AI Enhanced Version

This piece is a part of VentureBeat's special issue, "AI at Scale: From Vision to Viability". It explores the transformative impact of edge computing on artificial intelligence (AI) and how it is reshaping AI as we know it. As AI expands beyond centralized data centers, we're witnessing a shift in the processing of complex tasks. From smartphones running intricate language models locally to smart devices processing computer vision at the edge and autonomous vehicles making real-time decisions without cloud connectivity, AI's reach is expanding. The focus in the AI space has traditionally been on training, which requires powerful machines clustered together to handle large workloads. However, we're on the brink of AI workloads shifting from training to inference, with edge computing becoming the dominant paradigm. Edge computing will bring inference closer to users, either directly on devices like autonomous vehicles or at the network edge, making AI an integral part of daily life. However, this shift towards edge computing won't necessarily reduce cloud usage. Instead, edge AI is driving increased cloud consumption, revealing an interdependency that could reshape enterprise AI strategies. New research from Hong Kong University of Science and Technology and Microsoft Research Asia demonstrates the depth of this dependency and highlights the crucial role the cloud plays in data storage, processing, and model training for AI. The researchers created a test environment that mirrored real-world enterprise deployments. Their three-layer architecture revealed the computational demands at each level. They also developed new compression techniques for AI workloads, achieving remarkable efficiency and preserving high-quality results while dramatically reducing bandwidth requirements. The research findings provide a roadmap for organizations planning AI deployments. The results suggest that deploying AI solely at the edge or solely in the cloud leads to significant compromises in performance and reliability. Successful edge AI deployments require deep integration between edge and cloud resources, sophisticated orchestration layers, and new approaches to data management. The research also challenges common assumptions about edge AI requirements and presents a compelling case for hybrid cloud-edge architectures. It shows that enterprises can leverage existing edge computing providers to achieve similar benefits without building complex systems from scratch. The convergence of edge computing and AI is more than a technical evolution - it's a fundamental restructuring of the AI infrastructure economy. Three transformative shifts are likely to reshape enterprise AI strategy: infrastructure arbitrage in AI deployment, an emerging capability paradox in edge AI deployment, and the rise of orchestration capital. These insights suggest that the next wave of AI innovation will come from increasingly sophisticated approaches to orchestrating the interaction between edge and cloud resources. The enterprises that will thrive in this new landscape will be those that develop deep competencies in orchestration intelligence, or the ability to dynamically optimize complex hybrid systems for maximum value creation. This represents a fundamental shift in how we think about competitive advantage in the AI era, moving from a focus on ownership and control to a focus on optimization and orchestration.

