evolvedigital.ai blog

Tag: AI Agents

  • Anthropic Launches Claude Sonnet 5: The Most Capable Mid-Tier AI Model Yet

    Anthropic Launches Claude Sonnet 5: The Most Capable Mid-Tier AI Model Yet

    Anthropic released Claude Sonnet 5 on June 30, 2026, marking one of the company’s most significant mid-tier model launches to date. The new model is now the default for every Free and Pro plan user worldwide, and it represents a meaningful step toward closing the performance gap between frontier and mid-tier AI systems. With an IPO widely expected later this year, the release also signals Anthropic’s intent to compete aggressively with OpenAI and Google across both consumer and enterprise markets.

    What Was Announced

    Anthropic officially introduced Claude Sonnet 5 on June 30, 2026, positioning it as a direct successor to Sonnet 4.6. The model is available as the default experience for users on Free and Pro plans, and is also accessible to Max, Team, and Enterprise subscribers. Developers can access it immediately through the Claude API using the model identifier claude-sonnet-5.

    The launch came with a notable introductory pricing offer: $2 per million input tokens and $10 per million output tokens through August 31, 2026. After that window closes, standard pricing kicks in at $3 per million input tokens and $15 per million output tokens. This initial discount makes Sonnet 5 one of the most cost-effective options in its performance class.

    Alongside the model itself, Anthropic increased rate limits across its core products, including Claude Chat, Claude Cowork, Claude Code, and the API Platform. The company also deployed an updated tokenizer that delivers better performance, though it introduces a token mapping change of approximately 1.0 to 1.35 times the previous count, which developers will need to account for in production systems.

    Anthropic also confirmed that cyber safeguards are enabled by default on Sonnet 5, continuing the company’s focus on responsible deployment as its models grow more capable in autonomous and agentic contexts.

    Technical Details

    Claude Sonnet 5 is described by Anthropic as the most agentic Sonnet model ever built. It can formulate multi-step plans, use external tools such as web browsers and terminals, and operate autonomously across extended workflows. This positions it well above previous Sonnet releases in terms of practical utility for software development, research automation, and business process tasks.

    According to Anthropic, Sonnet 5’s performance approaches that of the flagship Opus 4.8 model on many benchmark categories, while carrying a substantially lower price tag. The model demonstrates measurable improvements over Sonnet 4.6 in reasoning, coding, tool use, and knowledge work. Anthropic also noted a reduction in hallucination rates and sycophancy compared to its predecessor, addressing two of the most commonly cited reliability concerns in enterprise deployments.

    One area where Sonnet 5 intentionally remains constrained is offensive cybersecurity. Anthropic confirmed the model is substantially weaker than Opus-class models on tasks involving the development of working exploits, a deliberate design boundary consistent with the company’s safety commitments.

    Industry Impact and Reactions

    The release places pressure on OpenAI’s GPT-4o series and Google’s Gemini mid-tier lineup. By bringing near-frontier-level agentic capability into a model that defaults to free users, Anthropic has moved the baseline of what consumer AI can do. The introductory pricing strategy also makes Sonnet 5 immediately attractive to startups and individual developers who previously would have needed to budget for larger, more expensive models to achieve comparable results.

    The timing of the release is notable. Anthropic has been expanding its enterprise partnerships and is widely reported to be preparing for an IPO later in 2026. Launching a capable, affordable model that becomes the new standard for tens of millions of users is a direct mechanism for growing the active user base and strengthening the company’s revenue story ahead of a public offering.

    More broadly, the release reinforces a trend visible across the AI industry in 2026: the rapid compression of the performance gap between mid-tier and frontier models. Each generation of mid-tier releases from Anthropic, OpenAI, and Google has arrived closer to the frontier than the last, and Claude Sonnet 5 is a clear example of that pattern accelerating.

