AI Agents Unlocked: Overcoming Web Barriers to Transform Digital Interaction

The promise of artificial intelligence has long captivated the collective imagination, evolving from science fiction to the tangible reality of sophisticated algorithms. For years, AI’s primary role for consumers was to answer questions, generate content, or automate basic tasks within specific applications. Today, however, we stand at the precipice of a much more profound transformation: the emergence of AI agents capable of taking delegated actions across the entire web. Imagine an AI that doesn't just tell you the best flight options but books them for you, negotiates the price, and integrates it seamlessly into your calendar. This vision, once distant, is rapidly becoming a reality, heralding a new era of digital convenience and efficiency.

Yet, as these advanced AI agents step out from behind the chat window and attempt to interact with the vast network of online services, they encounter an immediate and unexpected roadblock: websites are frequently blocking them. This isn't a minor technical hiccup; it's the defining challenge of the current AI revolution for consumers, threatening to fragment the very experience agents promise to streamline. As a groundbreaking U.S.-centric TechCrunch story highlighted on October 6, 2026, and updated on October 19, the "next hurdle for AI agents" is unequivocally "getting websites to let them in." This crucial insight shifts the focus from the internal intelligence of AI models to the external infrastructure, revealing that the decisive battle for consumer AI’s future will be fought not over processing power or algorithmic elegance, but over permission, authentication, liability, and standardized communication protocols between agents and businesses.

The Maturation of AI Agents: From Chatbots to Digital Delegates

The evolution of AI assistance for consumers has been swift and relentless. We’ve moved far beyond the rudimentary chatbots of a decade ago, which were largely confined to answering pre-programmed FAQs or guiding users through simple decision trees. Modern AI agents are a different breed entirely. They leverage large language models (LLMs) and sophisticated reasoning capabilities to understand complex instructions, break them down into multi-step tasks, and execute those tasks across various digital environments.

These capabilities are transforming the digital landscape, offering users the potential for unprecedented levels of automation and convenience. AI agents can now perform a dizzying array of functions: searching for specific products across multiple retailers, comparing prices and features, booking flights and hotels based on intricate preferences, reserving tables at restaurants, ordering groceries, managing email correspondence, and accurately filling out complex online forms. The vision is clear: to offload tedious digital chores to an intelligent assistant, freeing up human time and mental energy.

Early indicators suggest a strong public appetite for this advancement. While the deployment of these sophisticated AI agents remains constrained by issues of website access and, crucially, consumer trust, the traffic generated by agentic systems is already rising sharply. Cheq, an online security firm, reported a staggering 100-fold increase in agentic traffic over just 45 days, though it's important to note that agents still represent a relatively small fraction of overall online shoppers. This rapid escalation, however, underscores the profound shift occurring in how some users are beginning to interact with the web. Further solidifying this curiosity, Meta’s Muse agent reportedly surpassed 2.5 million downloads within its first two weeks, demonstrating a significant public interest in exploring the potential of these new digital tools. This initial enthusiasm, however, exists in tension with a lingering reluctance among many users to grant agents access to sensitive personal data, payment information, and critical online accounts – a barrier that website blocking only exacerbates.

The ultimate promise of these AI agents is to become an entirely new interface layer between consumers and online services. Instead of navigating dozens of different websites, each with its unique UI and workflow, a user could simply articulate a goal to their agent, and the agent would seamlessly handle the underlying interactions. This could democratize access, simplify complex tasks, and fundamentally alter how we experience the digital economy. But for this vision to materialize, the fundamental issue of website access must first be resolved.

The Unseen Barrier: Why Websites Block AI Agents

The core problem, as the TechCrunch story highlights, is that despite their advanced capabilities, AI agents are frequently met with digital roadblocks. Companies like Amazon have already been reported to block Meta’s Muse agent, while users themselves are experiencing restrictions when attempting to use agents on popular services such as Yelp, eBay, Zillow, Pizza Hut, Adidas, and various airlines. This widespread resistance isn't arbitrary; it stems from a complex interplay of existing web security paradigms, commercial interests, and legitimate concerns about liability and data privacy.