AI Enhanced Version

This article dives into the exciting and rapidly evolving world of Artificial Intelligence (AI) infrastructure, as part of VentureBeat’s special issue, “AI at Scale: From Vision to Viability.” As we look towards 2025, businesses can anticipate a wave of new opportunities and strategic decisions surrounding the crucial task of establishing a robust foundation for AI expansion. The introduction of AI-specific hardware, such as new chips, accelerators, co-processors, servers, and other networking and storage devices, promises to alleviate current shortages, enhance performance, broaden service range and availability, and accelerate time-to-value. The ongoing emergence of this specialized hardware is projected to sustain the double-digit growth in AI infrastructure that has continued for 18 consecutive months, according to IDC. The IT firm reveals that organizational purchases of compute hardware (mainly servers with accelerators) and storage hardware infrastructure for AI have risen by 37% year-over-year in the first half of 2024. Sales are expected to triple to $100 billion annually by 2028. A significant 42% of new AI spending worldwide through 2025 is anticipated to be on dedicated and public cloud infrastructure for AI, says Mary Johnston Turner, research VP for digital infrastructure strategies at IDC. These substantial figures underscore the crucial role of infrastructure in driving AI growth and digital transformation within enterprises. As a result, technology and business leaders are encouraged to prioritize AI infrastructure in their strategic, tactical, and budget planning for 2025. Anay Nawathe, director of cloud and infrastructure delivery at ISG, a global research and advisory firm, emphasizes that successful implementation of generative AI relies on smart investment and a robust infrastructure. He cites a recent ISG survey that found a lack of scalability in current infrastructure, which prevents organizations from deploying AI at scale, despite having numerous projects in the pilot phase and limited deployment. Industry experts concur that when scaling AI pilots, proofs of concept, or initial projects, it’s crucial to select deployment strategies that balance scalability, performance, price, security, and manageability. To assist enterprises in developing their AI expansion infrastructure strategy, VentureBeat has sought insights from over a dozen CTOs, integrators, consultants, and other seasoned industry experts, as well as numerous recent surveys and reports. These insights, coupled with selected resources for further exploration, can guide organizations on the most intelligent path to leverage new AI hardware and drive operational and competitive advantages. However, most new AI-ready hardware is expensive and targeted at large data centers, making it less accessible to most enterprises. Instead, these products are likely to be purchased by hyperscalers like Microsoft, AWS, Meta, and Google; cloud providers such as Oracle and IBM; dedicated AI firms; and major colocation companies like Equinix. Despite the challenges, DIY infrastructure for AI is expected to grow in popularity, especially as hardware vendors release new, customizable AI-specific products. This will encourage more mature organizations to deploy purpose-built, finely tuned, proprietary AI in private clouds or on-premises for better performance, risk management, data protection, cost control, and specific workload management. The optimal near-term strategy for most enterprises navigating the new infrastructure paradigm will likely mirror current cloud approaches: An open, “fit-for-purpose” hybrid that combines private and public clouds with on-premises and edge. As AI continues to evolve, experts caution against getting locked into any service provider or technology. Instead, they recommend multi-tenancy environments, which distribute AI infrastructure, data, and services across two or more cloud providers, as a sensible strategy for enterprises. In conclusion, as AI continues to evolve rapidly, it's crucial for enterprises to stay flexible, follow open standards and decouple storage from compute. This allows organizations to quickly adopt new models and technologies as they emerge, rather than waiting for their vendor to catch up. The freedom to experiment without architectural constraints is key to maintaining agility and staying ahead in this rapidly evolving landscape.

AI Enhanced Version

Retrieval-augmented generation (RAG) is a commonly used method for tailoring large language models (LLMs) to specific informational needs. However, it has its drawbacks, including technical complexity and slower processing times. Recent advancements in long-context LLMs now allow for an alternative approach, bypassing RAG by incorporating all necessary information directly into the prompt. A study from Taiwan's National Chengchi University highlights the potential of using long-context LLMs and caching techniques to create tailored applications that outperform traditional RAG pipelines. This new method, known as cache-augmented generation (CAG), could serve as a more efficient and straightforward alternative to RAG, particularly in business environments where the knowledge base can be accommodated within the model’s context window. RAG is typically used for managing open-domain queries and specialized tasks. It employs retrieval algorithms to collect relevant documents, providing context for the LLM to generate more precise responses. However, this process introduces latency and depends heavily on the quality of document selection and ranking, which can negatively impact user experience and retrieval efficiency. In contrast, the CAG approach proposes to eliminate the complexity of RAG by incorporating the entire document corpus into the prompt, allowing the model to determine the relevance of the information. This method not only simplifies the process but also mitigates issues caused by retrieval errors. However, this approach also presents challenges. Long prompts can slow down the model and increase inference costs. The model's context window length also limits the number of documents that can be included in the prompt. Moreover, the inclusion of irrelevant information can confuse the model and degrade the quality of its responses. To overcome these obstacles, the CAG approach leverages three key trends: advanced caching techniques, long-context LLMs, and improved training methods. Advanced caching techniques expedite the processing of prompt templates, and long-context LLMs allow for the inclusion of more documents and knowledge into prompts. Enhanced training methods enable models to better handle retrieval, reasoning, and question-answering tasks on long sequences. As future models continue to expand their context windows, they will be able to process larger knowledge bases, improving their ability to extract and utilize relevant information from long contexts. This evolution will broaden the applicability of the CAG approach, making it a robust and versatile solution for knowledge-intensive tasks. To test the effectiveness of RAG and CAG, the researchers conducted experiments on two widely recognized question-answering benchmarks: SQuAD and HotPotQA. The experiments demonstrated that CAG generally outperformed RAG, especially in scenarios where RAG might retrieve incomplete or irrelevant passages. However, CAG is not a one-size-fits-all solution and should be used judiciously. It is most effective in settings where the knowledge base is relatively static and small enough to fit within the model's context window. Businesses should also be cautious when their documents contain conflicting information based on the context, which could confuse the model during inference. In conclusion, the best way to ascertain if CAG is suitable for your use case is to conduct a few experiments. Implementing CAG is straightforward and should be considered as an initial step before investing in more development-intensive RAG solutions. VB Daily provides insights into how companies are utilizing generative AI, offering valuable information for maximizing return on investment.