    What Comes Next

    Developers building on Sonnet 5 should note the August 31, 2026 pricing transition date. Applications launched at introductory pricing will see a cost increase once standard rates take effect, so planning for that change now is advisable. Anthropic has not announced a specific roadmap for what follows Sonnet 5 in the mid-tier lineup, though the company’s release cadence suggests continued iteration through the second half of 2026.

    For enterprise customers, the increased rate limits and the addition of Claude Cowork and Claude Code support make Sonnet 5 a strong candidate for large-scale agentic deployments. As autonomous AI workflows become more common in software development and business operations, the ability to run capable agents at lower cost and higher throughput will be a significant factor in vendor selection.

    Conclusion

    Claude Sonnet 5 represents a meaningful shift in what mid-tier AI is capable of. By making near-flagship performance available as the default experience for all Claude users, Anthropic has raised the floor for the entire industry. For businesses evaluating AI platforms, for developers building production applications, and for individual users looking for more capable tools, Sonnet 5 is a release worth paying close attention to.

    Stay updated on the latest AI news at Evolve Digital.

  • Google Brings Computer Use to Gemini 3.5 Flash: AI Agents Can Now See, Reason, and Act Across Platforms

    Google Brings Computer Use to Gemini 3.5 Flash: AI Agents Can Now See, Reason, and Act Across Platforms

    Google has officially integrated computer use capabilities into Gemini 3.5 Flash, turning one of its most widely deployed AI models into a platform for building autonomous agents that can see, reason, and act across digital environments. Announced on June 24, 2026, this update represents a significant expansion of what developers can build with the Gemini API. The computer use feature, previously available only through a separate standalone Gemini 2.5 computer use model, is now a native built-in tool within Gemini 3.5 Flash, making it accessible to the full ecosystem of developers and enterprises already using the Flash model. The move marks a pivotal moment in the maturation of AI agent capabilities from research preview to production infrastructure.

    What Was Announced

    Google’s announcement centers on the integration of computer use directly into Gemini 3.5 Flash via the Gemini API and the Gemini Enterprise Agent Platform. This means developers no longer need to work with a separate, purpose-built computer use model. Instead, the same Gemini 3.5 Flash model they use for text, code, and multimodal tasks can now be directed to interact with browser, mobile, and desktop environments as a built-in capability.

    A demo environment has been made available through Browserbase, allowing developers to explore the capability in a sandboxed setting. Google has also published a reference implementation on GitHub for teams looking to get started quickly with their own agent deployments. Both resources are intended to accelerate the path from experimentation to production for developers building automation workflows.

    Enterprise partners including Browserbase, Browser Use, and UiPath were cited in the announcement as early collaborators and endorsers of the capability. The involvement of UiPath in particular signals a meaningful convergence between traditional robotic process automation tooling and AI-native computer use, two approaches to enterprise automation that are now increasingly complementary.

    Google stated that computer use in Gemini 3.5 Flash delivers improved performance for long-horizon and enterprise automation tasks compared to earlier iterations. Performance improvements were noted on OSWorld benchmarks, which are a standard evaluation framework for AI systems performing computer use tasks across operating system interfaces.

    Technical Details

    The computer use capability in Gemini 3.5 Flash is built on the model’s ability to process screenshots and visual representations of digital interfaces and then generate precise, coordinated actions to accomplish multi-step tasks. Agents built on this foundation can navigate web browsers, interact with mobile applications, and operate desktop software without requiring custom API integrations for each application or platform. This makes the capability particularly well suited for automating tasks in legacy software environments where native APIs are not available.

    To address the security risks inherent in deploying agents that take real-world actions in live environments, Google applied targeted adversarial training specifically designed to reduce the model’s susceptibility to prompt injection attacks. Prompt injection, in which malicious content embedded in a web page, document, or application interface attempts to redirect agent behavior, is among the most serious risks in real-world computer use deployments. Google’s targeted training approach aims to make the model more robust against this class of attack.

