We Need to Talk About Context

The DM from my boss was waiting when I woke up Monday, after a weekend away where, for the first time in a while, I hadn't thought about work at all.

What did he mean? I felt a jolt of panic, and under it some curiosity, even a little hope. Things at the company were such that the message could be read a dozen ways. Half of us work Sundays, so a Sunday-night DM wasn't odd on its own. I just had no idea what had set it off. Monday was wall to wall with meetings, and the soonest I could get him was Tuesday. So I spent 36 hours managing the low hum of anxiety at the back of my skull, all because I had one sentence and no context to read it against.

Context is critical

I remember hearing Robert Plant sing that line about words having two meanings, back as a teenager, and thinking it was profound. Thirty-some years later I've spent more hours than I'd like to admit chewing on that one idle lyric.

Context shapes how we use language to make sense of the world. We pull it from everywhere: our five senses, our memories, the cultural cues around us. So when something breaks, we do the natural thing for a problem-solving animal and ask "what changed?" Doesn't matter if it's a car, a relationship, a recipe, or the computing systems holding up most of modern life. The question is always the same.

AI is context devoid at rest

If you didn't see the AI turn coming, the joke's on you. It's 2026 and this is a vendor blog (spoiler: the DM was positive, I still have a job), so of course AI is the whole point.

You've used AI by now. You know it's garbage in, garbage out, and that the prompt is both an art and a science. You also know you can hand your agent skills, memory, and the rest of the RAG toolkit to steer it toward sharper, more specific answers. All of that is world-building. You're handing the model an environment rich enough that it can work out which meaning you had in mind, and surface the sideways details that turn out to matter.

But the model only has that environment because you gave it one. On its own, an LLM is context devoid at rest. It carries no state specific to your world, your systems, or your last incident. It's brilliant and completely blank at the same time, waiting for you to ground it in something real.

AI is very good at writing code. It's decent at using code to investigate problems too. Code has one thing going for it: it's precise, where human language usually isn't. But code is only one face of a product. There's the system it runs on. There's the intent behind the code, argued out in conversation by the people who wrote it. There's the business reasoning that shaped the approach, and the trail of what's already been tried and failed.

That last one stings. If I were on stage I'd ask for a show of hands: who's watched an agent confidently attempt the exact thing you ruled out an hour ago? I'd set the over/under at 75%.

Whether you're chasing root cause or just trying to make something faster, all of that context is critical. It's the difference between crossing the country by horse and buggy or by plane.

groundcover is context

groundcover understood the weight of context long before the AI wave. Organizing telemetry around the workload mattered to us from day one, so the eBPF sensor captures it at the source and enriches it on the way into the backend: ClickHouse for traces and logs, VictoriaMetrics for metrics, all of it running on your own cloud. We tie deploy markers and change events to traces and logs at the workload level, and connect them to metrics through consistent, low-cardinality labeling. None of that is something you configure. The context is wired up out of the box, and it still is.

That's why the new agentic experience on the platform, groundcover Agent Mode, is so rich. The agent reasons over unsampled data by default, sampled intelligently only when you decide to, covering application telemetry, infrastructure telemetry, and business events in one place, on infrastructure you own.

It makes our agent genuinely good at reasoning over observability data. But telemetry is only part of the story. The answer to "what changed?" usually lives somewhere the platform can't see.

Introducing Connectors

So what if the agent could see the rest of the story too?

What if it could ask Linear which change shipped right before the latency spike? What if it could read the Slack thread where R&D worked out why they built the feature that way? What if it could read the code itself? Once the agent sees what you see, it stops guessing at why a problem exists and starts contributing to the fix.

That's what Connectors do. We built a connector ecosystem for your groundcover agent that reaches any MCP server, Slack, and your coding agents to pull in that context. Then we put the whole thing in Slack. The place teams already work.

You don't have to click the alert, jump to groundcover, and start an investigation anymore. You invoke the bot on the alert right there in the channel. It pulls the analysis, reaches into the connected systems for whatever context sharpens the diagnosis, and opens a PR with a proposed fix. All of it happens in the war-room thread where your on-call team is already talking through the incident, and the agent keeps the incident record as it goes.

Here's how that plays out. p95 latency on your checkout service jumps at 2:14, and the alert lands in your incident channel. You tag the bot. It pulls the latency breakdown from groundcover and sees the regression is isolated to one workload. It checks Linear and finds a change to that service merged and rolled out at 2:09. It reads the linked Slack thread and sees R&D swapped a synchronous call for one that fans out under load, plus the reasoning for why they thought it was safe. It opens a PR backing out the risky path and posts the why back in the channel. Five minutes, and you never left the conversation.

Remember that Sunday-night DM? I burned 36 hours on it because I had one sentence and nothing to read it against. No context, no idea what changed, just the fragment and the anxiety behind it. That's an agent staring at a telemetry spike with nothing else to go on. It has the symptom and nothing to reason against.

Connectors hand it the rest: the ticket, the thread, the deploy, the code. You stop stitching those together by hand and start directing the work. You're the architect now, not the record keeper.

Connectors launch and are generally available on June 29th, 2026. There’s a webinar on the connector release where you can ask groundcover product leaders your questions about exactly what it can power.