1. The Enduring "Bot" Problem: Good Bots vs. Bad Bots

The internet has long been a battleground against malicious automated programs. For decades, websites have developed sophisticated mechanisms to identify and block "bots" – automated scripts designed to scrape data, send spam, engage in credential stuffing attacks, launch Distributed Denial of Service (DDoS) attacks, or exploit vulnerabilities. These malicious bots represent a significant threat to security, data integrity, and operational stability for online businesses.

Current bot detection systems rely on various indicators:

• User-Agent Strings: The HTTP User-Agent header, which identifies the client software originating a request (e.g., "Mozilla/5.0 (Windows NT 10.0; Win64; x64) AppleWebKit/537.36 (KHTML, like Gecko) Chrome/100.0.4896.75 Safari/537.36"), is a primary identifier. Malicious bots often use spoofed or generic user agents, or user agents that clearly identify them as automated. Legitimate AI agents face the challenge of being perceived similarly if their user-agent strings are not appropriately designed or recognized.
• Behavioral Analysis: Websites analyze patterns of interaction. Bots often exhibit non-human behavior – extremely fast browsing, repetitive actions, accessing pages in an illogical sequence, or clicking in precise, unnatural locations. Distinguishing sophisticated AI agents that mimic human behavior from malicious bots becomes exceedingly difficult with these methods.
• IP Reputation: Blocking IP addresses associated with known botnets or suspicious activity.
• CAPTCHAs and reCAPTCHAs: These challenges are specifically designed to differentiate humans from automated programs, posing a significant hurdle for AI agents that cannot interpret and solve them reliably.

The fundamental dilemma is that from a technical perspective, a sophisticated AI agent performing a delegated task can appear remarkably similar to a malicious bot attempting to scrape data or exploit a service. Until there's a reliable, standardized way to distinguish an authorized, benevolent AI agent from its malicious counterpart, websites will err on the side of caution, blocking anything that deviates from expected human interaction patterns.

2. Commercial and Economic Concerns

Beyond security, websites have strong commercial incentives to control who (or what) accesses their platforms and how.

• Data Ownership and Value: Websites invest heavily in collecting, organizing, and presenting data (e.g., product listings, flight prices, restaurant reviews). They see this data as proprietary and valuable. AI agents, particularly those designed to aggregate information from multiple sources, could be perceived as bypassing the intended user experience or even "scraping" valuable data without permission or proper attribution.
• Customer Journey and Revenue Models: Many online services rely on specific customer journeys to drive revenue. This could involve displaying advertisements, guiding users through affiliate links, or encouraging direct purchases within their own ecosystem. An AI agent that bypasses these pathways, perhaps by directly accessing an API (if available) or by swiftly navigating to a checkout page without viewing ads, could disrupt these revenue streams. For instance, an airline wants a user to book directly on their site, not via an agent that aggregates data from competitors and might not prioritize their specific deals.
• Brand Experience and Control: Companies spend significant resources crafting their brand experience, user interface, and messaging. An AI agent might present information or complete tasks in a way that diverges from the brand's intended presentation, potentially diluting the brand experience or even misrepresenting the service.
• Competitive Disadvantage: If certain AI agents, especially those linked to competing platforms (like Meta's Muse on Amazon), gain unrestricted access, it could be seen as granting a competitive advantage or enabling data extraction that benefits a rival.

3. Security and Liability Considerations

The increased autonomy of AI agents introduces significant new security risks and complex legal questions of liability.

• Unauthorized Access: If an AI agent, acting on a user's behalf, gains access to sensitive accounts, what are the implications if that access is exploited or compromised? Websites are responsible for protecting user data and preventing unauthorized access, and allowing unverified agents could create new attack vectors.
• Who is Liable for Agent Actions? This is perhaps the most pressing legal concern. If an AI agent makes a mistake – books the wrong flight, orders the wrong product, or makes a fraudulent purchase – who bears the responsibility? Is it the user who delegated the task, the developer of the AI agent, or the website that processed the agent's action? Existing legal frameworks are largely designed for human-to-human or human-to-system interactions and are ill-equipped to handle the nuances of agentic autonomy. This ambiguity creates a strong incentive for businesses to block agents until clear liability frameworks are established.
• Data Privacy and Compliance: AI agents often require access to personal data, account credentials, and even payment information to perform their delegated tasks. How do these agents handle and store this sensitive data? Websites must comply with stringent data privacy regulations like the California Consumer Privacy Act (CCPA) in the U.S. (and GDPR internationally). Allowing unknown agents to interact with their systems creates new compliance challenges and potential privacy breaches.

