{"id":818,"date":"2025-12-04T14:47:01","date_gmt":"2025-12-04T06:47:01","guid":{"rendered":"https:\/\/www.itsmetoo.com\/?p=818"},"modified":"2025-12-04T14:47:01","modified_gmt":"2025-12-04T06:47:01","slug":"ibm-podcast-explores-2025-as-the-year-of-ai-agents-experts-discuss-challenges-opportunities-and-the-road-ahead","status":"publish","type":"post","link":"https:\/\/www.itsmetoo.com\/index.php\/2025\/12\/04\/ibm-podcast-explores-2025-as-the-year-of-ai-agents-experts-discuss-challenges-opportunities-and-the-road-ahead\/","title":{"rendered":"IBM Podcast Explores 2025 as the “Year of AI Agents”: Experts Discuss Challenges, Opportunities, and the Road Ahead"},"content":{"rendered":"\n

Recently, IBM\u2019s video podcast series\u00a0Mixture of Experts<\/em>released a special episode titled\u00a0\u201c2025 \u2014 The Year of AI Agents?\u201d<\/strong>\u200b Hosted by\u00a0Tim Hwang<\/strong>, the episode featured insights from three IBM experts: engineer\u00a0Chris Hay<\/strong>, Director of IBM\u2019s Open Source AI Innovation Program\u00a0Lauren McHugh Olende<\/strong>, and Vice President of Core AI and Watsonx.ai\u00a0Volkmar Uhlig<\/strong>. Together, they shared their perspectives on the current state and future trajectory of AI agent technology.Notably,\u00a0IBM (IBM.US)<\/strong>\u200b has seen its stock price rise\u00a041.2% year-to-date in 2024<\/strong>, outperforming the\u00a0Nasdaq Composite\u2019s 15.2%<\/strong>\u200b and the\u00a0S&P 500\u2019s 13.2%<\/strong>\u200b gains. The company now has a market cap of approximately\u00a0282.9 bi<\/em>ll<\/em>i<\/em>o<\/em>n<\/em>,w<\/em>i<\/em>t<\/em>h<\/em> Q<\/em>3 2025 re<\/em>v<\/em>e<\/em>n<\/em>u<\/em>e <\/em>g<\/em>ro<\/em>w<\/em>in<\/em>g<\/em> 916.3 billion<\/strong>, driven in part by a\u00a017% surge in its infrastructure segment<\/strong>.Here\u2019s a summary of the key insights shared by the IBM experts during the discussion:<\/p>\n\n\n\n

\ud83d\udd39 1. Consumer-Facing AI Agents: Still a Long Way Off<\/h2>\n\n\n\n

The panel agreed that consumer-grade AI agents are unlikely to see widespread adoption in the near term<\/strong>. Current technology still struggles to reliably handle complex, multi-step tasks in the real world. Moreover, there remains a huge gap between prototyping AI agents and deploying them at scale<\/strong>. A platform solution that dramatically lowers the barrier for non-technical users to create and deploy agents doesn\u2019t yet exist.<\/p>\n\n\n\n

\n\n\n\n

\ud83d\udd39 2. Can \u201cNatural Language to Agent\u201d Fully Bypass Traditional Development?<\/h2>\n\n\n\n

Volkmar Uhlig<\/strong>\u200b envisions that true mass adoption will come when users can describe complex tasks in natural language, and AI will autonomously translate those into executable agents \u2014 a form of direct \u201cnatural language to agent\u201d conversion<\/strong>. This, he argues, could largely bypass today\u2019s complex framework-building processes that require developer involvement.However, Chris Hay<\/strong>\u200b sounded a note of caution from a practical standpoint. Giving large language models (LLMs) too much freedom to invoke tools can easily lead to unpredictable or \u201cderailed\u201d behavior<\/strong>. Therefore, for the foreseeable future, a reliable agent system will still depend on a planning module (Planner)<\/strong>\u200b to define and strictly enforce execution steps. This requires a careful balance between the open-ended creativity of models and the deterministic needs of real-world tasks \u2014 something that goes far beyond simple natural language commands.<\/p>\n\n\n\n