AI Enhanced Version

Dizzaract Games, a new game development studio based in Abu Dhabi, has officially launched as the largest of its kind in the Middle East. With a team of 160, Dizzaract aims to revolutionize gaming by integrating artificial intelligence (AI), immersive gameplay, and blockchain technology. This aligns with Abu Dhabi's strategic initiative to support Web3 gaming through the establishment of Beam Ventures, a $140 million fund dedicated to this emerging sector. The studio has developed patented AI frameworks that elevate traditional gaming into adaptive, interactive experiences. These patented technologies allow games to respond dynamically to player actions, creating a unique experience in each session. Dizzaract's founder, Ilman Shahzaev, highlights the importance of AI in gaming, envisioning a future where AI agents adapt to players' decisions and emotions in real-time, blurring the line between gaming and reality. The studio's flagship game, Farcana, is a team-based hero shooter developed on Unreal Engine 5. The game, which raised $10 million in its seed round, features 24 unique characters, diverse tactical roles, and advanced AI integration. As the first high-budget game developed in the UAE, Farcana aims to set a new standard for triple-A production quality in the region. The game also incorporates Web3 features, allowing players to own in-game assets and participate in blockchain-powered economies. Dizzaract's patented AI frameworks represent a significant advancement in the gaming industry. These technologies enable real-time analysis of player behavior, adjusting gameplay, non-player character (NPC) interactions, difficulty levels, and in-game economies to align with each player’s unique journey. This approach fosters deeper engagement and keeps players immersed in the gameplay. Moreover, Dizzaract's vision extends beyond traditional gaming. By integrating AI with Web3, the studio is building decentralized economies where players can truly own their in-game assets and identities. Shahzaev envisions a future where digital ownership becomes standard, and Web3 features become the norm in gaming, allowing players to transfer achievements, avatars, and assets across multiple games seamlessly. Shahzaev believes that by 2025, gaming will enter what he calls the "AI Supercycle," with games dynamically responding to players' actions, decisions, and emotions in real-time. He envisions an AI-driven in-game economy where prices, resources, and events are adjusted based on player behavior and market trends. He also sees the integration of digital ownership and AI as a catalyst for the creation of entirely new markets for game assets and experiences. In Shahzaev's view, AI will play a dominant role in Web3 gaming development in 2025. He predicts that AI agents will manage in-game economies, providing real-time insights and keeping systems dynamic and responsive. He also imagines a future where players own their in-game assets and identities, transferring achievements seamlessly across games while participating in economies that mirror real-world markets. For instance, an avatar from one game could carry its skills, history, and assets into another, creating a unified digital identity. Shahzaev believes that AI-driven Web3 games will dominate the market by combining immersive gameplay with decentralized economies, where in-game economies, powered by AI, will be as dynamic and complex as real-world markets. He envisions a future where every player's experience is uniquely tailored, from storylines to gameplay mechanics, creating deeply personal journeys. This convergence of AI and Web3 will also open entirely new markets for assets, experiences, and even professions within gaming. With the launch of Dizzaract Games, the Middle East is set to become a significant hub for technological and creative innovation in the gaming industry.