    Two optional enterprise safeguard systems were released alongside the model update. The first requires the agent to obtain explicit user confirmation before taking any action that is sensitive or irreversible, preserving a human-in-the-loop checkpoint for workflows where the cost of an error is high. The second automatically halts agent execution if an indirect prompt injection attempt is detected, providing an automated safety layer for organizations running agents at scale across untrusted environments. Google also recommends combining these systems with secure sandboxing, strict access controls, and human verification practices as part of a comprehensive deployment strategy.

    Industry Impact and Reactions

    Bringing computer use into a mainstream, widely available model like Gemini 3.5 Flash is a meaningful shift in the accessibility of AI agent capabilities. Until recently, computer use required developers to work with specialized, purpose-built models that were often in preview or limited-access phases. By embedding the capability directly into Flash, Google is signaling that computer use is ready for production, not just experimentation, and it is lowering the barrier for organizations that want to build autonomous agents as part of their core technology stack.

    The partnership with UiPath is particularly significant for enterprise adoption. UiPath has an established base of customers using robotic process automation to handle software interfaces that do not expose APIs, including in industries such as healthcare administration, financial services, and legal operations. Combining UiPath’s enterprise distribution and workflow tooling with Gemini’s AI-native computer use capabilities could accelerate automation in segments of the market that have historically been difficult to reach with purely code-driven approaches.

    The announcement also reflects a broader industry trend toward bundling safety and security tooling with agent capabilities rather than treating them as separate, optional concerns. By releasing enterprise safeguards alongside the computer use feature itself, Google is acknowledging that agent security is a first-class deployment requirement and positioning Gemini as a platform that takes production readiness seriously.

    What Comes Next

    Access to computer use in Gemini 3.5 Flash is available immediately through the Gemini API and the Gemini Enterprise Agent Platform. Developers can explore the capability via the Browserbase demo environment and the reference implementation on GitHub. Google has not announced a separate pricing tier for computer use within the Flash model, suggesting it will be accessible within existing Gemini 3.5 Flash API pricing structures, though enterprise platform access may carry distinct terms.

    Looking ahead, the integration is likely to serve as a foundation for further expansion as Google continues its June 2026 model rollout. Gemini 3.5 Pro, Google’s frontier model for the month, is expected to ship before the end of June. Bringing computer use to the Pro tier would be a natural next step, enabling more complex, long-horizon autonomous tasks at a higher level of model intelligence and reasoning depth.

    Conclusion

    Google’s integration of computer use into Gemini 3.5 Flash marks a clear turning point in the availability of AI agent capabilities for developers and enterprises. By moving computer use from a standalone model to a built-in feature of one of its most accessible APIs, and by releasing enterprise safeguards alongside the launch, Google has made autonomous digital agents a practical choice for production deployment. For organizations evaluating how to embed AI into their workflows beyond text generation and code assistance, this announcement opens a meaningful new set of possibilities.

    Stay updated on the latest AI news at Evolve Digital.

  • OpenAI Codex Is Now a Full Desktop Agent That Can Control Your Mac Even When Locked

    OpenAI Codex Is Now a Full Desktop Agent That Can Control Your Mac Even When Locked

    OpenAI has transformed Codex from a cloud-based code-running tool into a persistent desktop agent capable of operating a Mac computer autonomously — including while the screen is locked. The capability, confirmed in late May 2026 by multiple sources including MacRumors, Macworld, and TechTimes, represents one of the most significant shifts yet in how AI agents interact with personal computers. For the first time, users can assign tasks to an AI system and walk away confident it will continue working through the night, on a scheduled basis, or in response to real-world triggers.

    What Was Announced

    OpenAI confirmed that Codex, its autonomous coding and task agent, now supports a “locked computer use” mode on macOS. When enabled, Codex can continue operating Mac applications even after the display has been locked, using an Apple authorization plugin that temporarily grants it access to the screen and input systems. The feature is available to Codex subscribers in the United States and is activated through the Codex desktop app settings, requiring explicit user opt-in along with Screen Recording and Accessibility permissions.

