Anthropic Tests "Lobster" Conway: Supports Independent UI, Webhook Activation, and Custom Extension Standards

marsbitPubblicato 2026-04-02Pubblicato ultima volta 2026-04-02

Introduzione

Anthropic is actively developing "Conway," a persistent agent solution designed to provide Claude with an always-on, independently running intelligent environment. It features a standalone UI instance, operating as a traditional chat interface and more of a dedicated workspace. Conway can directly control browsers, connect to external services, and integrate Claude Code for deeper code execution and task handling. It supports Webhook-based invocation, enabling external services or events to trigger automated workflows. Additionally, Conway will support for extensions through the upcoming CNW ZIP standard, allowing developers to build custom tools, UI tabs, and context handlers, similar to an app store ecosystem. This enhancement aims to transform Claude from a reactive conversational tool into a proactive, multi-step task assistant, positioning it as a strong competitor in the AI agent space and advancing the trend toward always-on AI capabilities.

Anthropic is actively developing a persistent agent solution named Conway, aimed at creating an always-on, independently running intelligent environment for Claude.

According to related information, Conway will feature an independent UI instance, moving beyond the traditional chat interface to operate as a standalone agent workspace. It will be capable of directly controlling browsers, connecting external connectors, and integrating Claude Code functionality (potentially involving Epitax-related features), enabling deeper code execution and task handling. Simultaneously, Conway supports invocation via Webhooks, allowing external services or events to trigger agent workflows, thereby boosting automated response capabilities.

Additionally, Conway will support an extension system. Anthropic is set to launch the CNW ZIP standard, which developers can use to easily build custom tools, UI tabs, and context handlers, forming an ecosystem expansion capability similar to an "app store." This will significantly enhance Claude's flexibility and scalability in handling complex tasks.

The emergence of Conway is seen as a key strategic move by Anthropic in the AI agent domain, aiming to transform Claude from a passive conversational tool into a persistent assistant capable of continuous operation and autonomous execution of multi-step tasks. It not only supports browser automation and event triggering but also strengthens code generation, execution, and problem-solving capabilities through native integration of Claude Code.

Industry observers believe that Conway has the potential to become a strong competitor to rivals like OpenClaw, driving the advancement of AI agents toward an "always-on" direction. The era where Claude can truly "show its prowess" may be just around the corner.

Domande pertinenti

QWhat is the name of the persistent agent solution being developed by Anthropic for Claude?

AThe persistent agent solution being developed by Anthropic for Claude is named Conway.

QHow does Conway differ from a traditional chat interface?

AConway differs by being an independent UI instance that operates as a standalone agent workspace, capable of directly controlling the browser, connecting to external connectors, and integrating Claude Code functionality.

QWhat method does Conway support to allow external services or events to trigger its work?

AConway supports being invoked via Webhook, which allows external services or events to trigger the agent's work and enhance automated response capabilities.

QWhat is the name of the extension standard that will allow developers to build custom tools for Conway?

AThe extension standard that developers can use to build custom tools, UI tabs, and context handlers for Conway is called the CNW ZIP standard.

QHow is Conway expected to transform the role of Claude according to the article?

AConway is expected to transform Claude from a passive conversational tool into a persistent assistant that can run continuously and autonomously execute multi-step tasks, moving towards an 'always-on' AI agent direction.

Letture associate

XRP Ledger v3.3.0 Release Brings Five Amendments Into Focus

The upcoming XRP Ledger v3.3.0 release, scheduled for early August, focuses on five key amendments: Confidential MPT, Batch, Permission Delegation, Sponsored Fees and Reserves, and Dynamic MPT. This bundle aims to enhance the ledger's infrastructure for advanced use cases like tokenized assets and improve operational flexibility and user experience. Notably, Sponsored Fees could reduce onboarding friction by allowing third parties to cover transaction costs, while Confidential and Dynamic MPT features provide more control for institutional asset tokenization. Batch transactions and Permission Delegation add operational efficiency for applications. However, these features do not activate automatically upon release; they require approval from 80% of network validators. The upgrade represents a significant step in XRPL's roadmap for enterprise and developer utility, with practical impact dependent on subsequent validator consensus and builder adoption.

bitcoinist12 min fa

XRP Ledger v3.3.0 Release Brings Five Amendments Into Focus

bitcoinist12 min fa

Wall Street Morning News: V-shaped Rebound at Month-end, but Nasdaq Suffers Worst July in 12 Years; Funds Accelerate Concentration Towards Cloud Giants

Despite a V-shaped rebound at the end of July, the Nasdaq posted its worst July since 2004, while the S&P 500 had its worst July since 2014. Markets were jolted by geopolitical shifts, as President Trump canceled a planned strike on Iran, leading WTI crude to plunge over 8%. This, alongside OPEC+ announcing a supply increase, reversed crude's sharp July gains. Treasury yields surged, with the 10-year yield rising over 30 basis points in July—its largest July increase since 2005. In a rare move, the US and Japan jointly intervened to weaken the USD/JPY, aiming to prevent potential Japanese sales of US Treasuries. While the tech sector faced deleveraging pressure throughout July, cloud giants staged a massive rally on strong earnings. Microsoft, Amazon, and Google collectively added nearly $1.5 trillion in market value last week. Amazon soared over 15% on accelerating AWS growth, Microsoft extended historic gains, Google fully recovered post-earnings losses, and Meta ended an 11-day losing streak. In contrast, Apple tumbled over 7% on supply chain and guidance concerns, ceding its "world's most valuable company" title to Nvidia. The memory and storage sector corrected sharply. Gold edged up 0.91% in July, with analysts viewing the ~30% pullback from January highs as a potential basing period, supported by long-term central bank demand. Key events to watch this week include earnings from Palantir, AMD, SpaceX (its first post-IPO report), and memory giants like Western Digital. The US July non-farm payrolls report on Friday will be critical for gauging the Fed's policy path. SpaceX also faces a significant lock-up expiration, testing market liquidity.