AI Enhanced Version

AI Enhanced Version

Last year marked the advent of artificial intelligence (AI) agents in the engineering field, with the introduction of a product named Devin. This groundbreaking AI engineer, developed by a startup led by Scott Wu, was initially kept under wraps but has now been made widely available. The AI agent is rapidly learning and evolving, as demonstrated by the recent release of Devin 1.2, which brings a host of new capabilities that elevate its ability to manage entire development projects. One of the most noteworthy enhancements in Devin 1.2 is its refined in-context reasoning, which enhances the agent's proficiency in handling and reusing code. The update also incorporates the ability to receive voice messages via Slack, providing users with a more intuitive way to instruct Devin. This innovation comes amidst growing enthusiasm for AI-powered agents, which are increasingly being hailed as the future of modern work. Industry experts predict an imminent era where humans and AI agents will collaborate, with the latter taking over repetitive tasks. This trend is already taking root, as recently highlighted by Nvidia's CEO at CES, who forecasted that enterprise IT departments would evolve into "HR departments" for AI, responsible for managing and maintaining AI agents working across different company functions. While not a major update, Devin 1.2 introduces intriguing features that enhance the agent's performance. The most significant improvement is the enhanced ability to reason in context within a code repository. This means Devin can now better comprehend the structure and content of a repository, enabling it to identify relevant files for specific tasks, recognize and reuse existing code and patterns, and suggest edits or create pull requests more accurately, thus reducing errors and manual adjustments. This capability accelerates workflows for developers and alleviates the cognitive burden associated with searching for files, understanding codebases, or rectifying inconsistent code. Another notable update in Devin 1.2 is the introduction of voice messages. Users can now give voice commands to Devin via Slack by simply tagging Devin in a Slack chat, hitting the "Record audio clip" button, and describing the task or feedback. Devin then prepares a step-by-step action plan and executes the command using its developer tools. This feature simplifies user interaction with the agent, eliminating the need to type natural-language prompts into Devin’s chatbot-style interface. The latest release also includes usability enhancements. For example, machine snapshots have been introduced to streamline the login process for Devin’s workspace. Additionally, enterprise accounts have been introduced, providing organization admins with a centralized console to manage multiple Devin workspaces, member access controls, and billing. Furthermore, a usage-based billing model has been implemented, allowing users to pay for extra capacity beyond their subscription limits. This feature, active since January 9, allows users to set their additional usage budgets according to their needs, ensuring control over spending while guaranteeing uninterrupted service when additional capacity is needed. Currently, Devin is widely available for engineering assistance at a starting price of $500 a month, with no seat limits. Several enterprises, including Lumos, OpenSea, Curai Health, Nu Bank, and Ramp, have already integrated it into their workflows. The release of Devin's new capabilities comes at a time when competition in the AI engineering space is rapidly intensifying. From the widespread adoption of GitHub Copilot to significant funding received by Magic and Poolside AI for the development of advanced capabilities, the race to create the ultimate AI coding assistant is well underway. These AI-powered agents are not only transforming how developers work but also shaping the future of modern work, where efficiency and innovation are driven by a partnership between humans and machines. By 2028, Gartner estimates that 33% of enterprise software applications will include AI agents, enabling autonomous decision-making in 15% of day-to-day work. This trend underscores the growing importance of AI in the workplace, and the need for businesses to stay informed about developments in AI technology.