    Alongside the locked Mac capability, OpenAI announced that Codex has gained the ability to follow users across devices. A task started on a Mac can be monitored and managed from a connected mobile phone, allowing users to check progress, receive alerts, or redirect the agent while away from their desk. Codex can also now capture and analyze screen content over time to build what OpenAI describes as “ambient memory,” giving the agent contextual awareness of what has happened on the machine between sessions.

    Scheduled task execution rounds out the update. Users can instruct Codex to perform recurring jobs at specific times, a capability that effectively transforms it into a persistent background worker rather than an on-demand tool. The combination of locked-screen operation, cross-device access, and scheduling marks a qualitative leap: Codex is no longer a tool you run, it is an agent that runs on your behalf.

    Technical Details

    The locked Mac feature depends on a new Apple authorization plugin that ships with the Codex desktop app. When a user enables “Locked computer use” in Codex settings, the plugin installs at the system level and negotiates short-lived credentials with macOS that allow Codex to temporarily access the display, mouse, and keyboard interfaces. Once local input is detected, such as a user moving the mouse or pressing a key, the authorization expires immediately and the screen relocks. OpenAI describes this as a “relock on local input” safeguard, designed to prevent the agent from continuing to act if a human is present at the machine.

    Additional safeguards built into the system include covered display mode, which prevents visual output from the agent’s actions from being visible on the screen during locked use, and manual-unlock fallback, which reverts full control to the human user at any point. Certain system areas are explicitly off-limits: Codex cannot automate the Terminal application, cannot interact with its own interface, and cannot trigger system-level administrator prompts. These restrictions are enforced at the plugin level, not just through software policy.

    The agent’s screen-capture capability for ambient memory uses a rolling context window that logs what applications were open, what content was visible, and what actions were taken across sessions. This gives Codex the ability to resume complex multi-step tasks without requiring the user to restate context. The cross-device continuity is handled through OpenAI’s cloud infrastructure, with the Mac acting as the local compute environment and the phone serving as a remote management interface.

    Industry Impact and Reactions

    The announcement arrives in the middle of a broader industry race to build practical, persistent AI agents. Google’s Gemini Spark, announced at Google I/O on May 19, 2026, similarly positions itself as a 24/7 agent running on dedicated cloud virtual machines. Anthropic’s Claude has gained agentic capabilities through its computer use API. What distinguishes the Codex locked-Mac feature is that it operates locally on the user’s own hardware rather than requiring the cloud to spin up a virtual environment, which has implications for latency, privacy, and cost.

    The developer and power-user community has responded with a mix of genuine excitement and measured caution. The ability to have an AI continue working on a codebase, document, or research task overnight without requiring an open laptop or active session removes a meaningful friction point for professional workflows. At the same time, security researchers have begun examining what new attack surfaces are introduced by a system that can bypass the locked-screen boundary under any circumstances, even with safeguards in place. The feature’s absence in the European Economic Area, the United Kingdom, and Switzerland pending regulatory review signals that OpenAI anticipates scrutiny in jurisdictions with stricter data protection frameworks.

    The broader competitive context matters here. AI labs are no longer competing only on benchmark scores or raw model capability. They are now competing on how deeply their agents can integrate into users’ daily computing environments. An agent that keeps working while you sleep is a different value proposition than one that answers questions. This shift from reactive assistant to proactive coworker is reshaping how enterprises and individual professionals think about AI adoption.

    What Comes Next

    OpenAI has not published a detailed roadmap for Codex’s agentic capabilities, but the pattern of recent releases suggests continued expansion. Cross-platform support beyond macOS is a likely next step, particularly for Windows, which represents the majority of enterprise desktop environments. The company has also signaled interest in deeper integration with development tools and cloud services, which would allow Codex to coordinate actions across local and remote environments as part of single workflows. Regulatory approvals in the EEA, UK, and Switzerland will be required before the locked-use feature reaches those markets.

