Real-Time Task Status Updates Using AI

🤖 MCP: What It Means For UX Designers. With practical guidelines on what it means and what designers now can do in AI products ↓ 🚫 LLMs can’t access real-time data without re-training. ↳ E.g. LLMs can’t answer "What’s the weather is today?". ✅ When a user asks it, RAG retrieves “fresh” context. ✅ It adds context to user’s query, sends it back to LLM. ↳ E.g. "What’s the weather is today?" + "Chicago, 54 F" 🤔 RAG is retrieval-only, can’t trigger actions or workflows. ✅ MCP gives AI real-time access to tools, data, actions. ✅ Any product can set up an MCP server for ChatGPT etc. ✅ It describes available tools, data sources, internal tasks. ↳ E.g. update calendar, send email, add a record, import. ✅ MCP = instruction manual telling LLMs how to use tools. ✅ User sends a query → AI looks up if any tool is a match. ✅ If needed, AI asks for permission to access an MCP server. ✅ It reads instructions, then triggers an action based on query. ✅ Users can access your tools via AI systems of their choice. MCP (Model Context Protocol) sounds like a merely technical feature that gives AI access to tools, actions and live data streams in your product. But what it actually provides is a way to integrate your products in any AI product that a user chooses to use. As Addy Osmani noted, for example, with Zapier MCP, an AI agent can perform any action that Zapier supports, from sending Slack messages, creating Google Calendar events, updating CRM records, to initiating e-commerce orders. And that’s what allows for fast automation with a pipeline of AI agents. If you are selling products, ChatGPT or Claude could filter, sort and showcase your products as users ask for them. With MCP, AI could access Jira, Notion and GitHub to provide real-time status updates. For sensitive data, users could access their private data with established privacy guardrails and access codes. AI agents could break down a complex task like booking a ticket into a series of small tasks and complete them by one, without ever visiting a website at all — if the platform provides an MCP server with access to that tool. As we added URLs for search engines to crawl, now we can add “features” for AI to use. This also opens the door for fine-grained personalization and customization. But it also requires transparency and control over how user’s data and user’s queries are flowing between AI agents and tools. And for designers, it probably means more interactions outside of the UI, and more integrations with AI. Meet the world of MCP-automated workflows — e.g. from Figma to code, design system maintenance and quick prototyping. That’s a quite profound change — and a change that might make user experience blazingly fast and perfectly seamless for users — and often without UI interactions at all. (Useful resources in the comments below ↓)

I burned $47 in API calls watching an agent check the same API endpoint every 30 seconds for 6 hours straight. It was supposed to monitor a deployment status. Instead, it just… kept checking. No breaks. No strategy. Just pure, expensive anxiety. That's the problem with current AI agents: they're brilliant at one-time tasks but absolutely terrible at waiting. Microsoft Research just released something that fixes this: SentinelStep. It's now open-sourced in their Magentic-UI system, and honestly, this changes how we think about agent workflows. Here's what makes it work: The system breaks monitoring into three components: actions (what to check), conditions (when to stop), and polling intervals (how often to check). Simple concept, but the execution is clever. Dynamic polling is where it gets interesting. The agent doesn't blindly check every minute. It makes an educated guess based on task urgency. Monitoring quarterly earnings? Less frequent checks. Tracking an urgent email? More aggressive polling. Then it adjusts based on observed patterns. Now, here's my take on what's probably happening behind the scenes: The system likely maintains a state snapshot after the first check, basically freezing what the agent knows at that moment. Think of it like taking a photo of the agent's brain. For each subsequent check, instead of carrying forward the entire conversation history (which would expand the context window), it loads the frozen snapshot, performs the new check, compares the results, and determines whether the condition is met. The polling adjustment probably uses something straightforward maybe exponential backoff with task-specific multipliers. If nothing changes after a few checks, wait longer next time. If patterns emerge (like "emails usually arrive between 9-11 AM"), the interval shrinks during those windows. No fancy ML needed, just sensible heuristics. Context management is the real win here. Without it, a 2-day monitoring task would accumulate thousands of tokens of redundant checks. With state snapshots, each check stays isolated and lightweight. They tested it with SentinelBench and showed success rates jumping from 5.6% to 33-39% for 1-2 hour tasks. But here's what I think matters more than those numbers: where you'd actually use this. Imagine monitoring CI/CD pipelines that take hours to complete, tracking competitor pricing that updates sporadically, or watching for specific social media mentions across days. These aren't hypothetical—they're tasks we currently handle with clunky cron jobs or manual checking. pip install magentic-ui right now and start experimenting. The foundation is solid, though you'll want to test thoroughly for production use cases (Microsoft's transparency note calls this out explicitly). This feels like one of those unglamorous infrastructure pieces that quietly enable a whole new category of automation. Not flashy, but exactly what we needed.

