I finished working one coding agent at a time after attempting this bizarre new IDE

AI coding brokers can write options, refactor code, and monitor down bugs surprisingly nicely. But the usual workflow nonetheless feels moderately proscribing: begin an agent, anticipate it to complete, overview the end result, and repeat the method.

But there is a new IDE in the marketplace, and it takes a special method. Instead of constructing the agent sooner, Superset IDE lets a number of brokers work concurrently with out trampling over each other. And when you strive that mannequin, the single-agent workflow will start feeling obsolete.

An orchestration layer, not a seventh agent to decide on

Bring your personal Claude Code, Codex, or Cursor subscription

Despite being described as an agentic IDE, Superset is not competing instantly with Claude Code, Codex, Gemini CLI, or OpenCode. It is an orchestration layer that runs these instruments and organizes their work. It helps quite a few command-line agents, including Claude Code, OpenAI Codex CLI, Cursor Agent, GitHub Copilot, Gemini CLI, and OpenCode, whereas different CLI-based brokers can run with out particular configuration.

That distinction implies that Superset would not drive me to decide on one mannequin supplier or switch my workflow into one other proprietary assistant. I can proceed utilizing my current agent subscriptions and API keys, choosing a special agent for every activity if crucial. Superset would not proxy mannequin calls, so any mannequin utilization and related prices stay with the unique supplier.

The interface resembles a mixture of an IDE, terminal supervisor, Git consumer, and mission-control dashboard. The uncommon half is that every activity turns into a separate workspace moderately than one other dialog inside the identical challenge listing.

Git worktrees do the isolation, not a proprietary sandbox

Disposable workspaces, one department and listing per activity

Underneath the polished interface, Superset depends on a comparatively peculiar Git function: worktrees. Every workspace receives its personal department, working listing, terminal periods, and detected ports. Consequently, one agent can restore authentication whereas one other updates dependencies and a 3rd experiments with UI redesigns.

Because the brokers function in separate worktrees, they do not edit the identical working copy or fill one listing with conflicting uncommitted adjustments. Superset describes these workspaces as cheap and disposable: create one for a activity, overview the department, and take away the workspace after merging it.

Creating one is simple too. I can select a challenge, create a workspace from a brand new department, current department, or pull request and launch the popular agent with a immediate. Superset then retains the terminal, agent session, recordsdata, ports, and Git adjustments related to that job collectively.

Once the agent finishes, I can examine the diff inside Superset, proceed the dialog, open the workspace in one other editor, or ship the department by the standard pull request course of. The device provides a coordination layer with out attempting to switch Git.

Waiting received cheaper, not the brokers smarter

Racing two brokers on the identical immediate and merging the winner

Now since you may run a number of brokers collectively, you may break up a challenge into unbiased jobs. Instead of asking one agent to refactor an software, replace its assessments, revise the documentation, and repair an unrelated bug sequentially, I can assign every activity to a special workspace.

Superset additionally helps racing brokers. What which means is that I may give the identical drawback to Claude Code and Codex, let each produce unbiased branches, and merge the higher implementation. The official documentation explicitly recommends utilizing remoted workspaces for equivalent prompts and evaluating the ensuing diffs.

This would not make brokers extra clever. It makes ready inexpensive. When one agent is putting in dependencies or working a prolonged check suite, I can overview one other agent’s adjustments or reply to a different one which may want clarification.

That mentioned, you’ll want to be taught to assume like a coordinator moderately than a programmer working with a sidekick. Prompts want clearer boundaries, duties want express acceptance standards, and huge jobs should be divided into items that may be safely developed on separate branches.

A CLI, an MCP server, and a distant machine doing the work

Windows is not supported but, and macOS stays the actual goal

Superset’s command-line device exposes the identical backend utilized by the desktop software, permitting scripts and CI jobs to create workspaces, launch duties, set off automations, and management host machines. Its MCP server nevertheless, goes additional by permitting an agent to create and handle different workspaces, successfully turning one coding agent right into a coordinator for a number of employees.

Scheduled automations can deal with repetitive jobs like dependency updates, concern triage, changelog preparation, or routine audits. Each run produces a workspace that is still out there for inspection and follow-up moderately than silently modifying the principle checkout.

Remote entry can also be one other compelling function, particularly in case you’ve received a spare PC, residence server, or highly effective workstation. A workspace can run on one other machine whereas its terminals, agent periods, diffs, and ports stay accessible by the desktop app. That makes Superset notably attention-grabbing for local-first setups the place the laptop computer acts as a management panel and a bigger machine performs the precise work.

Five branches arrive sooner than you may overview them

Only Git-tracked recordsdata carry over, so setup self-discipline issues

Parallelism sounds nice for productiveness however introduces a brand new bottleneck: you. Five brokers can generate 5 branches sooner than you’d have the ability to rigorously overview them. This implies that poorly divided duties also can produce overlapping adjustments that should ultimately be mounted, even when their working directories stay remoted.

Worktrees additionally require some setup self-discipline. Only Git-tracked recordsdata are carried into a brand new workspace, so dependencies, ignored surroundings recordsdata, and different native assets may want lifecycle scripts. Development servers also can contend for ports as a result of Superset detects listening ports however would not routinely allocate a singular vary to every workspace.

Platform availability is one other limitation. Superset formally helps macOS, provides an experimental Linux AppImage, and would not present a Windows model on the time of writing, though there’s a waitlist open.

Last however not least, the free plan contains limitless native workspaces and initiatives for one consumer. Remote entry, automations, group options, and integrations require you to get at the least the Pro subscription priced at $20 per consumer month-to-month or $15 per consumer with annual billing.

The single-agent workflow is the one which feels out of date now

You turn out to be the bottleneck, not the mannequin

Superset’s actual innovation is not a better mannequin or one other chat panel. It is the belief that coding brokers have turn out to be processes value scheduling, isolating, monitoring, and reviewing.

Running a number of brokers will not go well with each challenge, and it will not rescue imprecise prompts or get rid of cautious code overview. But as soon as duties can progress on separate branches on the identical time, working one coding agent and gazing its terminal begins feeling out of date.