I Gave My AI Agent Access to Everything

Añadido: 2026-05-15

[00:00:00]In this video, I want to test something that becomes more obvious the better AI coding tools get. Because right now, writing code is honestly not the hard part anymore. With tools like Cursor, Cloud Code, Wing Surf, Lava Bowl, or V 0, you can go from idea to working feature insanely fast. But most engineering work does not actually start in your editor. It starts in your messaging app like Slack. A customer reports a bug, someone else links a ticket and remembers a decision from 3 months ago, another person points to a runbook in Notion, and before anyone can fix anything, an engineer has to reconstruct the context from Slack, GitHub, Linear, Docs, monitoring tools, and old conversations. And this is the part most AI coding tools still do not really solve. They can write code once you explain the task, but they usually do not know why the task exists, what triggered ticket explains it, or what decision shaped the system in the first place. So, even though coding is getting faster, context is still painfully manual. Recently, I came across Code Rabbit Agent and they were kind enough to partner with the channel for this video. You might already know Code Rabbit because they're well-known for AI code reviews, but Code Rabbit Agent is different. Instead of being another AI assistant you open separately, it lives inside Slack [music] and connects to the tools your engineering team already uses like GitHub, Jira, Sentry, and even AWS.

[00:01:39]Basically, the idea is that it becomes a second brain for your engineering team.

[00:01:44]Not just something that writes code, but something that understands the context around the code. Cursor helps you work inside the codebase. [music] Code Rabbit Agent helps the team move from conversation to investigation to PR to follow up losing the context in between.

[00:01:59][music] So, let's test it with a realistic workflow. Imagine users are reporting that checkout is suddenly extremely slow. Normally, someone checks Sentry, looks through recent GitHub changes, opens the linear ticket, searches Notion for runbook, and asks in Slack if anyone remembers why checkout was configured this way. And only after that can someone actually fix the issue.

[00:02:22]That is the hidden tax of engineering work. It is not always the code. It is the context. So, I'm going to mention Code Rabbit directly in the Slack thread. @CodeRabbit, investigate why checkout latency increased. Check recent code changes, related tickets, monitoring data, and any docs connected to checkout. And this is where the difference becomes clear. Instead of copying a Slack message into an AI coding tool and manually explaining the entire system, Code Rabbit starts from the team's actual operating context. It can look at the repo, recent [music] PRs, linked tickets, docs, monitoring data, and the conversation happening in Slack. [music] So, the starting point is not a blank prompt. In this example, it finds that the latency spike started after recent pull request changed part of the checkout configuration. But, the useful part is not just finding [music] the file. The useful part is that it connected the alert, the PR, the ticket, and the docs, then explains the situation inside the same Slack thread where the team is already talking. That matters because real investigations are usually messy. Everyone has a different piece of the puzzle, and half the work is building one shared mental model before touching the code. Now that we know the likely cause, I'm asking it to create a fix. Open a PR that fixes the checkout latency issue, but keep the change scoped only to the configuration that caused the regression. Don't touch unrelated checkout logic. And this is important because in real engineering work, the goal is not just to make a change. It is to make the right change with the right scope. During an incident, you do not want an AI agent rewriting half the service. You want the smallest, safe fix. So, Code Rabbit opens a PR with a specific configuration change, links it back to the original context, [music] and explains why the change is scoped that way. At this point, it doesn't even feel like a chatbot generating code, but more like an engineering teammate that can follow the thread, investigate the system, and handle the connective work around the fix. And I think that is the main difference. Most AI coding tools are editor first. You go to the tool, explain the task, give it context, and then it writes code. Code Rabbit agent is more Slack first. The work starts where the conversation starts. And for teams, that matters because Slack is usually where problems appear, where decisions get made, and where context gets lost. After the PR is open, someone usually asks why this happened and what we should do to prevent it. Normally, that means more manual work. You go back through the thread, check the ticket, look at the monitoring timeline, write a summary, maybe create a postmortem, update linear, and add something to the runbook. So, I'm asking Code Rabbit to do that from the same thread. Summarize the root cause, create a linear ticket for the follow-up, and draft a short postmortem with timeline, impact, fix, and prevention steps. And this is easy to underestimate. Writing a postmortem is not hard. The reason people skip it is because it requires gathering all the context after everyone is already tired of the issue. But if the agent has followed the work from the beginning, the fix, the reasoning, and the follow-up can become part of the team's memory instead of disappearing into an old Slack thread. That matters because tribal knowledge is one of the biggest hidden problems in engineering teams.

[00:05:43]Every company has that one person who knows why the billing service is weird, why deployments have a special exception, or why a certain folder should never be touched. Code Rabbit can capture that context as the team works.

[00:05:54]So, future tasks do not always start from zero. Another important part are guardrails. Anytime an AI agent can interact with real engineering tools, you need scopes, permissions, access control, and visibility. Not every agent should touch every repo, and not every person should be able to run every action. Code Rabbit agent is built around scoped controls, so teams can limit what it can access, which tools it can use, where it can operate, and how usage is tracked. That part is not flashy, but it is the difference between a cool demo and something a real company could actually use. My honest impression is that Code Rabbit agent is not trying to be a cursor inside Slack. It feels more like an agentic workflow layer for engineering teams. And that is why the second brain positioning makes sense. It remembers what matters, connects information across tools, and makes sure the next person does not have to rediscover the same thing again. So, if your team lives in Slack and feels the pain of context being scattered across GitHub, Linear, Jira, Notion, Sentry, and everything else, Code Rabbit agent is probably worth checking out. I'll put the link in the description and the pinned comment. Anyways, thanks for watching and until next time, happy coding.