\n\n\n\n

\ud83d\udd39 3. From Proof of Concept to Scale: Key Challenges<\/h2>\n\n\n\n

The conversation highlighted three critical challenges<\/strong>\u200b in moving AI agents from concept validation to large-scale deployment:<\/p>\n\n\n\n

    \n
  1. 1.Reliability & Control:<\/strong>\u200bEnsuring that agents can reliably execute plans in complex environments without going off-track or generating hallucinations will require\u00a0mature frameworks and \u201cguardrail\u201d technologies<\/strong>.<\/li>\n\n\n\n
  2. 2.Cost Efficiency:<\/strong>\u200bVolkmar Uhlig<\/strong>\u200b stressed that for agents to replace human labor or tackle previously unmanageable tasks, their costs must drop\u00a0exponentially<\/strong>. Currently, their use remains limited to\u00a0high-value, highly controlled scenarios<\/strong>.<\/li>\n\n\n\n
  3. 3.Infrastructure & Ecosystem:<\/strong>\u200bThere\u2019s a need for\u00a0\u201cagent cloud platforms\u201d<\/strong>\u200b that simplify deployment, operation, and monitoring, as well as potentially specialized\u00a0optimization models for planning<\/strong>\u200b to reduce reliance on expensive frontier models.<\/li>\n<\/ol>\n\n\n\n
    \n\n\n\n

    \ud83d\udd39 4. What Will the Future AI Agent Industry Look Like?<\/h2>\n\n\n\n

    Lauren McHugh Olende<\/strong>\u200b drew a parallel between today\u2019s agent development and the customized AI model landscape of a decade ago \u2014 where each project essentially started from scratch. She believes future breakthroughs may come from the emergence of reusable \u201cfoundation agents\u201d<\/strong>, or from companies that deeply specialize in a specific use case (similar to how AWS evolved from addressing its own needs) and eventually abstract that into a general-purpose platform.Volkmar Uhlig<\/strong>, meanwhile, argued that dominance in this space will hinge on two core capabilities<\/strong>:<\/p>\n\n\n\n

      \n
    • \u2022Who can deliver the\u00a0best reasoning and planning capabilities at the model level<\/strong>, and<\/li>\n\n\n\n
    • \u2022Who can achieve the\u00a0most extreme cost optimization at the infrastructure level<\/strong>, making AI agents ubiquitous.<\/li>\n<\/ul>\n\n\n\n
      \n\n\n\n

      \ud83d\udd39 5. Interview Highlights (as compiled by Bright Company<\/em>)<\/h2>\n\n\n\n

      \ud83c\udfa4 Q: How close are we to a \u201cone-click\u201d consumer experience with AI agents?<\/h3>\n\n\n\n

      Lauren McHugh Olende:<\/strong>\u200bIf we use the evolution of LLMs as a benchmark, the path becomes clearer. The Transformer paper came out in 2017<\/strong>, BERT and GPT-1 in 2018<\/strong>, and it wasn\u2019t until 2022<\/strong>\u200b that ChatGPT<\/strong>\u200b became widely accessible via web and mobile. That\u2019s roughly a four-year journey<\/strong>\u200b from lab breakthrough to mass adoption.Today\u2019s AI agents are more akin to LLMs in 2018<\/strong>\u200b \u2014 they\u2019ve moved beyond pure research, but we haven\u2019t yet seen a \u201ckiller app\u201d<\/strong>\u200b or a truly simple, consumer-friendly product. We have \u201cBERT-level\u201d demos<\/strong>\u200b that validate concepts but aren\u2019t ready for non-technical users. The big question is: will it take agents another four years to go mainstream? Or might capital, compute, and attention compress that timeline? Alternatively, if agents prove more complex to engineer than LLMs, the timeline could be longer.<\/p>\n\n\n\n

      \n

      \u201cConsumer-facing agents may remain hindered by the bottleneck of natural language interaction.\u201d<\/em><\/p>\n<\/blockquote>\n\n\n\n

      \n\n\n\n

      \ud83d\udee0\ufe0f Q: What\u2019s holding back the developer ecosystem?<\/h3>\n\n\n\n