The reluctance of companies like Yelp, eBay, Zillow, Pizza Hut, Adidas, and various airlines to embrace AI agents is therefore multifaceted. It’s not simply a technological oversight; it’s a strategic decision rooted in protecting their platforms, data, revenue, and legal standing in an evolving digital landscape.

The Shifting Battleground: From Intelligence to Infrastructure

The emergence of AI agents has fundamentally altered the landscape of consumer AI. The decisive battle, as highlighted by TechCrunch, is no longer solely about building more intelligent models or refining algorithms. That foundational work continues, but the immediate bottleneck has shifted to the infrastructure that governs how these intelligent agents can actually operate in the real world. The new front lines are defined by permission, authentication, liability, and standardized agent-to-business communication. If these critical systems are not reliably established, the grand vision of AI agents as a seamless interface layer will remain just that – a vision, fragmented by incompatible access protocols and stifled by mistrust.

1. Permission: The Gateway to Action

At its core, permission defines what an AI agent is authorized to do and where. This isn't just about a user giving consent for their agent to act; it's about websites explicitly granting permission for agents to interact with their services.

• Granular Consent: The current model of "agree to terms and conditions" is insufficient. Users need granular control over their agents: What types of tasks can it perform? Which specific websites can it access? How much data can it share?
• Website Whitelisting/Blacklisting: Websites need mechanisms to explicitly permit known, authorized agents while continuing to block malicious ones. This moves beyond simply trying to guess if a visitor is human or bot.
• Dynamic Permissions: Permissions might need to be dynamic, changing based on the context of a task or the user's real-time instructions.

2. Authentication: Proving Identity and Authority

Once permission is conceptually granted, how does a website verify that an AI agent is who it claims to be, and that it genuinely represents an authorized user?

• Agent Identity: An authorized AI agent needs a digital identity that websites can verify. This goes beyond a simple user-agent string. It might involve digital certificates, verifiable credentials, or cryptographic keys.
• User Authorization: The website also needs assurance that the agent is acting with the explicit and current authorization of a legitimate user. This could involve secure protocols similar to OAuth, where a user grants an agent specific scopes of access to their account on a particular website, without the agent ever directly handling the user's password.
• Beyond Passwords: Traditional username/password authentication is not designed for agentic interaction and poses security risks if agents are storing or transmitting these credentials. A more secure, token-based system is imperative.

3. Liability: Defining Responsibility in the Age of Agents

The question of who is responsible when an AI agent makes a mistake or facilitates fraud is a monumental legal and ethical challenge.

• Errors and Omissions: If an AI agent misinterprets an instruction and books the wrong flight, or makes an accidental purchase, who bears the financial cost?
• Fraud and Abuse: If an agent's credentials are compromised, or if a malicious actor leverages an agent, who is liable for any fraudulent transactions or data breaches?
• Regulatory Frameworks: New laws and regulations may be needed to assign liability to AI agent developers, platform providers, or the end-users, much like product liability laws govern physical goods. Without clear legal precedents, businesses will remain hesitant to open their gates to agents.

4. Standardized Agent-to-Business Communication

For AI agents to truly operate as a seamless interface layer, they cannot rely solely on "screen scraping" or reverse-engineering website UIs. This approach is brittle, prone to breaking with minor website updates, and often blocked. A robust, scalable future requires a standardized language or protocol for agents to communicate with online services.

• Structured APIs: The ideal solution would be for websites to offer well-documented, machine-readable Application Programming Interfaces (APIs) specifically designed for agent interaction. These APIs would allow agents to programmatically access data, submit forms, and complete transactions in a structured, secure, and efficient manner.
• Common Protocols: Beyond individual APIs, the industry needs common protocols or schema that agents can understand and use across multiple services, even if the underlying website implementations differ. This is where an "open standard" becomes critical. It would define a common way for agents to announce themselves, request permissions, authenticate, and exchange data securely.

If these four pillars – permission, authentication, liability, and standardized communication – can be reliably established, AI agents have the potential to unlock a new paradigm of digital interaction. If they remain fragmented and unresolved, consumer AI will be perpetually limited by the walls of individual websites, failing to live up to its transformative promise.