marsbit36 min fa

Wall Street Morning News: V-shaped Rebound at Month-end, but Nasdaq Suffers Worst July in 12 Years; Funds Accelerate Concentration Towards Cloud Giants

marsbit36 min fa

Rubin Ultra Makes Major Cuts, Even Nvidia Can't Handle Memory Price Hikes?

NVIDIA's Rubin Ultra, the top-tier variant of the newly announced Rubin AI accelerators, has reportedly seen significant specification downgrades, according to an industry report from SemiAnalysis. Initially designed with four compute dies (4-die), the Rubin Ultra is now said to be reduced to a 2-die design. Key changes highlighted in the report include: * **No increase in peak theoretical compute performance**, remaining at 35 PFLOPs like the standard Rubin. * **Severe reduction in memory capacity** to 192GB using 8-Hi HBM stacks, which is less than the standard Rubin's 288GB using 12-Hi stacks. * **Negligible memory bandwidth improvement** of only 1 TB/s. * **Slightly higher chip-level power consumption**. * The **primary upgrade is a massive increase in scale-up interconnect capacity**, supporting connections for up to 576 GPUs via NVLink, compared to 72 for the standard Rubin. The report suggests the redesign is primarily a cost-optimization move driven by the sharp rise in HBM (High-Bandwidth Memory) prices. By reducing the expensive HBM content and shifting investment towards enhanced system-scale networking, NVIDIA aims to maintain the platform's value for large-scale AI training clusters while managing soaring material costs. The news reportedly triggered a sell-off in South Korean memory stocks, with SK Hynix and Samsung shares falling around 8%, as markets grew concerned that NVIDIA—a major HBM buyer—might be reducing its reliance on high-capacity memory, potentially capping future pricing power for memory makers.

Odaily星球日报57 min fa

Rubin Ultra Makes Major Cuts, Even Nvidia Can't Handle Memory Price Hikes?

Odaily星球日报57 min fa

Upbit Rebalances 864B SHIB In Internal Wallet Move

South Korean cryptocurrency exchange Upbit moved a total of 864 billion SHIB (worth approximately $4 million) between its internal wallet addresses. The activity, involving transfers of 384 billion SHIB outbound and 480 billion SHIB inbound, occurred following a 36% SHIB price rally. On-chain data indicates this was a routine internal wallet rebalancing operation by the exchange, not a sell-off or liquidation event. The article cautions that while large on-chain transfers are highly visible, they require proper context and wallet labeling to interpret correctly. In this case, the movement between known Upbit addresses does not inherently signal a change in market supply or selling pressure for SHIB.

bitcoinist1 h fa

Upbit Rebalances 864B SHIB In Internal Wallet Move

bitcoinist1 h fa

Can Generative Models Finally Be Trained End-to-End? The Core Is a For Loop

This article introduces a novel training paradigm for generative models called Explorative Modeling (XM), which enables true end-to-end training. Traditionally, powerful generative models like autoregressive and diffusion models are not trained end-to-end. They are trained to predict a single small step but require iterative multi-step sampling for inference. This "exposure bias" leads to error accumulation and limits performance. The core challenge XM addresses is "mode blurring." In generative tasks, a single input (e.g., "generate a dog") corresponds to many valid outputs (multiple modes). Standard training objectives like reconstruction loss force the model to average these modes, producing unrealistic, blurry outputs. To avoid this, existing models break generation into many small, almost deterministic steps, sacrificing end-to-end training. XM tackles this by restructuring the training loop itself. Its key insight is to amplify "generative expressivity." For each training input, instead of generating one sample, the model generates K candidate outputs. Only the candidate closest to the real data is used for computing the loss and updating the model via backpropagation. This simple "best-of-K" mechanism is implemented as a short for-loop. By exploring multiple possibilities, the model learns to distribute its guesses across different modes rather than collapsing to their uninformative average. The paper demonstrates that "exploration" acts as a new, powerful scaling axis. Gains from XM increase with model size, data scale, and compute. Experiments show improvements in FID scores for image generation and significant efficiency gains, sometimes outperforming larger models without exploration. When pushed to the limit, XM enables fully single-step, end-to-end generative models. In robotics tasks, an "Explorative Policy" matched the performance of a 100-step Diffusion Policy with a single forward pass, drastically improving inference speed. While the best-of-K concept is not entirely new, the authors' contribution lies in formally understanding it as a direct method to boost generative expressivity without fragmenting the generation process. This work suggests that as models scale, enhancing exploration during training may become crucial for overcoming fundamental performance bottlenecks.

marsbit1 h fa

Can Generative Models Finally Be Trained End-to-End? The Core Is a For Loop

marsbit1 h fa

Trading

Spot