If you follow artificial intelligence news, you may have heard about the "Code with Claude" conference that Anthropic held in San Francisco on May 6, 2026. Among the announcements of the day, a feature called Dreaming particularly caught the attention: it allows an AI agent to review its own errors between two work sessions and adjust its memory automatically, without human intervention.

The principle is simple, but the results are starting to be striking. Without changing a single line of code, and without retraining any models, companies using Claude agents have seen their performance improve from one session to the next. Harvey, a company specializing in AI tools for law firms, reported an approximately 6x increase in its task completion rate after activating Dreaming. A figure worth examining.

1. The problem that Claude Dreaming solves

Until now, the major weakness of AI agents in a professional context was their lack of persistent memory between sessions. An agent could perform a task correctly one day, then repeat the same mistake the next day, because it had no record of what had happened previously.

Let's take a concrete example: an agent who helps draft legal contracts. If this type of document requires a specific format, references to certain clauses, or a particular writing style, the agent would have to relearn these details with each new session. The result: more time wasted correcting repetitive errors, more back-and-forth with users to clarify settings, and reduced productivity.

Dreaming directly addresses this problem. It's a scheduled process that runs in the background while the agent is idle, much like a night's sleep consolidates the day's learning. This process reviews previous sessions, identifies recurring patterns (frequent errors, user preferences, client-specific constraints), removes duplicates from the agent's memory, and inserts new, practical rules based on accumulated experience.

2. Harvey's performance: when the numbers speak for themselves

Harvey was one of the first companies to have access to Dreaming in a pilot version before its public launch. This American company develops specialized AI tools for law firms, particularly for legal research, document drafting, and contract analysis.

Before Dreaming, Harvey's agents had a well-documented problem: they would forget the specifics of certain file types, lawyer preferences, and common mistakes to avoid from one session to the next. Each new case essentially started from scratch in terms of contextual learning.

After activating Dreaming, Harvey measured an approximately 6x increase in his task completion rate. This gain stemmed from two main factors:

• The agents asked far fewer clarifying questions because they already knew the firm's preferences and constraints.
• The number of recurring errors had dropped considerably, since Dreaming had consolidated the lessons from past sessions into the agent's memory.

Harvey didn't modify his code or retrain his models. The agents improved themselves between sessions, thanks to Dreaming's automated process. According to VentureBeat, which covered the launch, it's precisely this autonomous and continuous nature that distinguishes Dreaming from traditional improvement approaches.

3. How does Dreaming work technically?

For those who are curious, here's how Dreaming is structured within the Claude agent architecture. The agent has an external memory space, separate from the language model itself. This space stores information about past sessions: what worked, what failed, user preferences, and the specifics of each task.

Dreaming is a maintenance layer that runs periodically on this memory space. It performs three main operations:

• Consolidation : removal of redundant entries and grouping of similar information to avoid contradictions.
• Updating : replacing outdated entries with more recent and relevant data, taking into account evolving preferences.
• Pattern extraction : identifying recurring patterns that allow the agent to anticipate needs and avoid known errors before they even recur.

The result is an agent that gradually improves at its task, without human intervention and without the cost of retraining. This is a different approach from traditional methods of improving AI models, which generally require time-consuming and resource-intensive data collection and training cycles.

Note that Dreaming was launched as a "research preview" at Code with Claude 2026. Access is on demand, and Anthropic is positioning the feature as exploratory rather than production-ready at this stage, with plans for progressively expanding access.

4. The two other new features announced with Dreaming

Dreaming wasn't the only new feature at Code with Claude 2026. Anthropic announced two other complementary features for its managed agents, which together form a coherent system:

• Outcomes : a structured feedback system after each task (success, failure, reason for failure) that provides Dreaming with high-quality data to work with. Without Outcomes, Dreaming wouldn't be able to clearly distinguish what went well from what went wrong, which would limit the effectiveness of its adjustments.
• Multi-agent orchestration : Agents can now delegate subtasks to other agents in parallel. A main agent can orchestrate several specialized agents simultaneously, reducing bottlenecks in complex, multi-stage projects.

