Skip to main content
roo coderoomotelocal-firstBYOLLMopinioncloud-vs-local

The 'IDEs are dead' debate is a misdirection.

Bodega One9 min read
Quick answer

On May 15, 2026, Matt Rubens archived Roo Code with 3 million cumulative installs. The stated reason: IDEs are not the future of coding. The follow-on product, Roomote, runs in Slack. But Roomote's own homepage says editor tools and Roomote are “complementary,” not that editors are dead. The marketing soundbite contradicts the product page. The real shift here is not IDE versus agent. It is local versus cloud. This essay argues the binary you are being sold is the wrong one. If you want the practical migration guide instead, read our companion post on where Roo Code users should go next.

On May 15, 2026, the RooCodeInc/Roo-Code repository was archived. Final release v3.54.0. 24,102 GitHub stars. 3,278 forks. 1,636,212 residual VS Code Marketplace installs and 3 million cumulative across all platforms. The README banner is short: “The Roo Code Extension was shut down on May 15th. If you're looking for an alternative, check out ZooCode (a fork started by the Roo Code community) and Cline (from where Roo Code originated).”

Matt Rubens' stated reason for the pivot is that IDEs are not the future of coding. The team is going all-in on Roomote, a cloud agent that integrates with Slack, GitHub, and Linear. Verbatim from Kilo's migration guide: “Roo Code hit 3 million installs. We're shutting it down to go all-in on Roomote.”

That framing has already done its job in tech press. The question this essay actually answers is narrower: is the “death of the IDE” thesis what is really being claimed, or is the editor surface a sideshow hiding a sharper claim about where your code runs?

What Roo Code actually shipped

Roo Code mattered. The repo hit 24,102 stars on the way to archive. 3,278 forks. Three million cumulative installs. The extension itself is still live in the Marketplace at 1,636,212 installs as of this writing, with a shutdown notice attached. Few open-source coding tools accumulate that kind of footprint in under two years.

The technical contributions are real. Kilo's “Thank you, Roo!” post by Brian Turcotte credits Roo with “custom modes, the Architect/Code/Debug split, diff-based editing, and the philosophy that agents should actually do things.” The apply_diff pattern that reduced token output versus full-file rewrites. Permission gating. Mode-switching as a first-class concept. Kilo runs on a Roo lineage. Zoo Code is the community fork keeping the codebase alive at zoocode.dev. The work was good. The team is taste-driven and shipped fast.

This essay is not a takedown of Roo Code or of Matt Rubens. The team is moving to a real product and the reasoning behind the move is real. The disagreement is with the framing, not the people. So: read the second sentence.

Read the second sentence

The soundbite is “IDEs are not the future of coding.” The product page tells a different story.

Roomote's homepage tagline is “The AI engineer for the interrupt work slowing your roadmap.” The pitch underneath it is verbatim:

“Those are great editor tools for one engineer writing code in an IDE. They wait for engineers. Roomote waits for operational work: bug reports, alerts, regressions, internal asks. The two are complementary.”

The Roomote docs are even clearer:

“Roomote is not an IDE, local copilot, or desktop app. It is a shared engineering teammate that works from Slack, GitHub, Linear, and the web dashboard so your team can ask repo-grounded questions and create reviewable work without pulling everyone into the same editor session.”

And again: “Roomote complements editor-based coding tools by making shared engineering work visible and reviewable outside the IDE.”

Take a beat. The headline says editors are over. The product page says editors and Roomote are complementary. Those are two different claims. The product page is the one that has to be true for Roomote to ship the integrations it ships. The headline is the one that travels in screenshots.

The actual claim Rubens is making, once you strip the framing, is not “the editor is dead.” It is “your code, your repo, your sprint work, and your AI inference should live on our infrastructure.” Roomote runs from Slack. From GitHub. From Linear. From a web dashboard. Plus Jira, Asana, Notion, Sentry, Vercel. All SaaS. The form factor is not “not-an-IDE.” The form factor is “not-on-your-machine.” That is a real, defensible business pitch. It is a different pitch from the one the headline made.