Towards a Solution: The Open Standard Initiative

Recognizing the critical nature of this infrastructure bottleneck, a consortium of major players is stepping forward to address the challenge head-on. The TechCrunch report highlights a significant development: companies including Meta, Walmart, Stripe, Sierra, Genesys, Rocket, NiCE, and Decagon are actively developing an open standard specifically designed to distinguish authorized user agents from malicious bots. This collaborative effort represents a pivotal moment, aiming to lay the groundwork for a more permissive and secure environment for AI agents.

The Vision of the Open Standard:

The primary goal of this initiative is to create a universally recognized and adopted framework that allows websites to confidently identify legitimate AI agents. This standard would facilitate secure, permissioned interactions, fostering trust between agent developers, users, and the businesses operating online services.

• Secure Identification: The standard would likely involve cryptographic methods or verifiable digital credentials to establish the identity and authenticity of an AI agent. This could be akin to how SSL/TLS certificates verify websites, providing a robust mechanism for agent validation.
• Permissioned Access: Building upon the concept of granular permission, the standard could define mechanisms for agents to request specific scopes of access, and for websites to grant or deny those requests based on their policies and the user’s authorization.
• Standardized Communication Protocol: The standard would likely include guidelines or a formal protocol for how agents should interact with websites, preferably through dedicated APIs rather than relying on UI automation. This would ensure stable, efficient, and machine-readable data exchange.
• Distinguishing Intent: Crucially, the standard aims to differentiate between a benevolent AI agent acting on a user's behalf (e.g., booking a flight) and a malicious bot (e.g., scraping prices or attempting a DDoS attack). This distinction is vital for websites to open their doors selectively.

Technical Considerations for the Standard:

The development of such an open standard will involve significant technical considerations:

• Beyond robots.txt: The current robots.txt protocol allows websites to request that bots not crawl certain pages, but it's purely advisory and easily ignored by malicious actors. The new standard needs to be enforceable and bidirectional, allowing websites to verify agent identity and agents to communicate their intent and capabilities.
• Specific HTTP Headers: The standard might define new HTTP headers that agents must include in their requests, containing encrypted tokens or verifiable credentials. Websites could then check these headers to authenticate the agent.
• Digital Certificates/Blockchain IDs: Each authorized agent could be issued a digital certificate or a unique, verifiable identity on a blockchain, providing a decentralized and tamper-proof way for websites to confirm its legitimacy.
• API-First Approach: A critical component of the standard would be to encourage (and provide guidance for) websites to develop dedicated APIs for agent interaction, rather than relying on agents to mimic human browser behavior. This provides greater control, security, and stability.
• Interoperability: The standard must be designed for maximum interoperability across diverse web technologies and platforms, preventing fragmentation.

Challenges and Hurdles for the Open Standard:

While the initiative is promising, achieving widespread adoption and functionality presents significant challenges:

• Broad Industry Adoption: Securing buy-in from thousands (or millions) of websites, especially smaller businesses, will be difficult. The standard needs to be easy to implement and offer clear benefits to websites.
• Security and Abuse Prevention: The standard itself must be robust against spoofing, exploitation, and abuse. A compromised standard would undermine its entire purpose. Constant vigilance and updates will be necessary.
• Balancing Control and Utility: The standard must strike a delicate balance between giving websites sufficient control over their platforms and enabling AI agents to provide valuable utility to users. Overly restrictive standards will limit agent functionality, while overly permissive ones will invite abuse.
• Neutrality and Vendor Lock-in: As an "open" standard, it must remain neutral and avoid favoring any particular AI agent developer or platform. This prevents vendor lock-in and encourages a competitive ecosystem.
• Evolving Threat Landscape: The nature of malicious bots is constantly evolving. The standard needs to be adaptable to new threats and attack vectors.

The collaborative development of this open standard, spearheaded by a diverse group of tech giants and innovative companies, signifies a crucial recognition that individual efforts won't suffice. The future of consumer AI hinges on a shared infrastructure, and this initiative represents the most promising path forward for overcoming the "permission problem."

The Promise of a Permitted Future: Agents as a New Interface Layer

If the challenges of website access, authentication, permission, and liability can be effectively resolved through an open standard, the potential benefits for consumers, businesses, and the broader digital economy are immense. AI agents could indeed become the seamless, intuitive interface layer envisioned, transforming online interactions from tedious chores into effortless commands.

