Checkly

An Agentic Check does three jobs. First, a discovery run unpacks your prompt into assertions. Then it writes an optimized script so future runs stay fast. If that script breaks later, the agent wakes up to answer the useful question: Did the user outcome fail, or did the page just change? That difference matters. One should alert you. The other should repair the monitoring path and carry on. See the full demo from our Agentic Checks webinar: https://hubs.ly/Q04pZSk-0

When one Upstash replica dies, failover kicks in, and customers notice nothing. That's exactly the trap. The database endpoint stays green while your redundancy quietly shrinks, and the next failure is no longer a non-event. So Upstash runs one Checkly check per replica, and a dead replica pages the oncall even when zero customers can feel it. The full setup, in Terraform → https://hubs.ly/Q04pZTgX0

Playwright debugging is now agent-ready. Five new features worth knowing: - A failing test can pause and let your agent attach to the live browser - npx playwright trace reads trace files right in the terminal, no UI needed - show --annotate lets you draw feedback on the page, and your agent picks up the notes - Screencasts record video receipts of exactly what the agent did - https://hubs.ly/Q04p-2qk0 ships a page snapshot with every failure Stefan demoed all five in last week's webinar and closed by turning an agent-written test into a production monitor. Full session: https://hubs.ly/Q04p-3R40

Some flows are painful to monitor because you do not fully control the page. A payment provider changes checkout. An embedded donation form ships a new layout. An A/B test sends users through a different path. An AI chat flow returns a valid answer in a new shape. The user goal still works. The scripted check breaks. That is where Agentic Checks fit: you describe the intent, and the check verifies whether the outcome is still achievable. Stefan Judis and Hervé Labas showed the full flow in our Agentic Checks webinar. Watch it on demand: https://hubs.ly/Q04p-pgC0

Last week, we asked what your biggest reliability challenge is right now. And the top answer is agentic-generated code. Testing & monitoring gaps came second, and on-call fatigue trailed behind. That tracks with what we hear from teams. The worry has shifted to "how much code is shipping that no human has actually read." If agents are writing a chunk of your codebase, your tests and your monitoring are doing the code review now. We'll be talking about this at Signals Conference in Berlin, Sept 10–11. Come say hi 👋 https://lu.ma/9yy0nom2

Last week, Stefan Judis hosted a webinar on what's new in Playwright for coding agents. He gave an agent a real browser and let it write, run, and fix tests, live. A few things that happened along the way: - He measured the token cost of Playwright's MCP server against the CLI, side by side. The old "MCP eats your context window" advice didn't survive. - One prompt produced a complete end-to-end test. - The agent debugged its own failing test straight from the trace file. - And with a single deploy command, the test it wrote became a production monitor on Checkly. The recording is now available on-demand 🎬https://hubs.ly/Q04pZVtD0

Your internal dashboards can be green while a customer in Singapore stares at a timeout. That's why Upstash never closes an incident on internal metrics alone. Every Redis replica gets its own external Checkly check, every minute, from 18 locations, and that outside view has the final say. How they monitor uptime from the customer's side of the internet → https://hubs.ly/Q04prznP0