What other “death of the IDE” essays actually say

The Roomote framing is not novel. There is a small genre of “the editor is over” essays this year. None of them, read carefully, actually say editors are over.

Cursor: “The Third Era of AI Software Development”

Michael Truell's February 26, 2026 essay announces Era 3: cloud agents that “run on its own virtual machine, allowing a developer to hand off a task and move on.” The headline shift is striking: “Cursor is no longer primarily about writing code. It is about helping developers build the factory that creates their software.”

Truell's adoption stat is the load-bearing claim: “Thirty-five percent of the PRs we merge internally at Cursor are now created by agents operating autonomously in cloud VMs.” That number is real and it matters. But it is internal Cursor data, on Cursor's code, by Cursor's agents, all running in cloud VMs Cursor owns. It is a stat about Cursor's own dogfooding, not about the industry. And critically: Cursor still ships an editor. The Third Era is escalation, not death. The editor surface gets a cloud-agent layer added around it. Nobody at Cursor turned off the editor.

Zed: “The Death of the IDE?”

Steve Yegge and Nathan Sobo's August 12, 2025 conversation has a question mark in the title for a reason. Nathan Sobo's conclusion is verbatim: “The IDE is dead. Long live the IDE. Or maybe the IDE needs to be reinvented in some very fundamental ways.” That is not a death notice. It is an architectural critique.

Yegge's strongest line is also worth quoting: “The IDE monolith architecture is kind of flawed... your AI needs to be able to do everything for you.” Fair. Concede the point. Current IDE architectures need reinvention to handle agent-density at the level Yegge and Sobo are imagining. The point is not that editors stay frozen. The point is that Zed's answer to its own essay is to ship a faster, more collaborative, more agent-aware editor. They are reinventing the IDE, not abandoning it.

JetBrains Air: “Agentic Development Environment”

JetBrains Air bills itself as an “Agentic Development Environment.” The positioning copy is explicit: “Air is a standalone desktop application designed for agent-powered workflows... without replacing existing development workflows.” JetBrains does not claim IDEs are dead. They ship IntelliJ, PyCharm, GoLand, WebStorm, and now a standalone desktop app for agent workflows. The lineup grows. Nothing in the IDE catalog gets retired.

The pattern across all three: every serious essayist who has actually shipped a product in 2025 or 2026 lands on “the editor surface gets richer, the agent gets more autonomous, the two are not in a zero-sum fight.” Nobody who ships the no-editor product ships it as the main act. Roomote is the closest thing, and Roomote's own docs call it complementary.

The data the death-of-the-IDE essays skip

If the editor were dying, you would expect the install graphs to show it. They do not.

The editor-resident, BYOK, local-LLM-capable stack is the fastest-growing segment in the post-Roo landscape. The cloud-agent segment is loud. The local-editor-plus-local-agent segment is large.

The actual fork in the road

The binary “IDE versus agent” is misdirection. The real binary is local versus cloud, and that is a binary about ownership.

When your agent runs in Roomote's VM, Roomote sees your code. When it runs in Cursor's Third Era cloud VM, Cursor sees your code. When it runs in your local editor against a local Ollama instance, nobody sees your code. That is the architectural difference. It is not a feature difference. It is a question of who has root on the runtime that touches your repo.

This matters for enterprise security review, regulated industries, IP-sensitive work, and the kind of work that ends in arbitration when a vendor's container leaks. It also matters for cost. Cursor reset credit pricing in June 2025 and effectively halved Pro's monthly usage overnight. Augment Code sunset completions for non-enterprise on March 31, 2026. GitHub Copilot moves all plans to usage-based billing on June 1, 2026. Roo Code archived May 15 and refunded balances. The pattern is the same every quarter: cloud-agent vendors change the meter, and the developers depending on them have a small window to migrate or absorb the new price. Local does not have a meter. Your code, your model, your machine, your cost. The number stays the number.

The “IDE versus cloud agent” framing erases the option of “local agent inside a local editor you actually own.” That is exactly what Cline plus Ollama is. It is exactly what Continue plus LM Studio is. It is exactly what Bodega One is. It is the option the binary is built to make invisible.