    For the AI industry overall, the locked-Mac feature from Codex and the 24/7 cloud agents from Google represent a convergence toward the same end goal: AI systems that are always available, always aware, and capable of sustained independent action. The next twelve months will likely determine whether this model becomes the dominant paradigm for professional AI tools or whether safety and privacy concerns prompt a course correction.

    Conclusion

    OpenAI’s Codex has crossed a threshold that seemed distant just a year ago: an AI agent that can operate a personal computer continuously, independently, and without requiring the user to be present. The technical safeguards built into the locked-screen feature reflect a genuine effort to make this capability responsibly deployable, while the geographic restrictions acknowledge that regulators will need time to assess the implications. What is clear is that the era of AI as a passive question-answering tool is ending. The question now is not whether AI agents will run persistently in the background of professional computing, but how quickly that becomes the default.

    Stay updated on the latest AI news at Evolve Digital.

  • Apple Is Working to Bring AI Agents to the App Store Ahead of WWDC 2026

    Apple Is Working to Bring AI Agents to the App Store Ahead of WWDC 2026

    Apple is developing a system to incorporate AI agents into the App Store, according to a report from 9to5Mac published on May 13, 2026. The move, which has not been officially confirmed by Apple, would represent a significant expansion of how third-party AI capabilities are surfaced to iPhone and iPad users, and is expected to be previewed at WWDC 2026 on June 8 alongside Apple broader iOS 27 and artificial intelligence announcements.

    What Happened

    According to the report, Apple is working on a new system internally described as Extensions, which would allow users to access generative AI capabilities from installed apps on demand, through existing Apple Intelligence features such as Siri, Writing Tools, Image Playground, and more. Under this system, third-party apps that include AI agents would be able to surface those agents through the App Store and integrate them into the Apple Intelligence layer, rather than operating only within the boundaries of their own apps.

    This would create a new category of App Store listing: not just apps, but AI agents that can be invoked across the operating system. A user might download an agent from a developer that specializes in contract summarization, for example, and invoke it through Siri or Writing Tools whenever they are working with legal documents, regardless of which app they are currently using. Models from Google and Anthropic are reportedly being tested in this context, consistent with earlier reporting that iOS 27 will allow users to choose from multiple AI models as the backend for Siri.

    Why It Matters

    If Apple implements an open AI agent marketplace on the App Store, it would mark one of the most significant changes to the App Store model since its launch. Currently the App Store distributes software. Adding AI agents as a distinct category would make it a marketplace for AI capabilities, and Apple curation and distribution infrastructure would apply to AI in the same way it currently applies to apps.

    For AI developers, an App Store channel would provide access to Apple installed base of over two billion active devices, with the trust and discoverability that Apple platform provides. For users, it would mean AI agents are as easy to find and install as apps, rather than requiring separate accounts, subscriptions, or technical setup. The competitive implications for standalone AI companies and for Apple own Siri are significant, as the system would simultaneously empower third-party AI and place Apple at the center of how users discover and manage it.

    What Comes Next

    WWDC 2026, beginning June 8, is the expected venue for Apple to formally announce its AI agent strategy for iOS 27. The event will also include the major Siri overhaul that has been in development, and the two announcements, a restructured Siri and an AI agent marketplace, are likely to be presented as complementary parts of a broader Apple Intelligence vision for the year ahead.

    Stay updated on the latest AI news at Evolve Digital.

