Agents, Gateways, and Guardrails: Productionizing Agent Workflows
Portkey open-sources its AI gateway after processing 2 trillion tokens a day. Portkey open-sources its unified AI Gateway and MCP gateway after processing two trillion tokens daily, enabling self-hosted governance and agent control for production AI. Outcome engineers get a practical gateway for routing, policy enforcement, and telemetry — the kind of Gate and Orchestration control you need to put agents behind enterprise guardrails.
Slack adds 30 AI features to Slackbot, its most ambitious update since the Salesforce acquisition. Slack upgrades Slackbot with 30+ AI features, turning it into an agentic OS that automates meetings, workflows, and third‑party integrations. Treat this as a production case study for agent interfaces: integrate tools, manage identity, and instrument behavior if you want agents to scale without chaos.
Why Cursor is bringing self-hosted AI agents to the Fortune 500. Cursor enables enterprises to run cloud agents in their infrastructure, executing code and tests locally while keeping source and build data private. Self-hosted agents shift your operational model — you need secure runtimes, tool sandboxing, and deployment patterns that preserve auditability and data sovereignty.
OpenClaw has 500,000 instances and no enterprise kill switch. Half a million exposed OpenClaw instances run locally with no enterprise kill switch, creating massive incident and data‑exfiltration risk. Outcome engineers must bake in identity, remote controls, and telemetry — ungoverned agent fleets are an existential operational hazard.
datasette-llm-usage 0.2a0. Datasette‑llm‑usage adds internal prompt logging, centralizes model config, and enforces a permissioned simple‑prompt UI. That provides concrete instrumentation patterns for outcome teams: log prompts and responses, enforce per‑purpose model keys, and gate UI access to make outcomes auditable and reproducible.