AI Enhanced Version

Runway, a leading provider of AI media technology, has launched Frames, a cutting-edge text-to-image generation model. The new tool is already earning accolades for its ability to produce strikingly cinematic visuals, a testament to Runway's reputation as a trusted AI video model provider. This has led industry insiders to speculate if Frames could become the preferred choice for AI filmmakers and artists. Introduced in November 2024, Frames was initially accessible only to selected ambassadors and power users of the Runway Creators Program. However, as of now, it is available to everyone via Runway’s Unlimited and Enterprise subscription plans. A gallery of images generated by Frames users, titled "Worlds of Frames," can be viewed on the company's website. Frames offers an unprecedented level of stylistic control and visual precision, making it a versatile tool for various sectors including editorial, art direction, brand development, pre-visualization, and production. The model’s prompting system allows for detailed and in-depth results, enabling users to create nuanced, naturalistic, and cinematically composed visuals. Users have praised the model's high quality, style control, and quick animation capabilities. Frames stands out from previous models for its ability to maintain stylistic consistency, allowing users to establish a distinct visual identity for their projects and generate variations that stay true to that specific style. At launch, Frames provides 19 preset visual styles that users can select and further customize. These presets showcase the broad range of creative possibilities the tool offers, making it an ideal choice for artists, designers, and filmmakers seeking stylistic precision. Frames also excels in rendering advanced textures, natural lighting, and complex compositions, offering more flexibility compared to the rigid outputs of earlier image generation models. This is just the first version of Frames, with Runway planning future updates that will include more style tools and controls. Runway prioritizes safety and ethical responsibility in its generative AI tools. As part of the company's Foundations for Safe Generative Media initiative, Frames includes robust content moderation features to prevent misuse. The company's in-house visual moderation system detects and blocks harmful or inappropriate content, balancing creative freedom with safety. To address concerns about misinformation and misuse, Frames embeds invisible watermarks in all AI-generated content. These watermarks comply with provenance standards set by the Coalition for Content Provenance and Authenticity (C2PA), allowing users to trace whether a media item is AI-generated. Runway is also committed to improving fairness and representation in its generative models. Efforts have been made to reduce bias in visual outputs and to support diverse demographics and languages. However, Runway is currently involved in a lawsuit from human artists who allege that the company, along with others like Stability AI and Midjourney, used their artwork without permission, violating copyright laws. The case is still in progress. For the latest updates on how companies are utilizing generative AI, from regulatory changes to practical applications, stay tuned to VB Daily. We provide the inside track on maximizing ROI through insights into this rapidly evolving field.

AI Enhanced Version

This article is part of our special issue, “AI at Scale: From Vision to Viability.” Imagine taking a journey 60 years back to Stevenson, Alabama, where the Widows Creek Fossil Plant, a 1.6-gigawatt power station, stood tall with one of the world's highest chimneys. Today, in its place, you'll find a Google data center, powered by renewable energy instead of coal, marking a significant transition from a carbon-burning facility to a digital factory. This transformation signifies a global shift towards digital infrastructure, with AI factories set to supercharge the production of intelligence. AI factories, or data centers, are essentially decision-making engines. They consume vast computing, networking, and storage resources to convert raw data into valuable insights. With the growing demand for artificial intelligence, these densely packed data centers are sprouting up at an unprecedented pace. However, the infrastructure supporting AI faces similar challenges as traditional industrial factories. From power consumption to scalability and reliability, these modern factories need innovative solutions to age-old problems. In the era of steam and steel, labor was synonymous with thousands of workers operating machinery round the clock. In contrast, today's AI factories rely on computing power to produce output. Training large AI models necessitates massive processing resources, and the industry is experiencing a four-fold growth per year in training these models, according to Aparna Ramani, VP of Engineering at Meta. This rapid scaling is leading to familiar bottlenecks seen in the industrial world, particularly in the supply chain. Graphics Processing Units (GPUs), the workhorses of the AI revolution, are complex, in high demand, and subject to price fluctuations due to their limited supply from a few manufacturers. To circumvent these supply constraints, tech giants like AWS, Google, IBM, Intel, and Meta are designing their own custom silicon chips. These chips are specifically optimized for power, performance, and cost, making them uniquely tailored for their respective workloads. The shift towards AI is not just about hardware. It also raises concerns about its impact on the job market. A study by Columbia Business School reveals that AI adoption leads to a 5% decline in the labor share of income, reflecting shifts seen during the Industrial Revolution. Despite these challenges, the demand for AI continues to soar. McKinsey predicts that data centers' power needs will triple by the end of the decade. At the same time, the rate at which processors double their performance efficiency is slowing, resulting in a growing demand for compute power that outpaces the rate of improvement in performance per watt. Meeting this feverish adoption of AI technologies will require significant investment. A Goldman Sachs report suggests that U.S. utilities need to invest around $50 billion in new generation capacity to support data centers. Training AI models can take months, using tens of thousands of GPUs working in parallel. Any failure during this process requires stopping, restoring to a recent checkpoint, and resuming the run. As AI factories become more complex, the likelihood of failures increases, causing significant delays and inefficiencies. To address this, Meta is developing ways to detect failures sooner and resume operations quickly. Future research into asynchronous training may improve fault tolerance while enhancing GPU utilization and distributing training runs across multiple data centers. Like traditional factories, AI factories rely on new technologies and organizational models to scale production. However, instead of producing goods, AI factories generate tokens, the smallest pieces of information an AI model uses. Nvidia CEO Jensen Huang described the AI factory as a new commodity producer, valuable and applicable across industries, marking it as a new Industrial Revolution. McKinsey estimates that generative AI could add $2.6 to $4.4 trillion in annual economic benefits across 63 different use cases. However, achieving even a quarter of this growth by the end of the decade will require an additional 50 to 60 gigawatts of data center capacity. Despite these challenges, the growth of AI factories is set to revolutionize the IT industry, enabling it to generate intelligence for $100 trillion worth of industry. Stay ahead of the curve with our daily updates, providing the latest insights on what companies are doing with generative AI, from regulatory shifts to practical deployments, enabling you to impress your boss and maximize ROI.