      Lauren McHugh Olende:<\/strong>\u200bFor experimentation, it\u2019s actually quite exciting. With no-code tools like LangFlow<\/strong>, users can visually assemble agents via drag-and-drop, avoiding the risks of coding first and then discovering missing data or misaligned semantics.For more advanced users, there are options like LangChain, LangGraph, CrewAI, AutoGen, and Semantic Kernel<\/strong>\u200b \u2014 some offer high abstraction and ease of use, others give full control. But the real pain point comes when you try to deploy outside a controlled environment<\/strong>\u200b \u2014 integrating inference services, hosting logic, and connecting everything. There\u2019s no \u201cone-click deploy\u201d solution yet<\/strong>, and developers still have to build the full tech stack themselves.Volkmar Uhlig:<\/strong>\u200bThat\u2019s one of the key barriers. We don\u2019t yet have a \u201cready-to-use\u201d agent solution<\/strong>. The true \u201cShopify moment\u201d hasn\u2019t arrived<\/strong>\u200b \u2014 where anyone can say, \u201cGive me an agent,\u201d and it just works.IBM has been experimenting internally \u2014 turning business process descriptions written in natural language directly into executable LangFlow files. Once we reach the point where anyone can describe a problem in natural language and have an agent auto-generated without coding<\/strong>, that\u2019s when it becomes truly consumer-facing. Imagine saying, \u201cTurn on the lights when I get home,\u201d and the system instantly generates the automation \u2014 no setup required.Right now, the interface is still like a \u201cbaby programmer\u201d tool<\/strong>\u200b \u2014 built for those who can code. But if we can translate natural language business logic into agents as smoothly as we do with code today<\/strong>, the mass-market moment will arrive. The interface just isn\u2019t there yet.<\/p>\n\n\n\n

      \n\n\n\n

      \u2699\ufe0f Q: What\u2019s needed for production-grade AI agents?<\/h3>\n\n\n\n

      Chris Hay:<\/strong>\u200bMoving from POCs or MVPs to scale is hard because consumer behavior is unpredictable<\/strong>. To safely deploy LLMs directly to consumers, you need guardrails<\/strong>\u200b \u2014 either via guard models or deterministic workflows to keep them on track.Approaches like text-to-planning<\/strong>\u200b are emerging, with tools such as Claude Code, Cursor, Windsurf<\/strong>, and Manus<\/strong>\u200b using planners<\/strong>\u200b to break down complex requests before execution. Projects like Manus<\/strong>\u200b decompose tasks with a planning agent, then execute them step-by-step.This \u201cplan first, execute later\u201d<\/strong>\u200b model is essential. Giving an LLM access to hundreds of tools without structure often leads to tool overuse and derailment<\/strong>. Even with a plan, models might skip tools, forget updates, or confidently hallucinate answers based on internal memory \u2014 causing compounding errors.Therefore, deterministic frameworks<\/strong>\u200b are needed to enforce step-by-step execution. But today, these frameworks are still built manually by developers, not natively integrated into platforms or models. For mass adoption, such frameworks must be embedded into the stack.<\/p>\n\n\n\n

      \n\n\n\n

      \ud83d\udcb0 Q: Who will win in the AI agent economy?<\/h3>\n\n\n\n

      Volkmar Uhlig:<\/strong>\u200bTwo key challenges define the competitive landscape:<\/p>\n\n\n\n

        \n
      1. 1.Which models to use?<\/strong>\u200bToday\u2019s frontier models can easily go off-track when given too many tools. Avoiding such failures still relies on their\u00a0dense reasoning capabilities \u2014 which are expensive<\/strong>. We may soon see\u00a0specialized \u201cplanning models\u201d<\/strong>\u200b that focus solely on generating correct plans, but we\u2019re not there yet.So for now,\u00a0frontier models are the only option \u2014 albeit costly<\/strong>. We\u2019ll likely see smaller, cheaper models dedicated to planning.<\/li>\n\n\n\n
      2. 2.How to execute \u2014 and where?<\/strong>\u200bMy belief \u2014 and IBM\u2019s product philosophy \u2014 is that\u00a0\u201cAI should be everywhere\u201d<\/strong>. It\u2019s not just about stacking H100s or H200s in data centers. Agents will run on\u00a0phones, edge devices, cloud, and on-prem<\/strong>.The key is who can make AI agents\u00a0cost-efficient first<\/strong>. Many business processes in places like Portugal still rely on manual labor. We want agents to take over repetitive tasks, freeing humans for higher-value work \u2014 and to provide services in areas that were previously unaddressed. This is fundamentally a\u00a0cost optimization race<\/strong>.We need infrastructure that can drive the\u00a0cost per task down by 10x to 100x<\/strong>. Right now, agents are only viable in\u00a0high-value, labor-intensive, and controlled environments<\/strong>. Once models become more powerful and affordable, agents will become as ubiquitous as electricity and water.<\/li>\n<\/ol>\n\n\n\n
        \n\n\n\n

