Booking.com Launches Lola AI Travel Agent as Omnissa, Komprise and CData Expand Agent Controls

According to Engadget, Lola promises personalized recommendations, live availability and pricing, direct booking, preferred rates and benefits. It can recommend and reserve a hotel, restaurant, events, activities or an entire trip. Booking.com's launch partners include ResX, CLEAR, BLADE, Aero, SeatGeek, GetYourGuide, Nuitee and Ten Group, along with Booking's brands Booking.com, OpenTable, KAYAK, Priceline, Agoda and FareHarbor. The app searches real-time availability and pricing, lets users compare options and book in the same conversation, understands natural language, asks follow-up questions and learns about users to make recommendations more personal.

Lola is free, but some services require a subscription. The Plus plan costs $5 per month and VIP costs $25 per month, both with 12-month commitments. Subscribers get up to 15% discounts on four- and five-star hotels, complimentary Gold status for OpenTable reservations, priority access to events, premium airport services and a personal concierge, according to Engadget. Engadget noted that Meta's Muse arrived first with similar capabilities and that its launch caused travel booking stocks to fall over fears it could bypass dedicated travel sites. AI agents can also make mistakes: a law firm cited an AI concierge that booked a non-refundable hotel in Naples, Florida, instead of Naples, Italy, where the traveler was flying, and AI assistants can show a 'confirmed' status while a booking is still pending.

SiliconANGLE reported that at its Omnissa ONE 2026 conference in Orlando, Omnissa LLC unveiled AI agents for IT teams and virtual desktop users, a managed cloud PC service and an AI governance product called Omnissa Elara. The company said employees are adopting AI tools faster than IT departments can approve them. Its State of Digital Workspace 2026 report found that use of AI assistant apps across enterprise endpoints grew nearly 1,000% in 2025, with roughly three-quarters coming from unsanctioned tools. Chief Product Officer Bharath Rangarajan said AI "is compressing the distance between insight and action," and companies need to see and control what AI is doing before giving it more autonomy.

Three of the new agents are for IT administrators. The Digital Employee Experience agent traces a worker's problem to its root cause and recommends a fix based on existing playbooks or earlier Workspace ONE Assist support sessions. The Windows App Lifecycle agent handles application packaging and tests each app before an administrator reviews it. The third works with Workspace ONE Vulnerability Defense, ranking vulnerabilities by priority and preparing patches, with a human kept in the loop before deployment. Omnissa said large language models are speeding up flaw discovery and the time to exploit a flaw has dropped below a day, so the agent is meant to help IT teams patch quickly. Administrators can also connect outside agent tools through a hosted server built on the Model Context Protocol, with access limited by existing roles and permissions.

On the virtual desktop side, Horizon Delegate runs inside an employee's Horizon virtual machine and carries out tasks under that worker's approved identity and permissions, and the work can continue after the employee disconnects. Omnissa Cloud PC is a desktop-as-a-service offering that includes underlying compute plus Omnissa's endpoint management and digital employee experience tools. Elara is pitched as "the authority layer" for AI governance and high-impact enterprise actions, pulling signals from separate systems so IT and security leaders can see how AI apps, models and agents are used and apply policies before consequential actions go through. Phil Hochmuth, research vice president for endpoint management and enterprise mobility at International Data Corp., said IT leaders need visibility into human and AI-driven activity and controls to govern actions at machine speed. Omnissa did not give availability dates; Cloud PC will run first on Amazon Web Services, with other cloud providers to follow. Omnissa was spun out of VMware in 2024 after Broadcom acquired VMware and sold its end-user computing business to KKR for $4 billion.

SiliconANGLE reported that Komprise Inc. launched its Universal File MCP tool to address what co-founder and President Krishna Subramanian called "MCP bloat." MCP, developed by Anthropic, gives AI agents a common standard to connect with third-party data and software, but Subramanian said almost every technology company has released a dedicated MCP server, overloading AI with multiple tool definitions and lowering accuracy and speed. He said MCP bloat also raises the cost of running agents by increasing token consumption, citing a McKinsey study that found almost 60% of agentic token consumption is driven by "response refinement," the behind-the-scenes token loops where sub-agents process data repeatedly.

Komprise said Universal File MCP rightsizes AI responses so only the most important unstructured data is fed to agents. It identifies the right data sources for a request, enriches information with context and governs it with user-specific access permissions. The tool builds on Komprise capabilities including the Global Metadatabase for a consistent schema across storage resources and AI Preparation & Process Automation for extracting contextual metadata. Noise filters eliminate irrelevant data before agents process it. Subramanian said a clinician could ask a large language model to find a set of pathology images, with the query filtering by KAPPA-enriched context and user permissions, while an information security professional could discover non-compliant files or search archived research data by project keywords. Komprise said the tool works with any data source, whether managed by Komprise or not. Todd Dorsey, an analyst at Data Center Intelligence Group, said Komprise is simplifying how AI accesses unstructured data at scale and making it more efficient.

SiliconANGLE reported that CData Software Inc. introduced Connect AI Gateway, a platform for controlling how AI agents choose models, use tools and access company data. The product entered early access and extends CData's Connect AI platform and managed MCP offering. It gives IT teams one place to register models, MCP servers and agents and to set rules for what each can do. A context engine draws on connected system structures, business definitions and knowledge from user interactions, which CData said can improve answer accuracy while limiting data exposure and model costs. The gateway enforces a person's permissions when an agent retrieves data or acts on that person's behalf, with policies for specific rows and columns and an audit trail showing which prompt, model, tool and policy contributed to a response.

CData said the context engine can reuse definitions maintained in tools such as Fivetran's dbt and Microsoft's Power BI, as well as information about fields and relationships in source systems. It can capture less formal knowledge from documents, conversations and corrections to agent responses, holding it in a context graph outside individual AI models so an organization can apply it when changing models. People can review the graph and decide what becomes shared context. Marie Forshaw, CData's senior vice president of product marketing, said, "You don't want AI redefining revenue every time you ask a question about revenue." The gateway includes routing and control functions for models, MCP tools and agents, and can set token budgets and route requests according to policies. Will Davis, CData's chief marketing officer, said routing will initially be policy-based and eventually intelligent enough to select the lowest-cost model for a task. CData said its data layer can filter, join and aggregate records before sending results to a model to reduce context window use. In a company-run test of 378 enterprise queries, Connect AI answered 98.5% correctly, compared with 65% to 75% for other MCP providers it tested, and a separate test found a cost difference of up to 175-fold between models that produced the same correct answer. Those are CData benchmarks, not independent tests or documented customer savings. CData said the context engine is not intended to replace a full data catalog or semantic layer, and human review will govern which lessons from agent interactions become shared knowledge.