The short answer is: probably not in principle, probably yes in practice. The gap between those two answers is doing most of the philosophical work in the field right now.
Two different categories of claim
Artificial general intelligence describes a property of a system: the ability to perform, across a wide range of domains, at or above some baseline of human capability. It is a claim about what the system can do.
Vibe computing describes a property of the interaction: the user describes an outcome and the system carries out the operations. It is a claim about how the user addresses the system.
Stated like that, the two are orthogonal. A system can be very general without being addressable in natural language at all (consider an unsupervised theorem prover). A system can be addressable in natural language for a narrow set of tasks and not be remotely general. The properties vary independently in principle.
OpenAI’s framing
OpenAI states the goal openly. Their charter defines AGI as “highly autonomous systems that outperform humans at most economically valuable work” and frames the mission as building that, safely. Under this definition, vibe computing is not a category mistake against AGI. It is a substrate. If your AGI cannot operate the computer in front of you to do the work you would have otherwise done yourself, it has not outperformed you at much that is economically valuable. Products like Codex and ChatGPT Atlas are intermediate steps along this path. Sam Altman has been explicit that “agents that do real work for you” are the through-line, not a side quest.
Anthropic’s framing
Anthropic is more careful with the word. Their core views avoid making AGI an explicit destination and frame the work as building “safe, beneficial AI systems” whose capabilities are advanced enough to matter. When they shipped Claude Computer Use in 2024, the framing was research-preview-with-caveats, not agi-is-here. Their CEO Dario Amodei prefers the term “powerful AI,” and his essay on what it could do is closer in shape to a thought experiment than a product roadmap.
For Anthropic, vibe computing is one good use case of capable AI, but it is not the goal. The goal is the AI being good.
Where the convergence is real
Here is where the in-principle distinction starts losing its grip. The current generation of vibe computing systems is built on large language models with vision and tool-use, which happen to be the most general artifacts the field has produced. The systems that come closest to operating arbitrary applications are the same systems that come closest to general capability. As a result, progress on one currently looks like progress on the other, and the labs building “agents” and the labs building “AGI” are largely the same labs.
This is a feature of this present moment, not a property of the concepts. It is possible to imagine a future system that is more general but addressed only through formal interfaces, or one that is fully vibe-addressable but narrowly scoped to a single domain. Both already exist in partial form. But if you point to the actual products people are currently shipping under the AGI banner, Codex, Atlas, Claude Computer Use, Operator, Manus, Perplexity Personal Computer, every one of them is, by structure, a vibe computing system.
An opinion
The version of AGI most people would actually recognize if they got it, an assistant on the device in front of them that they can ask anything and that quietly takes care of it, is vibe computing taken to its limit. It is the version Sam Altman gestures at when he says “magic intelligence in the cloud” without explaining further. It is the version OpenAI’s charter describes when it talks about “most economically valuable work.”
By that practical, end-user-facing definition: yes. Vibe computing and AGI converge in the system you would actually want to use, even though they remain different claims about the world. Anthropic is right that they should not be conflated. AGI is a capability threshold, vibe computing is an interaction shape, and treating them as identical lets people overclaim. But OpenAI is right that in the products they intend to ship, the two stop being separable.
The interesting question is therefore not “is AGI vibe computing.” It is “is the vibe-computing limit reachable, and is it a good idea to reach it.” The first is an empirical bet that the next few years will settle one way or another. The second is a question worth having before, not after.