operations: Beyond reads, every interaction that matters - 🧩 Automatic, programmatic checkers: No hand-waving—if your agent fails, you'll know 📉 Key Research Insights: - 🤖 Frontier LLMs still stumble—GPT-5-medium scored just ~52% on the hardest tasks - 🧠 Smarter tool calls matter—context and reasoning win, not brute-force - 🪄 Robustness and long-context still open challenges—weak points exposed in planning - 🌐 Local APIs outperform remote—data bottlenecks affect success 🚧 Why Should AI Developers Care? MCPMark shows exactly where your agentic stack fails in real scenarios. - Diagnose context/pruning issues - Guide agent and framework design - Know your model's true limits before shipping! 💻 Curious? Test it yourself: Explore the code on GitHub: 👉 https://lnkd.in/d7wSAAxi 💡 As an AI Architect, I see MCPMark as a launchpad for resilient, context-smart agents. But here's my challenge to the community: What new strategies or architectures will finally deliver reliable, multi-tool agents at scale? #AIResearch #LLM #MachineLearning #AIEngineering #DeepLearning #AgenticAI

Will Scardino

Accelerant • 8K followers

𝗥𝗲𝗽𝗹𝗶𝘁 𝗔𝗴𝗲𝗻𝘁 𝟯: 𝗧𝗮𝗸𝗶𝗻𝗴 𝗔𝘂𝘁𝗼𝗻𝗼𝗺𝘆 𝘁𝗼 𝘁𝗵𝗲 𝗡𝗲𝘅𝘁 𝗟𝗲𝘃𝗲𝗹 🔥 If you thought AI agents were cool before, Replit’s latest release, Agent 3, is a serious leap forward. It’s built to do more on its own — test, fix, build, automate — all with minimal supervision. Here’s what you should know👇 🔗 https://lnkd.in/eVXcsH6F 𝗪𝗵𝗮𝘁 𝗶𝘀 𝗔𝗴𝗲𝗻𝘁 𝟯? Agent 3 is the newest version of Replit’s AI agent that helps you move from idea → app more effortlessly. It doesn’t just suggest code; it builds more of the workflow for you, monitors its own work, tests, fixes, and can even create other agents/automations. 𝗞𝗲𝘆 𝗙𝗲𝗮𝘁𝘂𝗿𝗲𝘀 & 𝗨𝗽𝗴𝗿𝗮𝗱𝗲𝘀 • 𝗘𝘅𝘁𝗲𝗻𝗱𝗲𝗱 𝗔𝘂𝘁𝗼𝗻𝗼𝗺𝘆 / 𝗥𝘂𝗻𝘁𝗶𝗺𝗲: Agent 3 can run for up to ~200 minutes autonomously. That means fewer interruptions and more of the “set it and forget it” feel when you have well-defined tasks.  • 𝗔𝗽𝗽 𝗧𝗲𝘀𝘁𝗶𝗻𝗴 + 𝗦𝗲𝗹𝗳-𝗖𝗼𝗿𝗿𝗲𝗰𝘁𝗶𝗼𝗻: It periodically runs real tests in a browser: clicking through UIs, submitting forms, checking APIs, etc. If it finds bugs, it automatically fixes what it can and retests. • 𝗕𝘂𝗶𝗹𝗱 𝗠𝗼𝗱𝗲𝘀: Design-first or Full-stack: You can start by just prototyping the front end/design, or go straight into building full functionality. More flexibility depending on how well you know what you want. • 𝗔𝗴𝗲𝗻𝘁𝘀 & 𝗔𝘂𝘁𝗼𝗺𝗮𝘁𝗶𝗼𝗻𝘀: One of the cooler bits — Agent 3 can build other agents/automations. Think bots (Slack, Telegram), scheduled tasks like daily reports, or workflows integrating tools like Notion, Linear, etc. 𝗪𝗵𝘆 𝗜𝘁 𝗠𝗮𝘁𝘁𝗲𝗿𝘀 𝗦𝗮𝘃𝗲𝘀 𝗧𝗶𝗺𝗲 & 𝗥𝗲𝗱𝘂𝗰𝗲𝘀 𝗖𝗼𝗴𝗻𝗶𝘁𝗶𝘃𝗲 𝗟𝗼𝗮𝗱: Less babysitting. When the agent can test & fix on its own, you don’t need to constantly switch back and forth. 𝗟𝗼𝘄𝗲𝗿 𝗕𝗮𝗿𝗿𝗶𝗲𝗿 𝗳𝗼𝗿 𝗡𝗼𝗻-𝗖𝗼𝗱𝗲𝗿𝘀: If you’re more idea-driven than detail-driven, Agent 3 helps translate ideas into working apps with fewer technical steps. 𝗕𝗲𝘁𝘁𝗲𝗿 𝗔𝗽𝗽 𝗤𝘂𝗮𝗹𝗶𝘁𝘆: Built-in testing & iteration means more bugs caught earlier, more robust behavior out of the box. 𝗔𝘂𝘁𝗼𝗺𝗮𝘁𝗶𝗼𝗻 & 𝗦𝗰𝗮𝗹𝗶𝗻𝗴: With the ability to build agents/automations, you can delegate recurring tasks. Consider automating your workflow, not just building a single app. Thanks for reading ✨𝐀𝐈 𝐏𝐌 𝐁𝐘 𝐃𝐄𝐒𝐈𝐆𝐍. ♻️ Repost to help someone level up with AI. 🚀 Subscribe on Substack to become an Agentic AI PM.

Suryadeep Chatterjee

Insight Global • 3K followers

LLMs aren’t just tools. They’re users. And we need to start treating them like it. The way we build software is about to change—again. Not because of another framework or trend, but because AI agents (especially LLMs) are becoming real actors in our systems. Here’s what I have learned designing with them, not just around them: 🔹 Design for Partial Autonomy Think less “talking to a terminal,” more “Iron Man suit.” LLMs need structure. Apps should: Handle memory (they forget everything). Orchestrate multiple models, not just chat. Include GUIs for quick verification. Offer users an autonomy slider—tap to suggest, dial up to act. 🔹 Fast Generation → Fast Verification LLMs make mistakes. That’s not new. What matters is how fast we catch and correct them. Visual diffs beat long explanations. Keep changes small, auditable. Vague prompts = vague results. Guide users to be concrete. 🔹 Your Infra Isn’t Ready (Yet) LLMs can’t “click here.” They need: Markdown docs. Executable commands, not instructions. Ingestion tools that reshape your repo into something readable. Protocols built for agents, not humans. This isn’t theoretical—it’s the work happening right now. And it’s reshaping how we think about UI, infra, and even documentation. We’re not just adding AI to products. We’re building for a new kind of user. And it’s changing everything. We’ve rewritten software for mobile, cloud, and APIs. Now we rewrite it for AI. #AIProductDesign #LLMInfra #EngineeringLeadership #HumanInTheLoop #PartialAutonomy #FutureOfSoftware #AgentOps #LLMEngineering #TechLeadership