{"id":126235,"title":"Stacktrace - Give agents autonomy without losing control","tagline":"Real-time steering for AI agents","body":"**TL:DR;** Stacktrace supervises agents used by your employees and customers. It detects risks across security, infrastructure and cost and lets teams block unsafe actions, require approval or steer agents.\n\n<https://youtu.be/2FCDlUCL_KE>\n\n### **Why we built this**\n\nAgents are moving from tools employees use at the terminal to systems they delegate entire business workflows to. As agents take on longer horizon tasks and run unattended, they make decisions without someone reviewing them. \n\nThe same challenge extends to personal agents that customers use to interact with your business. These agents need enough access to get things done without doing anything the customer or business hasn’t approved.\n\n**Just because an agent has permission to do something doesn’t mean it should.**\n\nToday companies manage this risk by restricting what agents can access or which employees can use them. But these restrictions limit the impact agents can have. We think businesses should be able to give agents more autonomy, with visibility and controls into their actions. \n\nThat’s why we built Stacktrace.\n\n### **What Stacktrace does**\n\nStacktrace supervises agent sessions in real time, detects actions that violate user intent or company policy, and can block them, require approval, or steer the agent back on track.\n\n**See:** Understand what agents are doing, which tools they use, and what they cost. Detect sensitive data exposure, destructive infrastructure changes, and stuck runs\n\n**Control:** Set policies for plugins, MCP servers, and skills. Block unapproved or vulnerable components and apply token budgets across model vendors.\n\n**Steer:** Block unsafe actions before they execute, require approval, or steer the agent toward a safer way to finish the task. \n\nStacktrace extends this approach to customers’ agents like Muse and Instinct, interacting with your business. It sits in front of your customer-facing MCP server, enforces customer permissions, and requires explicit confirmation for consequential actions.\n\n### **Who we are**\n\n![uploaded image](/media/?type=post&id=126235&key=user_uploads/3627108/68b56ce9-7bfc-47af-acc8-5a46b300e64a)\n\nWe’ve spent our careers building infrastructure, security, and platforms for systems where failures have real consequences.\n\n[Vinod](https://www.linkedin.com/in/vinodkone) led Apache Mesos, the open-source infrastructure used by companies including Apple, SpaceX, Netflix, Uber, and Airbnb. He previously led engineering teams at Harness, Twilio, Twitter, and Mesosphere.\n\n[Micheal](https://www.linkedin.com/in/michealb) led infrastructure, security, and AI platforms at Airtable, and previously led developer platform and cloud infrastructure organizations at Databricks, Pinterest, and Twitter.\n\n[Prasanth](https://www.linkedin.com/in/prasanthganta) spent 15 years at Amazon, building foundational services across AWS and Alexa, and GenAI systems at Amazon Ads.\n\n### **Our ask**\n\nIf you run agents with access to sensitive data or production systems, we’d love to show you Stacktrace.\n\nBook a demo at <https://stacktrace.ai>","slug":"Wq3-stacktrace-give-agents-autonomy-without-losing-control","created_at":"2026-11-05T19:00:00.083Z","updated_at":"2026-11-17T09:10:27.226Z","total_vote_count":3,"url":"https://www.ycombinator.com/launches/Wq3-stacktrace-give-agents-autonomy-without-losing-control","share_image_url":"//bookface-static.ycombinator.com/assets/ycdc/yc-og-image-c440a0ad1dacfb86eeeb343717479cc54d256614449b4ef719977a0a451f8bc8.png","company":{"id":35096,"name":"Stacktrace","slug":"stacktrace","url":"https://stacktrace.ai","logo":"https://bookface-images.s3.amazonaws.com/small_logos/e79ba1ad9464ac22023bbed994e2299d2ae8addf.png","batch":"Fall 2026","industry":"B2B","tags":["DevSecOps","Cybersecurity","Infrastructure","AI"],"search_path":"https://bookface.ycombinator.com/company/35096"}}