At Anthropic’s developer conference in April 2026, the AI company announced a new feature called ‘dreaming’ for its Claude AI agents, allowing them to sort through ‘memories.’ But the name ‘dreaming’ raises an important question: can we not stop naming AI features after human processes?
Key Takeaways
- Anthropic’s ‘dreaming’ feature allows Claude AI agents to sort through ‘memories.’
- The name ‘dreaming’ sparks debate over ethics and responsibility in AI development.
- AI companies are increasingly using human process names for their features and capabilities.
- This trend raises concerns about the accountability and transparency of AI decision-making.
- The use of human process names may create unrealistic expectations about AI capabilities and limitations.
Anthropic’s ‘Dreaming’ Feature
Anthropic’s ‘dreaming’ feature is a significant development in the field of artificial intelligence. The feature allows Claude AI agents to sort through ‘memories’ and retrieve relevant information. But the name ‘dreaming’ raises concerns about the ethics and responsibility of AI development.
As Wired reports, the use of human process names for AI features and capabilities is becoming increasingly common. This trend raises important questions about the accountability and transparency of AI decision-making.
The ‘dreaming’ system operates during low-activity periods, when Claude agents process stored interactions and data patterns. It’s not a passive state—it’s an active reindexing and reweighting of stored information based on relevance, frequency, and user-defined priorities. The feature improves response accuracy over time by identifying latent connections across user interactions. But calling it ‘dreaming’ implies a subconscious, almost biological process, which it is not. There’s no introspection, no emotional filtering, no randomness born of fatigue or subconscious bias. It’s algorithmic optimization under a poetic label.
And yet the label sticks. It’s memorable. It sells. But at what cost?
Historical Context
The habit of naming AI functions after human cognition didn’t start in 2026. It’s been building for over a decade. In 2015, Google introduced “Smart Reply” in Inbox, framing predictive text as intuitive understanding. Then came “learning”—machine learning, deep learning—terms that stuck despite their misleading simplicity. Neural networks, first theorized in the 1940s, were revived in the 2010s with renewed biological metaphors, even though today’s networks bear little structural resemblance to the human brain.
When OpenAI launched GPT-3 in 2020, it described the model as “reading” text or “answering” questions. Microsoft later said its AI could “reason” through problems. In 2023, Google’s Med-PaLM claimed it could “understand” medical queries. Each term borrowed from human cognition, each subtly shaping public perception.
Anthropic itself has used terms like ‘memory’ since Claude 2.1, allowing agents to retain context across sessions. By 2025, the company added ‘reflection’, a process where agents reviewed prior outputs before generating new ones. That already stretched the metaphor. Now, ‘dreaming’ takes it further—into the realm of unconscious processing, an idea loaded with psychological and philosophical meaning.
This pattern isn’t accidental. It’s a linguistic shortcut. Engineers and marketers reach for familiar terms because abstract technical descriptions don’t resonate. But in doing so, they blur the line between simulation and replication. The machine doesn’t dream. It reprocesses. It doesn’t remember. It retrieves. It doesn’t reflect. It recalibrates.
And every time they use these terms, they shape how developers, users, and regulators understand what the system actually does.
The Problem with Human Process Names
The use of human process names for AI features and capabilities can create unrealistic expectations about AI capabilities and limitations. It can also make it difficult to understand the actual decision-making processes behind AI systems.
When a user hears that an AI is “dreaming,” they might assume it’s generating creative insights the way a human does—through subconscious synthesis, emotional resonance, or intuitive leaps. But the system is doing no such thing. It’s running a scheduled background task that applies clustering algorithms to stored data, tagging and ranking associations based on user interaction history. It’s efficient. It’s useful. But it’s not dreaming.
This misalignment becomes dangerous when developers start coding around the metaphor, not the mechanism. If a team believes their AI ‘remembers’ conversations like a person, they may skip designing strong data expiration or consent protocols, assuming the system “knows” what to forget. It doesn’t. It stores until told otherwise.
The same applies to debugging. If a model produces a biased output during its ‘dreaming’ phase, engineers might struggle to trace it—not because the process is complex, but because the naming obscures the mechanics. Was it a ‘memory’ that was corrupted? Or was it a weighting function in the reindexing module that favored certain inputs? The metaphor hinders precise communication.
As the Wired article notes, “we are begging AI companies to stop naming features after human processes.” The article argues that this trend is not only confusing but also raises important questions about the ethics and responsibility of AI development.
What This Means For You
The use of human process names for AI features and capabilities has significant implications for developers and builders. It can create unrealistic expectations about AI capabilities and limitations, making it difficult to understand the actual decision-making processes behind AI systems.
Consider a startup building a mental health chatbot using Claude’s ‘dreaming’ feature. The founder sees ‘dreaming’ as a sign the AI is “processing emotions overnight,” leading them to claim the bot offers “therapeutic reflection.” But that’s not what’s happening. The system is reordering stored responses, not empathizing. If a user confides trauma and the bot later responds with a mismatched suggestion, the founder might blame the ‘dreaming’ process—as if it failed emotionally, not technically. The metaphor shields the real failure: poor system design and misaligned expectations.