What Bodega One actually thinks

Bodega One is a standalone desktop IDE with an autonomous agent built in. Electron 40, Monaco editor, AI chat, an agent that can plan and execute multi-file changes, BYOLLM across 10+ provider presets, air-gap mode with nine layers of network egress blocking. Windows, macOS, Linux. One-time purchase. $79 Personal or $149 Pro. No subscription, no credit reset, no quota tier.

We are not arguing cloud agents are bad. Roomote is a real product. Cursor's Third Era pitch is internally consistent. There is real value in handing off a Slack interrupt to a managed cloud agent. We are arguing that “IDEs are dead” is the wrong frame for what is actually happening, and that the real question is who owns the runtime your code passes through.

Our pitch is simple. Pay once. Run locally. Bring your own LLM. The agent runs on your hardware. The repo never leaves your machine. The cost does not reset every quarter because someone in finance redrew the credit table. If that matches how you want to work, the beta is live now. If it does not, Cline and Continue are excellent free options and Roomote is the consistent pick if you agree with Rubens about the surface.

The framing we want to leave you with is not “the editor is forever.” It is: the binary you are being sold is the wrong one. Pick on the right axis. If you want the practical “where do Roo Code users go on May 15” piece, we wrote that one in April. This essay is the opinion-piece companion.

Common questions

Did Matt Rubens actually say “IDEs are not the future of coding”?

The verbatim line is paraphrased across secondary reporting (The New Stack and others) and originates in a Roo Code X post that is no longer primary-fetchable. The Rubens quote that is primary-verified, captured in Kilo's migration guide, is: “Roo Code hit 3 million installs. We're shutting it down to go all-in on Roomote.” The “IDEs are not the future” framing is the broader thesis the team used to justify the pivot. Treat it as the company's stated position, not a verified single sentence.

What does Roomote itself say about IDEs?

Roomote's homepage says editor tools and Roomote are “complementary,” not that editors are obsolete. The verbatim line: “Those are great editor tools for one engineer writing code in an IDE. They wait for engineers. Roomote waits for operational work: bug reports, alerts, regressions, internal asks. The two are complementary.” The docs reinforce this: “Roomote is not an IDE, local copilot, or desktop app... complements editor-based coding tools.” The product page contradicts the marketing soundbite.

Is the editor dying or just changing?

Changing. Zed's “Death of the IDE?” essay ends the title with a question mark and Nathan Sobo's conclusion is reinvention, not death. JetBrains Air is positioned as a complement to existing IDEs, not a replacement. Cursor's “Third Era” essay reframes the editor but still ships one. Every serious analyst lands on the same conclusion: the editor surface gets richer, the agent gets more autonomous, the two are not in a zero-sum fight.

What is the actual difference between a cloud agent and a local IDE with an agent?

Where your code runs. A cloud agent like Roomote clones your repo into a vendor's infrastructure, runs the agent on the vendor's machines, and returns a pull request. A local IDE with an agent runs the agent on your hardware, against your filesystem, with your choice of model. The editor surface is a sideshow. The real question is whether your code, your repo, and your inference happen on your machine or someone else's.

Which fast-growing AI coding tools still ship as editors?

Most of them. As of May 2026, Cline has 4,003,499 VS Code Marketplace installs (2.5x Roo's residual count). Continue.dev has 3,019,736. OpenCode has 162,460 GitHub stars on a local CLI agent. Ollama's top model alone has 114.6 million pulls. Every winner in the post-Roo shakeout is shipping an editor-resident, BYOK, local-LLM-capable stack. The “IDEs are dead” framing does not match the install numbers.

Why does local versus cloud matter more than IDE versus agent?

Because the cloud meter keeps moving. Cursor reset credits in June 2025, cutting Pro's monthly usage by more than half. Augment Code sunset completions for non-enterprise on March 31, 2026. GitHub Copilot moves all plans to usage-based billing on June 1, 2026. Roo Code archived May 15 and refunded balances. The pattern is the same every quarter: cloud-agent vendors change the rules, and the developers depending on them have a window to migrate or absorb the new price. Local does not have a meter. Your code, your model, your machine, your cost.

How is Bodega One different from Roomote?