With the new #Microsoft #Agent #Framework, developers can now build AI agents that handle complex, multi-step workflows with persistent state and structured execution. 🕒 A key challenge is supporting long-running operations, like multi-turn reasoning, tool integration, and maintaining conversation context. Traditional web apps face issues such as HTTP timeouts, connection drops, and scalability limits when handling these advanced AI workflows. ⚡ Azure #App #Service addresses these challenges by enabling an asynchronous request-reply pattern. The API responds immediately with a task ID, while background workers handle the heavy processing. Task status and results are tracked in Azure Cosmos DB, and Azure Service Bus ensures reliable message delivery. This approach avoids timeouts, supports real-time progress tracking, and scales efficiently. ✈️ Blog features an AI Travel Planner application built with the Agent Framework and hosted on Azure App Service. The solution combines REST API and background worker in a single deployment, uses Azure AI Foundry for persistent agents and threads, and demonstrates how to inspect agent behavior directly in the Azure portal. ✨ This architecture supports rapid innovation, easy updates, and integration of new AI capabilities, making it suitable for production-ready AI applications that require reliability and scalability. 🔗 Blog: https://lnkd.in/dKxBvQQG 🔗 Microsoft Agent Framework: https://lnkd.in/d3XDQ3-7 🔗 Azure App Service: https://lnkd.in/dxHgUb3k 🔗 Sample Application: https://lnkd.in/dUJqzhU2 #Azure #AI #AzureAI #AppService #AzureAIFoundry #AgentFramework

We’re seeing a fundamental shift in how people use software. Up until now, we’d brainstorm in one place with AI, then do the manual work of turning those ideas into projects somewhere else. With the new Asana app in Anthropic Claude, that gap closes. Your chat becomes an execution layer where you can get live status updates, surface blockers, and see what’s behind schedule. Then, with a click, you pull those insights from your conversation into a new Asana project, along with a clear roadmap that follows the decision framework and addresses the immediate priorities surfaced in your status overview. Asana is the context and institutional memory for AI; by grounding Claude in Asana Work Graph data, we ensure the model has context like specific company policies and knowledge—rather than just "guessing" in a vacuum—so conversations shift to action. In the demo video, you can see how: - Priority levels are assigned based on urgency and framework guidance  - Critical tasks like executive approval are clearly marked, with immediate items captured in a dedicated section  - Tasks are assigned to you where you’re the clear owner, with others left unassigned for team routing.