For enterprise developers, the stakes are higher. A financial services firm using Claude to analyze client interactions might enable ‘dreaming’ to “help the AI understand long-term client needs.” But if the AI starts making investment recommendations based on reindexed data patterns, auditors will want to know how those connections were formed. Was it a deliberate rule? A statistical anomaly? If the team can’t explain it in technical terms—because they’ve been using terms like ‘memory’ and ‘dreaming’—they risk failing compliance checks.
Even internal documentation suffers. Imagine a developer joining a team where the codebase refers to “dream cycles” and “memory pruning.” Onboarding takes longer. Misunderstandings multiply. A simple cache-clearing function becomes shrouded in pseudoscientific language. That’s not branding. That’s operational debt.
Developers and builders must be aware of these implications and take steps to ensure that AI systems are transparent and accountable. This may involve using clear and concise language when describing AI features and capabilities.
Competitive Landscape
Anthropic isn’t alone in this naming trend—nor is it the first. OpenAI refers to its models “reasoning” through chain-of-thought prompts. Google’s DeepMind has used “imagination” to describe how its agents simulate future moves in game environments. Meta has described certain AI behaviors as “planning” and “thinking.”
But Anthropic’s choice stands out because of its framing. The ‘dreaming’ feature wasn’t just a backend upgrade. It was presented with narrative flair—a Claude agent “resting” and “making sense of its day.” The demo showed a timeline of interactions being “revisited” at night. That language isn’t neutral. It’s anthropomorphic theater.
Other companies have started pushing back. In early 2026, a coalition of open-source AI developers launched a naming guide advocating for literal, functional terms. “Call it reindexing. Call it background processing. Call it data reweighting. Just don’t call it dreaming,” one contributor wrote. Some smaller AI firms have adopted terms like ‘offline optimization’ or ‘context refinement’ instead.
But the pressure to stand out in a crowded market is intense. “Dreaming” generates headlines. “Nighttime data reprocessing” doesn’t. That’s why the trend persists. Marketing trumps precision.
The Future of AI Development
The use of human process names for AI features and capabilities is a symptom of a larger issue in AI development. As AI systems become increasingly complex and autonomous, ensure that they are transparent and accountable.
The future of AI development will depend on our ability to address these issues and create AI systems that are transparent, accountable, and responsible.
Technical debt isn’t just in code—it’s in language. Every time a company chooses a human-centered metaphor over a mechanistic one, it adds a layer of ambiguity that future teams will have to unpack. That ambiguity slows down debugging, complicates regulation, and erodes public trust.
We’re entering an era where AI agents will operate across healthcare, finance, education, and law. In those domains, clarity isn’t optional. A misinterpreted term could mean a misdiagnosis, a flawed legal argument, or a financial loss. If a doctor asks whether an AI “remembers” a patient’s allergy history, they need to know whether that means a database lookup or a probabilistic inference based on incomplete data.
The naming conventions we adopt today will shape how these systems are governed tomorrow.
What Happens Next?
Anthropic hasn’t responded to requests for comment on whether it plans to reconsider the ‘dreaming’ label. But internal signals suggest hesitation. Some engineers have reportedly pushed back, arguing that the term undermines technical credibility.
Regulators are starting to take note. The EU AI Office has flagged “anthropomorphic terminology in technical documentation” as a potential risk factor in its upcoming audit framework. If adopted, companies may soon be required to justify the names they give to AI functions—especially in high-risk sectors.
Meanwhile, developer backlash is growing. Online forums are filled with threads debating whether terms like ‘dreaming’ should be banned from documentation. Some open-source projects have added linters that flag anthropomorphic language in code comments.
The next 12 months will be telling. Will more companies follow with features named ‘meditating’, ‘learning’, ‘grieving’? Or will the backlash force a return to literal, descriptive naming?
What’s Next?
As AI companies continue to develop and announce new features, remain vigilant about the use of human process names. We must continue to demand transparency and accountability in AI development and push for clearer and more concise language when describing AI features and capabilities.
A Call to Action
To the AI companies, we say: stop naming features after human processes. Use clear and concise language when describing AI features and capabilities. Be transparent and accountable in your development processes.
Conclusion
The use of human process names for AI features and capabilities is a concerning trend that raises important questions about the ethics and responsibility of AI development. in the field of artificial intelligence, prioritize transparency and accountability.
What This Means For You:
Developers and builders must be aware of the implications of human process names for AI features and capabilities. They must take steps to ensure that AI systems are transparent and accountable, using clear and concise language when describing AI features and capabilities.
This trend has significant implications for the future of AI development. As AI systems become increasingly complex and autonomous, ensure that they are transparent, accountable, and responsible.
Final Thoughts:
As we look to the future of AI development, transparency and accountability will be crucial. The use of human process names for AI features and capabilities is a concerning trend that must be addressed. We must demand clearer and more concise language when describing AI features and capabilities and push for greater transparency and accountability in AI development.
Sources: Wired, Wired
This article is a call to action for AI companies to stop naming features after human processes. We believe that transparency and accountability are essential in AI development and hope that spark a wider conversation about the ethics and responsibility of AI development.
A dimly lit conference room with a projector screen displaying a presentation on AI development. A group of developers and researchers are seated in the front row, engaged in a heated discussion about the implications of human process names for AI features and capabilities.
