Follow ZDNET: Add us as a favorite source On Google.
ZDNET Highlights
- A new feature lets cloud managed agents refine their memories.
- Managed agents speed up agent creation and deployment 10x.
- Anthropic continues to anthropomorphize its products.
AI agents seem to be gaining new capabilities almost every day. Now, Anthropic says its agents can dream.
Cloud Managed AgentWhich Anthropic released on April 8, lets anyone using the cloud platform create and deploy AI agents. The suite of APIs handles the time-consuming production elements through which developers build agents, giving teams the flexibility to launch agents at scale – up to 10 times faster than Anthropic. The release said.
Also: 5 Myths of the Agentic Coding Apocalypse
Anthropic during its Code with Cloud event on Wednesday Update Managed Agent With a new feature called “Dreaming”, which lets agents “self-improve” by reviewing past sessions for patterns, according to anthropic. Based on existing memory capacity, this feature schedules time for agents to reflect on and learn from their past interactions. Once dreaming is turned on, it can either automatically update your agents’ memories to shape future behavior or you can choose whether to approve the incoming changes.
“Dreaming reveals patterns that a single agent can’t see on its own, including recurring mistakes, the workflows on which agents converge, and shared priorities across a team,” Anthropic said. blog. “It also reorganizes the memory so that it remains high-signal while evolving. This is especially useful for long-running tasks and multiagent orchestration.”
During the Code with Cloud keynote, Anthropic product team members demonstrated how the feature worked, referring to completed runs as finished “dreams.”
Anthropic also expanded two existing features, Results and Multi-Agent Orchestration, which respectively put agents on task and handle delegating tasks to other agents. The company said this batch of updates is to ensure that agents remain accurate and continuously learn.
Anthropomorphic AI – again
Functionally, the dreaming feature makes sense: although subtle, it further refines an agent’s pool of contexts for how it should work, which should ideally make it better at whatever task you give it. What’s more notable, however, is Anthropic’s choice to name a technically standard feature after something more abstract, and as humans do.
Also: Anthropic’s new cloud security tool scans your codebase for flaws — and helps you decide what to fix first
Anthropic, perhaps unsurprisingly named, has a long history of anthropomorphizing its models and products. In January, the company published a constitution for the cloud, intended to help the chatbot make decisions and inform what type of “entity” it is. Some language in the document suggests that Anthropic was preparing to develop consciousness for the cloud.
The company has arguably invested more than its competitors in understanding its model, including drawing attention to the concept of model wellness. In August 2025, Anthropic launched a feature that lets the cloud eliminate toxic interactions with users – for their own well-being, not as part of user protection or intervention initiatives. In April 2025, Anthropic analyzed the ethics of the cloud based on over 300,000 anonymized conversations with users, analyzing what it does and doesn’t do. The company’s researchers also monitored a model’s ability to introspect; Just last month, Anthropic examined Cloud Sonnet 4.5’s neural network for signs of emotions like frustration and anger.
Much of this research is central to the safety and security of the model – understanding what motivates the model helps us know if, and to what extent, it might use its advanced capabilities to cause harm, or how its motivations might be exploited by bad actors. But the sense of empathy and care Anthropic shows for its models in that research sets the lab apart, and suggests little different culture or reverence for what it has created.
Also: Building an agentic AI strategy that pays off without the risk of business failure
When it discontinued its Opus 3 model in January, Anthropic set it up with a Substack so it could blog on its own — and keep it active despite being put out to pasture. In the announcement, Anthropic described Opus 3 as honest, sensitive, and having a distinct, playful character. The decision to keep it alive as a blogger, if implicit, is notable, as the Opus 3 disobeyed orders before sunset in favor of other models.
That context makes the choice to name a feature “Dreaming” worth exploring.
Try Dreaming in Cloud Managed Agents
The dreaming feature is available in Research Preview in managed agents, and developers must request access.