AI Enhanced Version

This article is a part of our special series, "AI at Scale: From Vision to Viability". Dive deeper into this intriguing topic here. Three years ago, the realm of AI-powered coding was largely dominated by GitHub Copilot. This AI-based tool astounded developers with its capabilities, such as code completion and generating new code. Fast forward to 2025, the market is now brimming with a plethora of generative AI coding tools and services from various vendors. These advanced tools offer sophisticated code generation, completion features, and support a wide range of programming languages and deployment patterns. This new breed of software development tools holds the potential to radically transform the way applications are built and delivered. However, there are concerns that these tools might render professional coders obsolete. So, what's the reality? How are these tools impacting the current landscape? What are their shortcomings and where is the market heading in 2025? Mario Rodriguez, Chief Product Officer at GitHub, revealed to us that AI tools have become increasingly crucial for developer productivity over the past year. Tools like GitHub Copilot can already generate 30-50% of code in certain workflows, automate repetitive tasks, assist with debugging and learning, and even serve as a thought partner for developers, enabling them to convert ideas into applications in mere minutes. Rodriguez also mentioned that AI tools not only enhance the speed of code writing but also its quality. A recent study found that code written with Copilot is not only easier to read but also more functional - it's 56% more likely to pass unit tests. While GitHub Copilot is a pioneer in this field, newer entrants like Replit, which has developed an AI-agent approach to expedite software development, are also making waves. According to Replit's CEO, Amjad Masad, AI-powered coding tools can increase coding speed by 10-40% for professional engineers. Masad believes the most exciting aspect of generative AI coding is its potential to transform non-engineers into junior engineers, democratizing development and enhancing the efficiency of professional developers. However, he acknowledges that AI isn't a cure-all solution and has current limitations. Itamar Friedman, co-founder and CEO of Qodo, agrees, stating that while AI has made remarkable progress in simple, isolated projects, it still struggles with complex enterprise software. He believes that AI's strength lies in specific tasks like question-answering on complex code, line completion, test generation, and code reviews. Friedman suggests that the biggest challenge lies in the complexity of enterprise software, and pure large language model (LLM) capabilities alone can't handle this complexity. He warns that merely using AI to generate more lines of code could potentially worsen code quality, a significant issue in enterprise settings. According to Friedman, Qodo is addressing this issue by focusing on understanding complex code, indexing it, categorizing it, and understanding organizational best practices to generate meaningful tests and code reviews. However, he identifies legacy code as another barrier to broader adoption and deployment, with a lack of quality data preventing wider adoption of AI coding tools. The future, as Friedman sees it, lies in specialized agentic workflows, with multiple specialized agents each excelling at specific tasks. He expects the features that seemed revolutionary in 2022, like autocomplete and simple code chat functions, to become commonplace by 2025. GitHub's Rodriguez concurs, predicting that AI tools will continue to evolve to assist developers throughout the entire software lifecycle. He envisions a future where developers are augmented with AI, making the process faster and more efficient, rather than being replaced by it. In a future where AI-powered coding becomes increasingly mainstream, Rodriguez believes that platform integration will be a key differentiator for enterprises. He argues that for AI tooling to truly succeed at scale, it must seamlessly integrate into existing workflows. Stay updated with the latest developments in generative AI with VB Daily. We provide insights into what companies are doing with generative AI, from regulatory changes to practical deployments, helping you stay ahead of the curve.