        \ud83c\udfc6 Q: Will a single model or platform dominate?<\/h3>\n\n\n\n

        Lauren McHugh Olende:<\/strong>\u200bThe winner will be whoever can make processes repeatable first<\/strong>. Today\u2019s agent development is like traditional AI a decade ago \u2014 every project starts from scratch. With agents, it\u2019s even harder because you\u2019re not just rewriting code, but tweaking natural language prompts<\/strong>\u200b to control tool usage and optimize quality.The inflection point for traditional AI came with foundation models<\/strong>\u200b \u2014 large pretrained models that could handle diverse tasks out of the box. Similarly, if we can define a \u201cfoundation agent\u201d<\/strong>\u200b with built-in planning and execution capabilities, and then fine-tune or configure it for different use cases, we eliminate the need to rebuild prompts from scratch each time.<\/p>\n\n\n\n

        \n

        \u201cThat\u2019s the key to moving from a craft-based model to a platform-scale one.\u201d<\/em><\/p>\n<\/blockquote>\n\n\n\n

        Will it be today\u2019s big players?<\/strong>\u200bShe doubts it will be a single model. Instead, success may come from multi-model orchestration<\/strong>\u200b combined with control mechanisms. The winner might not be an existing leader, but a dark horse that perfects a niche use case<\/strong>, then scales it through modular reuse \u2014 much like how AWS began with internal e-commerce needs and evolved into a universal cloud platform<\/strong>.<\/p>\n\n\n\n

        \n

        \u201cThe next platform might emerge from someone who masters one thing exceptionally well \u2014 and then finds a way to make it repeatable.\u201d<\/em><\/p>\n<\/blockquote>\n","protected":false},"excerpt":{"rendered":"

        Recently, IBM\u2019s video podcast …<\/p>\n","protected":false},"author":2,"featured_media":819,"comment_status":"open","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[9,8,3],"tags":[24,150,222],"class_list":["post-818","post","type-post","status-publish","format-standard","has-post-thumbnail","hentry","category-companies","category-deep-tech","category-investment","tag-ai","tag-ibm","tag-q3"],"_links":{"self":[{"href":"https:\/\/www.itsmetoo.com\/index.php\/wp-json\/wp\/v2\/posts\/818","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/www.itsmetoo.com\/index.php\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/www.itsmetoo.com\/index.php\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/www.itsmetoo.com\/index.php\/wp-json\/wp\/v2\/users\/2"}],"replies":[{"embeddable":true,"href":"https:\/\/www.itsmetoo.com\/index.php\/wp-json\/wp\/v2\/comments?post=818"}],"version-history":[{"count":1,"href":"https:\/\/www.itsmetoo.com\/index.php\/wp-json\/wp\/v2\/posts\/818\/revisions"}],"predecessor-version":[{"id":820,"href":"https:\/\/www.itsmetoo.com\/index.php\/wp-json\/wp\/v2\/posts\/818\/revisions\/820"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/www.itsmetoo.com\/index.php\/wp-json\/wp\/v2\/media\/819"}],"wp:attachment":[{"href":"https:\/\/www.itsmetoo.com\/index.php\/wp-json\/wp\/v2\/media?parent=818"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/www.itsmetoo.com\/index.php\/wp-json\/wp\/v2\/categories?post=818"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/www.itsmetoo.com\/index.php\/wp-json\/wp\/v2\/tags?post=818"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}