1. Enhanced Consumer Experience:

• Hyper-Personalization: Agents, with appropriate permissions, could learn user preferences at a profound level, anticipating needs and offering truly tailored recommendations and services – from travel plans to financial advice.
• Unprecedented Convenience and Time Savings: Imagine an agent managing all your subscriptions, finding the best deals on groceries, scheduling appointments, or handling returns across various e-commerce sites, all without you having to manually navigate each platform. This liberates significant user time.
• Accessibility: For users with disabilities, agents could bridge interaction gaps, making the web far more accessible by automating complex navigation or input tasks.
• Optimized Decision Making: Agents can quickly aggregate and analyze information from myriad sources, helping users make more informed decisions on purchases, investments, or services by cutting through noise and bias.

2. Benefits for Businesses:

• New Customer Acquisition Channels: Businesses that embrace agent access could reach new customer segments who prefer agent-driven interactions or who might have been deterred by complex UIs.
• Reduced Customer Service Load for Routine Tasks: Many customer service inquiries are repetitive. Agents could handle these routine interactions, freeing up human agents for more complex, high-value problem-solving.
• Data-Driven Insights (with consent): If users permit, agent interactions could provide businesses with aggregated, anonymized insights into consumer behavior and preferences, informing product development and marketing strategies.
• Innovation Opportunities: The ability for agents to interact seamlessly with services opens up entirely new categories of agent-enabled applications and services, fostering innovation in the digital economy.
• Improved Conversions: For tasks like purchasing, an agent could guide users through the process more efficiently, potentially leading to higher conversion rates for businesses.

3. Economic Impact in the U.S.:

• Productivity Gains: By automating tasks that consume significant human time, AI agents could contribute to substantial productivity gains across various sectors of the U.S. economy.
• Economic Growth: The creation of new agent-enabled services and the improved efficiency of existing ones could stimulate economic growth and job creation in the AI sector and related industries.
• Competitive Advantage: U.S. companies that successfully implement and leverage these agent technologies could gain a significant competitive edge globally, shaping the future of digital commerce and services.

This future, however, is not guaranteed. It hinges on successfully navigating not just the technical hurdles, but also the equally complex human elements of trust, privacy, and control.

Overcoming the Remaining Hurdles: Trust, Privacy, and Control

Even if an open standard effectively solves the technical problem of website access, the broader adoption of AI agents will still face significant non-technical barriers. The initial strong consumer curiosity demonstrated by Meta’s Muse downloads exists alongside a profound reluctance to grant agents access to payments, accounts, and personal data. Addressing these concerns is paramount for widespread trust and adoption.

1. Consumer Trust:

• Transparency: Users need to clearly understand what their agent is doing, why, and how. Opaque "black box" operations will breed mistrust. Agent UIs must be designed to show their thought process and actions.
• Reliability and Accuracy: Agents must perform tasks reliably and accurately. A single significant error can erode trust quickly.
• Ethical AI Design: Agent developers must adhere to strong ethical guidelines, ensuring agents act in the user's best interest, avoid manipulation, and are free from bias.
• Security Assurance: Users need strong assurances that their data and accounts are secure when handled by an agent. This involves robust encryption, secure authentication, and clear security audits.

2. Data Privacy:

• Minimization: Agents should only access and process the minimum amount of data necessary to perform a task.
• Data Residency and Storage: Where is user data stored by the agent? Who has access to it? These questions become even more critical with agents acting across multiple services.
• Compliance with Regulations: AI agent platforms operating in the U.S. must be meticulously compliant with privacy regulations such as the CCPA, ensuring users have rights regarding their data, including access, deletion, and the right to opt-out of data sales. Future regulations specifically addressing AI agent data handling may also emerge.
• User Control Over Data: Users must have absolute control over what data their agent can access, store, and share, and with whom.

3. User Control:

• Granular Permissions: As discussed, users need fine-grained control over their agents' capabilities – specifying which websites, what types of actions, and under what conditions.
• "Kill Switch" Functionality: Users must have the ability to instantly pause or terminate an agent's operation, revoking all permissions, should they lose trust or change their mind.
• Human Oversight and Intervention: While agents are autonomous, users should always have the option to review, modify, or override an agent's proposed actions. The system should be a partnership between human and AI, not a complete surrender of control.
• Auditing and Logging: Users should be able to view a clear log of all actions their agent has taken, providing full transparency and accountability.