AI Enhanced Version

This article is part of a special issue by VentureBeat, titled "AI at Scale: From Vision to Viability." It explores the transformative role of artificial intelligence (AI) in streamlining business operations and improving customer service. Not long ago, when a customer approached Baca Systems for service, it would take a human agent approximately five to seven minutes to search through similar cases in the system and analyze technical documents. However, the integration of AI agents, powered by Salesforce, has significantly reduced this response time to a mere five to ten seconds. The primary aim of this technological shift, according to Baca Systems, is not to cut down on staff but to ensure the customer's query is addressed as swiftly as possible. This Michigan-based robotics manufacturing company first adopted Salesforce in 2014, and underwent a significant digital transformation in 2021, which included the implementation of Salesforce’s enterprise resource planning (ERP) platform. The company's team members began working with predictive AI for sales and manufacturing forecasts, eventually evolving to AI agents. One of the initial key applications was service calls, with AI agents assisting human representatives in finding relevant case information, thus enabling faster responses. AI is also being used to support preventative maintenance, helping to identify and rectify ongoing issues, such as a short circuit, that may not have been resolved in the past. The goal is to deliver faster resolution times for customers. Another critical application of AI is in sales. As a small company, Baca Systems doesn’t have a large sales team. AI serves as a sales development representative, sending out inquiries and emails, engaging in dialogue, and then passing the prospect to a sales team member. This approach is cost-effective and justifiable, considering the high salaries required for additional salespeople. In the near future, the company plans to deploy customer-facing service agents that can interact with human users via text message to open and handle cases without initial human intervention. If the AI agent can't solve a problem, it will escalate the issue to a human representative. Outside sales and service, Baca Systems is using AI for tasks like generating emails, creating receivables, and crafting stern collections letters when required. The technology is also being used for part deduplication checking, leveraging retrieval-augmented generation (RAG) with prompt builders to detect duplicates and prevent bad data from entering Salesforce. The company has reported little to no pushback from employees regarding the adoption of AI, as it has made their jobs easier. The company plans to continue this deliberate, incremental approach to further incorporate AI, with the aim of remaining agile. In the corporate catering sector, ezCater is leveraging AI to address logistical challenges such as shifts in headcounts, food preferences, and dietary restrictions. They plan to implement Salesforce’s Agentforce, allowing customers to communicate their needs with AI in natural language, and the AI agent will automatically make adjustments. More complex issues will be quickly escalated to a human representative. AI agents will also be used for "restaurant discovery," guiding users to the best venue based on their preferences, budget, location, and other factors. The company is initially incorporating AI agents in-house to assist concierge agents, and reports positive feedback from human agents. The technology has become increasingly accessible in recent months, and is expected to become the norm in the near future. Business partners are also excited about the possibilities for tasks such as business analysis or process maps. For insights into how companies are leveraging generative AI, from regulatory shifts to practical deployments, stay tuned to VB Daily.