Different bet. Roomote is a cloud SaaS agent that runs in Slack and produces pull requests from someone else's infrastructure. Bodega One is a standalone desktop IDE with a local agent: Electron 40, Monaco editor, AI chat, autonomous coding agent, 10+ LLM provider presets, air-gap mode. One-time purchase at $79 Personal or $149 Pro. The product decision: own the runtime, own the editor, own the model. If you agree with the Roo team that the editor is the wrong surface, Roomote is the consistent pick. If you think the real fight is local versus cloud, we are the consistent pick.

Sources

Common questions

Did Matt Rubens actually say 'IDEs are not the future of coding'?
The verbatim line is paraphrased across secondary reporting (The New Stack and others) and originates in a Roo Code X post that is no longer primary-fetchable. The Rubens quote that is primary-verified, captured in Kilo's migration guide, is: 'Roo Code hit 3 million installs. We're shutting it down to go all-in on Roomote.' The 'IDEs are not the future' framing is the broader thesis the team used to justify the pivot. Treat it as the company's stated position, not a verified single sentence.
What does Roomote itself say about IDEs?
Roomote's homepage says editor tools and Roomote are 'complementary,' not that editors are obsolete. The verbatim line: 'Those are great editor tools for one engineer writing code in an IDE. They wait for engineers. Roomote waits for operational work: bug reports, alerts, regressions, internal asks. The two are complementary.' The docs reinforce this: 'Roomote is not an IDE, local copilot, or desktop app... complements editor-based coding tools.' The product page contradicts the marketing soundbite.
Is the editor dying or just changing?
Changing. Zed's 'Death of the IDE?' essay ends the title with a question mark and Nathan Sobo's conclusion is reinvention, not death. JetBrains Air is positioned as a complement to existing IDEs, not a replacement. Cursor's 'Third Era' essay reframes the editor but still ships one. Every serious analyst lands on the same conclusion: the editor surface gets richer, the agent gets more autonomous, the two are not in a zero-sum fight.
What is the actual difference between a cloud agent and a local IDE with an agent?
Where your code runs. A cloud agent like Roomote clones your repo into a vendor's infrastructure, runs the agent on the vendor's machines, and returns a pull request. A local IDE with an agent runs the agent on your hardware, against your filesystem, with your choice of model. The editor surface is a sideshow. The real question is whether your code, your repo, and your inference happen on your machine or someone else's.
Which fast-growing AI coding tools still ship as editors?
Most of them. As of May 2026, Cline has 4,003,499 VS Code Marketplace installs (2.5x Roo's residual count). Continue.dev has 3,019,736. OpenCode has 162,460 GitHub stars on a local CLI agent. Ollama's top model alone has 114.6 million pulls. Every winner in the post-Roo shakeout is shipping an editor-resident, BYOK, local-LLM-capable stack. The 'IDEs are dead' framing does not match the install numbers.
Why does local versus cloud matter more than IDE versus agent?
Because the cloud meter keeps moving. Cursor reset credits in June 2025, cutting Pro's monthly usage by more than half. Augment Code sunset completions for non-enterprise on March 31, 2026. GitHub Copilot moves all plans to usage-based billing on June 1, 2026. Roo Code archived May 15 and refunded balances. The pattern is the same every quarter: cloud-agent vendors change the rules, and the developers depending on them have a window to migrate or absorb the new price. Local does not have a meter. Your code, your model, your machine, your cost.
How is Bodega One different from Roomote?
Different bet. Roomote is a cloud SaaS agent that runs in Slack and produces pull requests from someone else's infrastructure. Bodega One is a standalone desktop IDE with a local agent: Electron 40, Monaco editor, AI chat, autonomous coding agent, 10+ LLM provider presets, air-gap mode. One-time purchase at $79 Personal or $149 Pro. The product decision: own the runtime, own the editor, own the model. If you agree with the Roo team that the editor is the wrong surface, Roomote is the consistent pick. If you think the real fight is local versus cloud, we are the consistent pick.

Related posts

Ready to own your tools?

Beta is live now. Join the waitlist for full launch.

Join the Waitlist →See Pricing
← Back to the blog