Together, these three features address the main challenges of using AI agents at scale: persistent memory (Dreaming), the ability to evaluate their own performance (Outcomes), and efficiency on complex tasks (multi-agent orchestration). It's an architecture designed so that agents become more reliable and autonomous as they are used.

5. Claude Code: the fastest-growing AI product in history

These new features come amid exceptional growth for Anthropic and for Claude Code in particular. Publicly launched in the spring of 2025, Claude Code has become, according to several industry sources, the fastest-growing software product in the history of the industry, surpassing the records set by the largest technology platforms at their inception.

The numbers speak for themselves: Claude Code now generates over $2.5 billion in annualized revenue and accounts for more than half of all enterprise spending on Anthropic platforms. In just one year, the product has gone from zero to a key player in AI-assisted development, with weekly active users doubling since the beginning of 2026.

More broadly, Anthropic's annualized revenue jumped from $9 billion in January 2026 to $47 billion in May 2026, according to information published by Fortune during the confidential filing of the IPO prospectus on June 1, 2026. The company now has more than 300,000 business customers, and the number of customers spending more than $100,000 annually has increased sevenfold in one year.

These figures illustrate an adoption that extends far beyond the circle of tech enthusiasts: companies of all sizes, across all sectors, are integrating Claude into their daily workflows. AI agents are no longer a technological curiosity; they are becoming a normal component of the work environment.

6. What this means for SMEs in Quebec

For SMEs in Trois-Rivières, the Mauricie region, or the Quebec City area, these advancements aren't yet available at the click of a mouse. Dreaming has restricted access, and enterprise AI tools like Claude still require specialized support for proper and secure deployment. But the path is clear, and organizations that start preparing now will have a head start.

The sectors most likely to benefit from persistent AI agents in the near future are:

• Professional firms (law, accounting, consulting), where agents can remember client preferences and the specifics of each case from one meeting to the next.
• Automated customer service, where an agent who remembers past errors and customer habits can handle routine requests without human escalation.
• Document management, automated invoicing and the production of recurring reports, where repetitive format errors are particularly costly in terms of time.
• Cybersecurity and network monitoring, where persistent agents can better detect anomalies by relying on accurate behavioral history.

The question is no longer "Can AI do this?" The question has become "When do we start, and how?" Discover OKTO Solutions' IT services to assess how to integrate these technologies into your organization in a structured and tailored way.

Frequently Asked Questions

What exactly is Claude Dreaming?

Claude Dreaming is an automated process launched by Anthropic in May 2026. It runs in the background while an AI agent is not active, reviewing past sessions to consolidate memory, remove redundancies, and extract useful patterns. The goal is for the agent to perform better with each new session, without requiring any manual intervention in its configuration.

Is Dreaming available to all businesses right now?

No. At launch on May 6, 2026, Dreaming was only available in "research preview," meaning on request to selected partners. Anthropic plans to gradually expand access over the coming months. Interested companies can submit an access request directly through the Claude.ai platform.

How are Claude agents different from a classic AI assistant?

A typical AI assistant answers one-off questions and generally has no memory between conversations. A Claude agent can perform complex, multi-step tasks, use external tools, remember past interactions thanks to persistent memory, and now improve between sessions thanks to Dreaming. This is the difference between an advisor consulted only once and a collaborator who learns and adapts over time.

A development to watch closely for your SME

Claude Dreaming represents a concrete step towards AI agents that do more than simply answer questions; they improve over time, just like an employee gaining experience on the job. For Quebec SMEs, now is the perfect time to start exploring how these tools can be integrated into your operations, before your competitors get ahead. TheOKTO Solutions can guide you from the initial audit to implementation, ensuring a smooth and secure process. Contact us to discuss your automation and artificial intelligence needs.