EP 520: IBM Think 2025 - AI Updates that could shape Enterprise Work
🎯 Summary
Podcast Episode Summary: EP 520: IBM Think 2025 - AI Updates that could shape Enterprise Work
This 47-minute episode of the Everyday AI Show focuses on the significant generative AI announcements made at the IBM Think 2025 conference, specifically highlighting updates to the Watsonx AI platform designed to drive practical, agentic workflows within large enterprises. The host, Jordan Wilson, contrasts consumer-facing AI (like ChatGPT) with the robust, data-integrated solutions necessary for enterprise operations.
1. Focus Area
The primary focus is the IBM Watsonx platform, particularly advancements in Watsonx Orchestrate concerning AI Agent creation and automation. Key themes include the shift from simple LLM interaction to autonomous, workflow-executing agents, the importance of integrating proprietary enterprise data, and the growing viability of Small Language Models (SLMs).
2. Key Technical Insights
- Agentic Workflow Automation: Watsonx Orchestrate is evolving to create “doers”—autonomous AI agents that orchestrate complex workflows across disparate enterprise systems (like Salesforce, Oracle) using low-code/no-code interfaces.
- Rapid Agent Customization: New “build-your-own agent” capabilities allow non-technical users to create customized, data-connected agents in under five minutes, leveraging a drag-and-drop interface and immediate preview/reasoning model visualization.
- SLM Validation: IBM continues to be bullish on Small Language Models (SLMs), noting that modern SLMs (like Llama 4, Gemma 3) are now competitive (Tier 1B) for many business tasks, offering a flexible alternative to massive foundational models.
3. Business/Investment Angle
- Enterprise Data Integration is Key: Watsonx’s core value proposition remains its ability to securely connect to and operate on dynamic, up-to-date enterprise data sources, overcoming the limitations of consumer tools.
- Pre-built Domain Agents Drive ROI: The introduction of pre-built, fine-tuned domain agents (e.g., HR, Sales, Procurement) offers immediate, high-impact automation. The Better Business Bureau reported $1.5 million in annual cost savings from implementing these agents.
- Democratization of Development: The five-minute, no-code agent builder significantly lowers the barrier to entry for AI implementation, allowing business leaders (not just developers) to deploy solutions quickly, accelerating enterprise adoption timelines.
4. Notable Companies/People
- IBM: The central focus, highlighting CEO Arvind Krishna’s emphasis on SLMs and the strategic direction of the Watsonx suite.
- Amazon (AWS): Mentioned for its reported development of Kiro, a highly advanced, multi-modal AI code generation tool expected in 2025, signaling intense competition in the AI coding assistant space.
- Google: Noted for the release of the Gemini 2.5 Pro I/O edition, which now leads the web development leaderboard benchmarks, surpassing OpenAI models in specific coding metrics.
- OpenAI/Windsurf: Windsurf, a coding AI startup, was reportedly acquired by OpenAI for $3 billion, underscoring the high valuation in the AI tooling sector.
5. Future Implications
The conversation suggests the enterprise AI landscape is rapidly shifting from simple querying to agentic execution. The future of enterprise work will involve non-technical staff building and deploying specialized AI agents that autonomously manage complex, multi-step business processes by integrating across existing software stacks. Furthermore, the platform flexibility (supporting Granite, Llama, Mistral, Gemma) indicates a future where enterprises will utilize a diverse portfolio of models optimized for specific tasks rather than relying on a single monolithic LLM.
6. Target Audience
This episode is most valuable for Enterprise Technology Leaders (CTOs, CIOs), Business Unit Heads (HR, Sales VPs), AI Strategists, and IT Professionals working within large organizations that utilize or are evaluating IBM infrastructure. It is targeted at professionals needing to understand the practical, deployable applications of generative AI beyond consumer chatbots.
🏢 Companies Mentioned
💬 Key Insights
"Here's the thing that I really liked about Granite 4.0 Tiny: it can run on a consumer-grade GPU, which is funny because there are always arguments and companies always say, right, like, 'Oh, this can run on a consumer-grade GPU,' but this one actually can."
"All right, let's get to Tiny Granite 4.0. ... it is a small 7-billion parameter model from IBM in their Granite series that's more efficient and performs similar, IBM says, to larger models."
"there's a whole other layer because unlike, unlike agentic AI, when you're talking about a human using a large language model, it's at least a little easier to be like, 'Hey, what happened here? Let's trace back our steps.' With agents, it's a little trickier."
"When we talk about some of the main problems with AI adoption, not even just at the enterprise level, right, just even at the consumer level, it's always trust, transparency, data governance, etc., right? And being able, it sounds like a small thing, but being able to see the reasoning and see how a large language model is or is not accessing your data to see how it's handling the query."
"you can create customized AI agents in under five minutes with no coding required. That part to me is extremely impressive, right?"
"these agents are described as more "doers" that act autonomously and orchestrate workflows across your enterprise, right? A lot of the AI, you know, especially the large language models that we use right now, we don't think of them as necessarily autonomous because we, the human, still for the most part have to go in and feed them information, right? Whereas with Watson X Orchestrate, it's much more agentic, right?"