A instrument I’ve been utilizing for some time, Roo Code, an extension for Visible Studio Code, just lately decided that obtained me interested by the place agentic software program growth is definitely headed.
Roo Code is an AI-powered coding extension for VS Code, and like related instruments within the house it allows you to hook up with an LLM and work with an AI assistant proper inside your growth surroundings. I’ve discovered it genuinely helpful for agentic coding workflows, so it was a little bit of a shock when the staff introduced they had been stopping energetic growth on the extension itself. Replace: it seems the staff has reversed this resolution and that growth actively continues. In any case I nonetheless discover it wonderful, and use it with Kimi 2.6 Cloud (through Ollama) as one in all my eight agentic coding assistants.
As a substitute, they’re doubling down on one thing referred to as Roo Distant.
It’s not out there but, they usually’re at present in a pre-prototype section, accumulating e mail addresses and gauging demand. The idea is intriguing although: slightly than requiring a separate coding surroundings, the brokers come to you, wherever you already work. Their wager, at the very least initially, is that place is Slack.
The premise is easy sufficient. You’re already in Slack with a mixture of people in these channels, and now you add brokers to the combo. You assign them duties, they go off and do issues, they usually come again and report. It’s a imaginative and prescient of the agentic future the place autonomous AI employees function alongside people in the identical collaborative areas we already inhabit.
I take into consideration this type of factor lots, and whereas I see the enchantment, I’m undecided I absolutely purchase the premise, at the very least not as a substitute for what now we have now.
The Observability Downside
Right here’s my difficulty with the Slack mannequin as a major growth surroundings: visibility.
As a developer, as a product supervisor, or as anybody enjoying an orchestration position, you want to have the ability to see what’s occurring. When an agent disappears into the workflow and comes again with a outcome, you’ve misplaced the thread. You don’t have any window into what it modified, what it touched, or what selections it made alongside the way in which. In some contexts that’s nice and also you simply need the end result, however in software program growth that opacity is commonly an issue.
That is exactly the place the IDE earns its maintain. It’s not “stunning” in the way in which an iPhone is gorgeous, nevertheless it’s stunning in the way in which a well-designed instrument is gorgeous: it does precisely what it’s speculated to do, which is talk state.
While you’re working in an IDE like VS Code or Cursor, every little thing is seen. The information you’re engaged on, the standing of your adjustments, the git graph with its whole model historical past, branching, diffing, all of it. When an agent makes a change contained in the IDE, you’ll be able to select to belief it and transfer on, or you’ll be able to drill down, examine precisely what modified, and confirm it did what you anticipated. That stage of observability is tough to duplicate in a chat-based interface like Slack. And for this reason I feel the Built-in Productiveness Surroundings (IPE), which is basically utilizing an IDE to do non-coding issues, is an thought with legs.
Cursor, by the way in which, is a residing demonstration of this. Its success speaks to how a lot builders nonetheless worth having a wealthy visible surroundings round their code.
Three Fashions, All Legitimate
So the place does this go away us? I feel we’re heading towards three distinct however complementary fashions for agentic growth, and all three will (and do) coexist.
Mannequin 1: The IDE. The normal growth surroundings, enhanced with AI. Brokers function inside your workspace, and also you preserve full observability over what they’re doing. VS Code, Cursor, and related instruments belong right here. This mannequin isn’t going away, and if something, it’s getting stronger.
Mannequin 2: The Slack Mannequin (Human-Agent Collaboration). Brokers working alongside people in shared communication platforms like Slack, Discord, and Telegram. The brokers take part within the stream of labor, tackle duties, and report again. It is a legitimate and genuinely helpful sample, particularly for workflows which can be much less about code inspection and extra about job execution and communication. Consider it because the “assembly room” for people and brokers.
Mannequin 3: Orchestrated Agentic Workflows (LangGraph, CrewAI). That is the place issues get extra refined. Instruments like LangGraph and CrewAI will let you construct hybrid workflows which can be half deterministic and half probabilistic. You outline the construction of how work ought to stream, however LLMs deal with the reasoning and technology inside that construction. It’s the mixture of the reliability of conventional programming with the flexibleness of language fashions.
These methods supply observability too, simply in a distinct kind than the IDE. You possibly can hint the workflow, see the place brokers had been within the pipeline, and examine inputs and outputs at every node. For analysis workflows, speculation testing, and multi-step information processing, this mannequin is exceptionally effectively suited. A thesis analysis pipeline, as an illustration, maps naturally onto this type of graph-based structure.
What This Means
The framing that brokers will merely exchange the IDE, that we’ll all simply work in Slack with our AI teammates, misses one thing necessary about how builders, and information employees extra broadly, really have to relate to the work.
It’s not nearly getting a outcome. It’s about understanding what occurred, having the ability to confirm it, study from it, and course-correct. That relationship between human and agent, the place the human stays an knowledgeable orchestrator slightly than a passive recipient, requires surfaces that help statement, not simply communication.
Roo Distant could discover its area of interest, and the Slack mannequin is actual and price taking significantly. However the IDE will stay central to how severe growth work will get finished, and the graph-based orchestration layer will develop in significance as workflows develop into extra advanced.
Three fashions, all rising, all legitimate. The fascinating query isn’t which one wins, it’s studying when to achieve for which one.
Clinton Stark is co-founder of Stark Insider and StarkMind, a human-AI collaboration laboratory. He covers expertise, movie, and the humanities from Silicon Valley.