🚛 𝗪𝗵𝗮𝘁 𝗶𝗳 𝗳𝗹𝗲𝗲𝘁 𝗺𝗮𝗻𝗮𝗴𝗲𝗿𝘀 𝗰𝗼𝘂𝗹𝗱 𝘀𝗰𝗵𝗲𝗱𝘂𝗹𝗲 𝘁𝗵𝗲𝗶𝗿 𝗿𝗲𝗽𝗮𝗶𝗿𝘀 𝗯𝗲𝗳𝗼𝗿𝗲 𝘁𝗵𝗲𝘆 𝗯𝗿𝗲𝗮𝗸 𝗱𝗼𝘄𝗻? That’s exactly what this AI-powered, agent-based workflow does. Here's how it works: 1️⃣ Sensors on vehicles detect change (like low oil pressure). The signal is sent instantly through: • 𝗞𝗮𝗳𝗸𝗮 (best for huge fleets, high-speed data) • 𝗘𝘃𝗲𝗻𝘁𝗕𝗿𝗶𝗱𝗴𝗲 (ideal for smaller-scale, simpler systems) This kicks off the workflow. 2️⃣ An event processing system calls the 𝗔𝗻𝗼𝗺𝗮𝗹𝘆 𝗗𝗲𝘁𝗲𝗰𝘁𝗶𝗼𝗻 (𝗠𝗟) system that triggers an anomaly. It predicts a likely failure. ✅ This is an 𝗠𝗟 𝗔𝗴𝗲𝗻𝘁 - it doesn’t use language, just pattern recognition. 3️⃣ The workflow gets activate. It triggers the 𝗦𝘂𝗽𝗲𝗿𝘃𝗶𝘀𝗼𝗿 𝗮𝗴𝗲𝗻𝘁 (LLM agent). It logs the issue and starts tracking the workflow. The supervisor agent managed the workflow, co-ordinatinf different agents and updating state. ✅ All task status and decisions are saved in 𝗗𝘆𝗻𝗮𝗺𝗼𝗗𝗕 → This keeps track of what’s done, what’s pending, and what’s next. 4️⃣ A 𝗣𝗹𝗮𝗻𝗻𝗲𝗿 𝗮𝗴𝗲𝗻𝘁 (LLM agent) checks the truck’s schedule. It finds a downtime slot for repairs that won’t affect deliveries. ✅ This is an 𝗟𝗟𝗠 𝗔𝗴𝗲𝗻𝘁. It talks to scheduling tools and updates the state in 𝗗𝘆𝗻𝗮𝗺𝗼𝗗𝗕. 5️⃣ A 𝗣𝗿𝗼𝗰𝘂𝗿𝗲𝗺𝗲𝗻𝘁 𝗮𝗴𝗲𝗻𝘁 (LLM Agent) checks what part is needed, finds a vendor, and places the order. It uses secure APIs + keys from 𝗦𝗲𝗰𝗿𝗲𝘁𝘀 𝗠𝗮𝗻𝗮𝗴𝗲𝗿. ✅ Order status is also saved in 𝗗𝘆𝗻𝗮𝗺𝗼𝗗𝗕 for tracking. 6️⃣ The system waits until the part arrives or the technician scans the truck to begin the repair. 7️⃣ A 𝘁𝗲𝗰𝗵𝗻𝗶𝗰𝗶𝗮𝗻 𝗮𝘀𝘀𝗶𝘀𝘁𝗮𝗻𝘁 𝗮𝗴𝗲𝗻𝘁 (LLM agent) pulls up step-by-step instructions. Helps the technician through the process. ✅ It uses 𝗥𝗔𝗚 (𝗿𝗲𝘁𝗿𝗶𝗲𝘃𝗮𝗹-𝗮𝘂𝗴𝗺𝗲𝗻𝘁𝗲𝗱 𝗴𝗲𝗻𝗲𝗿𝗮𝘁𝗶𝗼𝗻) to search the knowledge base → This could be 𝗢𝗽𝗲𝗻𝗦𝗲𝗮𝗿𝗰𝗵, 𝗪𝗲𝗮𝘃𝗶𝗮𝘁𝗲, or another vector DB with SOPs and repair logs. 8️⃣ Once the fix is complete, a final agent updates the system. ✅ It marks the job done in 𝗗𝘆𝗻𝗮𝗺𝗼𝗗𝗕 ✅ Sends an alert via 𝗦𝗡𝗦 𝗼𝗿 𝗙𝗶𝗿𝗲𝗯𝗮𝘀𝗲 Vehicle is back on the road. This can be done at scale for the whole fleet. 💡 𝗕𝘂𝘀𝗶𝗻𝗲𝘀𝘀 𝗯𝗲𝗻𝗲𝗳𝗶𝘁𝘀: • 📉 Less downtime • 🕑 Better scheduling • 🔧 Faster repairs • 💰 Lower costs • 📊 Smarter operations    ⚠️ 𝗡𝗼𝘁𝗲: This is a 𝗰𝗼𝗻𝗰𝗲𝗽𝘁𝘂𝗮𝗹 𝗱𝗲𝘀𝗶𝗴𝗻, not a ready-to-deploy solution. But it shows what’s possible when 𝗠𝗟 + 𝗟𝗟𝗠 𝗮𝗴𝗲𝗻𝘁𝘀 work together, maintaining state in a database, not just to chat, but to take action. How would you improve this solution? What other use cases do you think it could address? 👇 Let’s talk. #AgenticAI #FleetTech #MLops #RAG #OpenSearch #AgentBuild

I've been running 23 automated AI tasks since January. They process my inbox at 6 AM. Prep my client meetings by 8 AM. Ping me on Telegram when something needs attention. Run health checks across my entire client portfolio. All on autopilot. While I sleep. Two months ago, I showed this system to my newsletter audience. The most common response was some version of: That's incredible. But I can't build that. Fair. My setup took months. Custom PowerShell scripts. Windows Task Scheduler configurations. MCP server integrations. A safe-execution wrapper to kill zombie processes. Not exactly beginner-friendly. This month, Claude shipped 16 product updates in 21 days. 4 of them just made my custom setup available to everyone. No scripts. No configurations. No technical background required. Here's what's now possible: Upload 3,000 pages into one conversation. A full OM, five years of financials, rent rolls, comps, and your underwriting assumptions. All at once. No splitting. No losing context. Schedule AI tasks to run on autopilot. Deal screening at 6 AM. Portfolio rent roll checks every Monday. Lease expiration alerts daily. Set it once, walk away. Control your AI from your phone. Start an analysis at your desk, redirect it from a property tour, get results in the parking lot. One continuous thread that never resets. Get texted when it's done. Your AI finishes a 200-page OM analysis at 2 AM? Telegram ping with the summary. No more checking back every 20 minutes. Claude didn't get smarter this month. It got easier. The technical barrier that separated Level 1 from Level 3 just disappeared. I wrote the full breakdown in this week's newsletter - what each feature does, how to set it up for CRE, and a copy-paste prompt for every use case.