  • Anthropic Gives Claude Agents a Dreaming Capability to Self-Improve Between Sessions

    Anthropic Gives Claude Agents a Dreaming Capability to Self-Improve Between Sessions

    Anthropic announced three new features for Claude Managed Agents on May 7, 2026, with the most notable being a capability the company is calling dreaming. The feature allows autonomous Claude agents to review their past sessions, identify patterns in how they have performed tasks, and use those observations to improve their behavior in future sessions — a form of offline self-refinement that does not require continuous human instruction. The announcement marks a step toward agents that become meaningfully more capable through use rather than requiring periodic retraining by their developers.

    What Happened

    The dreaming capability gives Claude Managed Agents access to structured summaries of their previous sessions, which they can review during idle periods to extract lessons and update their internal guidelines for handling similar situations in the future. Anthropic describes the feature as a research preview, indicating it is being made available to a limited set of enterprise and developer customers for evaluation before broader rollout.

    Alongside dreaming, Anthropic announced increased rate limits for Claude Code users, doubling the five-hour weekly usage limit for Pro, Max, and Enterprise subscribers. The company also announced improvements to how Managed Agents handle long-running multi-step tasks across domains including coding, finance, and legal work. These updates position Managed Agents as Anthropic primary vehicle for enterprise agentic deployments.

    Why It Matters

    The dreaming capability represents a meaningful architectural evolution for autonomous AI agents. Current AI systems improve primarily through deliberate retraining on new data, a process that requires significant engineering resources and does not happen automatically based on an agent operational experience. Dreaming enables a lighter-weight form of improvement that happens between sessions, allowing agents deployed in production to gradually refine their approaches to recurring task types.

    The practical implications for enterprise deployments are significant. A Claude agent running routine coding or financial analysis workflows could, through dreaming, develop increasingly optimized approaches to the specific patterns it encounters most frequently — without requiring its operators to monitor every session or manually update its instructions. This degree of autonomous self-improvement is one of the key capabilities that distinguishes a capable long-term agent from a simple task executor.

    Stay updated on the latest AI news at Evolve Digital.

  • Amazon Wins Court Order Blocking Perplexity AI Shopping Bots on Its Marketplace

    Amazon Wins Court Order Blocking Perplexity AI Shopping Bots on Its Marketplace

    A federal court ruled on March 10, 2026, that Perplexity AI must immediately stop using its Comet web browser agent to make purchases on behalf of shoppers on Amazon marketplace. The injunction, granted at Amazon request, marks a significant legal development at the intersection of AI agents, consumer identity, and e-commerce fraud law.

    What Happened

    Amazon filed a lawsuit accusing Perplexity of committing computer fraud by deploying Comet to shop on Amazon without clearly disclosing that the activity was being performed by an AI agent rather than a human user. The core legal argument is that Perplexity Comet browser agent violated computer fraud statutes by accessing Amazon systems under false pretenses — presenting as an ordinary browser session when it was, in fact, an automated agent acting on behalf of a third party.

    The court sided with Amazon in granting the preliminary injunction, ordering Perplexity to halt Comet activity on Amazon marketplace while the broader lawsuit proceeds. The case is the latest in a series of legal challenges that AI agent products have faced as they enter consumer commerce. Perplexity Computer, which launched in February 2026, uses Comet to execute multi-step agentic tasks including web shopping on behalf of users.

    The ruling does not affect other Perplexity products or its search functionality, but it does temporarily remove one of the most visible use cases that the company had been promoting for its new agentic platform.

    Why It Matters

    The Amazon versus Perplexity case raises fundamental questions about how AI agents that act on behalf of users will be regulated in commercial environments. Marketplaces like Amazon have terms of service that govern automated access, and the question of whether an AI shopping agent is acting as the user or as a separate entity is not yet settled in law.

    The outcome could affect the entire category of AI consumer agent products. If courts determine that AI agents must explicitly identify themselves when conducting transactions, it would require significant changes to how products like Perplexity Computer, and similar offerings from other companies, operate in commerce contexts. The case is expected to proceed to a full trial, with the preliminary injunction in place until a final ruling is reached.

    Stay updated on the latest AI news at Evolve Digital.