--- title: "How far have AI Agents come?" newsletter: "MLOps Community" date: 2024-10-31 source: https://aaif.live/newsletters/mlopscommunity/2024-10-31-how-far-have-ai-agents-come --- # How far have AI Agents come? *Plus, New Tool Thursday, integrating real-time data, agent gems, and hidden gems.* *MLOps Community — Agentic AI Foundation, 2024-10-31* https://go.mlops.community/jxnr44 ## June, 2023 - ## MLOps Community Retrospective With our AI Agents conference just around the corner on November 13, let's take a quick look back at some past talks to see how far agents have come. From early-stage hurdles to recent advancements, these insights remind us of what’s been accomplished and hint at what’s still to come. June, 2023 - Travis Fisher on Reliable AI Agents [https://go.mlops.community/04xdky] Last year, Travis Fisher called out the challenges in building reliable AI agents, labeling most as “toys” due to issues like looping and UX gaps. Despite progress, many of these same challenges still shape today’s field. March, 2024 - Michelle Chan on Voice AI Challenges [https://go.mlops.community/8hxy6b] Michelle from Deepgram tackled voice agents in our AI in Production Conference, addressing latency, end-pointing, and clarity to make agents sound more human. The balance of speed and quality in voice AI remains an evolving challenge. May, 2024 - Tom Smoker on Reliable Multi-Agent Systems [https://go.mlops.community/s2x3a2] Tom discussed how knowledge graphs enhance RAG systems' reliability by making them "exactly wrong" as a step toward reliability. His multi-agent setup blends graphs with LLMs to filter relevant information, reducing errors and optimizing accuracy. June, 2024 - Shaun Wei on AI Agents in Customer Service [https://go.mlops.community/xebugh] In June, Shaun Wei shared insights on Rivia, a customer service AI, bridging Fisher’s and Michelle’s themes. His talk underscored agents’ real-world progress and ongoing obstacles, hinting at this year’s key discussions. October, 2024 - Raj Rikhy on The Impact of Agentic Workflows [https://go.mlops.community/0c7r1w] Just two weeks ago, Raj highlighted the importance of clearly defining agent environments, actions, and success criteria. He stressed starting small and testing to ensure agents achieve reliable, real-time performance while managing complexity. ## But that’s the goal of all agents, right? ## New Tool Thursday It's been a while since I wrote about new tools that I wanted to highlight. (Honestly that's probs due to what felt like a sea of sameness). But today is different. Friend of the pod and longtime community member, Sam Partee, just released something that I consider novel. As you probably know I've been knee-deep in AI agents over the past couple of months, organizing the AI Agents in production conference [https://go.mlops.community/jxnr44]. Because of that, I reconnected with Sam to learn about his agent authentication platform. The goal is to bring LLMs away from chat and actually get them to take action. But that’s the goal of all agents, right? I’ve been pretty outspoken about how agents are too unreliable to be useful these days. The value prop of his project is giving AI the ability to use tools more easily and letting the AI use your personal accounts to do things. Up until now, one real challenge (especially at scale) when dealing with agents has been to let them sign into your services and take actions on your behalf (without giving them god-mode permissions). So Sam is dead set on changing that. He told me he wants to make sure agents have the proper agency to do what they need without being a complete liability. The first step for that is creating really good authentication capabilities. They are still in private alpha, but if you sign up for the waitlist [https://go.mlops.community/lswk5q], mention you came from the community, and Sam will grant you instant access. ## Enriching LLMs with Real-Time Context There's a saying that 'the real challenge isn't the technology; it's integrating people and ideas.' I suspect those who've said that haven’t dealt with integrating real-time data into LLM-powered applications. This blog dives into how tools like LangChain and LlamaIndex can solve that. By combining feature stores with live data, we can make LLM outputs more relevant and accurate. The article covers how to optimize prompt engineering with real-world data, ensuring retrieval systems operate efficiently at scale. The focus is on enhancing LLM-driven applications to be more responsive, keeping performance and relevance sharp even as they scale. Maybe a future post will cover the "people and ideas" part. [Enriching LLMs with Real-Time Context](https://go.mlops.community/4wi3s1) ## Job of the Week [https://go.mlops.community/dhcdar](https://go.mlops.community/dhcdar) ## Hidden Gems Agents ## Hidden Gems Interested in partnering with us? Get in touch: partners@mlops.community Thanks for reading. See you in Slack [https://go.mlops.community/slack], YouTube [https://go.mlops.community/mgNff7], and podcast [https://go.mlops.community/ttftlf] land. Oh yeah, and we are also on X [https://go.mlops.community/twitter]. The MLOps Community newsletter is edited by Jessica Rudd [https://go.mlops.community/cetjhj]. --- Source: https://aaif.live/newsletters/mlopscommunity/2024-10-31-how-far-have-ai-agents-come