The successful integration of AI agents into daily life will depend not just on technological prowess, but on the careful cultivation of user trust through transparent design, robust privacy safeguards, and empowering control mechanisms.

The U.S. Context: Regulatory Landscape and Market Dynamics

The U.S.-centric nature of this challenge and its proposed solutions carries significant implications for innovation, regulation, and consumer protection within the American market. The rapid pace of AI development in the U.S., particularly by companies like Meta, Amazon, Google, and Microsoft, means that the solutions developed here could set global precedents.

• Regulatory Scrutiny: As AI agents become more prevalent, U.S. regulatory bodies such as the Federal Trade Commission (FTC), the National Institute of Standards and Technology (NIST), and potentially Congress, are likely to weigh in. Issues around consumer protection, data privacy, market competition, and liability will fall under their purview. There's a strong likelihood of new guidelines or even legislation aimed at governing AI agent behavior and business responsibilities.
• Competition and Innovation: The U.S. is a hotbed for AI innovation. The development of an open standard for agent access could either foster healthy competition among agent developers and online services or, if poorly designed, could inadvertently create new monopolies or barriers to entry. The diverse group of companies involved in the standard's development suggests an initial commitment to openness, but long-term vigilance will be necessary.
• Consumer Protection: U.S. consumers stand to benefit immensely from AI agents, but also face new risks regarding data privacy, potential scams, or errors. Robust consumer protection frameworks, both within the standard itself and through government oversight, will be crucial to ensure trust and widespread adoption.
• Economic Leadership: By solving this critical infrastructure bottleneck, the U.S. could solidify its position as a leader in the global AI economy, setting standards that influence how AI agents interact with the web worldwide.

The U.S. has the opportunity to lead in defining the secure and ethical operational norms for AI agents, transforming the digital landscape for its citizens and influencing global technological trajectories.

Conclusion

The journey of AI agents from rudimentary assistants to powerful digital delegates capable of performing multi-step actions across the web is one of the most exciting technological narratives of our time. This evolution promises unparalleled convenience, personalization, and efficiency for consumers, fundamentally altering how we interact with online services and manage our digital lives. However, as the U.S.-centric TechCrunch story from October 2026 so acutely articulates, this transformative vision faces an immediate and formidable obstacle: the pervasive blocking of these AI agents by the very websites they seek to serve.

This isn't merely a technical glitch; it's a profound strategic challenge that shifts the decisive battle for consumer AI from internal model intelligence to the external infrastructure governing permission, authentication, liability, and standardized communication. Websites, driven by legitimate concerns over malicious bots, commercial interests, security vulnerabilities, and ambiguous liability, currently err on the side of caution, creating a fragmented digital landscape that stifles agent adoption. The experiences of users encountering restrictions on services from Amazon to Yelp, eBay, and airlines underscore the urgency of this dilemma.

The collaborative initiative by companies like Meta, Walmart, Stripe, and others to develop an open standard for distinguishing authorized user agents from malicious bots represents a beacon of hope. This standard, once widely adopted, holds the key to unlocking the full potential of AI agents by providing the necessary mechanisms for secure identification, granular permission, and standardized communication protocols. If successful, it could pave the way for AI agents to truly become the new, seamless interface layer between consumers and the online world, ushering in an era of unprecedented digital efficiency and personalization.

Yet, technical solutions alone are not sufficient. The widespread adoption of AI agents also hinges on building and maintaining profound consumer trust. This requires unwavering commitments to transparency, robust data privacy safeguards in compliance with U.S. regulations like CCPA, and empowering users with absolute control over their agents' actions and data access. The complex questions of liability, too, must be resolved through clear legal and ethical frameworks to ensure accountability in this rapidly evolving domain.

The future of consumer AI in the U.S. and globally now critically depends on resolving this "permission problem." If we can collectively navigate these intertwined technical, commercial, legal, and ethical hurdles, AI agents will fulfill their promise, transforming our digital interactions from fragmented tasks into a harmonious symphony of delegated action. If not, the profound potential of AI agents will remain largely untapped, confined by the digital walls of a web unprepared for its most intelligent visitors. The stakes are immense, and the urgent work of defining how AI agents are "let in" is now the most critical frontier for the next generation of artificial intelligence.