# Claude's New Models, Industry Trends, and AI's Role in Customer Service

An episode of [The Anthropic AI Daily Brief](https://invisiblenetwork.ai/podcast/1784290225), 15 May 2025, 8:49.

Audio: https://media.podcastai.com/episodes/dirI4OZsF0C.mp3?u=1754073626

In this episode, we explore the capabilities of Anthropic's new Claude models, providing an overview of their advancements. We discuss industry trends pointing towards more autonomous AI systems and highlight Sendbird's successful integration of Claude, examining its impact on their operations. The episode further delves into the future of AI in customer service and legal tech, contemplating its transformative potential in these sectors. The discussion wraps up with a summary and closing remarks, emphasizing the key insights shared. We conclude with an outro and a reminder to subscribe for more updates on AI developments. (0:00) Introduction to today's topic (0:25) Overview and capabilities of new Claude models (1:56) Industry trends towards autonomous AI systems (2:21) Sendbird's AI success with Claude (3:49) Claude's impact on Sendbird's operations (5:20) AI's future in customer service and legal tech (7:58) Episode summary and closing remarks (8:24) Outro and subscription reminder

## Transcript

[0:00] What if artificial intelligence could fix its own mistakes, just like a seasoned
[0:04] professional?
[0:05] Welcome to the Anthropic AI Daily Brief, your go-to for the latest AI updates.
[0:12] Today is Thursday, May 15, 2025.
[0:15] Here’s what you need to know about Anthropic's groundbreaking new Claude models.
[0:20] Let’s dive in.
[0:25] Anthropic is reportedly gearing up to release the next generation of its Claude models, and
[0:29] they're set to redefine what AI can do on its own.
[0:33] Imagine this: these models not only perform tasks but also correct themselves when they
[0:38] stumble.
[0:39] It's like having an assistant who learns from every misstep without needing you to step in
[0:43] and guide them each time.
[0:46] The company plans to roll out new versions of Claude Opus and Sonnet in the coming weeks.
[0:51] Testers have shared that these models operate far more independently than their predecessors.
[0:56] The secret sauce here is their ability to blend independent reasoning with tool usage.
[1:02] When they hit a snag, they switch gears into a 'thinking' mode, analyze the situation, and
[1:07] figure out how to fix the issue, all on their own.
[1:11] Take, for example, a market analysis task for a Manhattan café.
[1:16] The Claude model begins by evaluating national trends, but it quickly realizes those aren't
[1:21] helpful.
[1:22] So, it pivots to analyzing local demographic data from the East Village, aiming to give more
[1:27] relevant insights.
[1:29] This ability to navigate complex challenges with minimal user intervention is a
[1:34] game-changer.
[1:36] These new models also shine in coding tasks.
[1:39] They don't just churn out code; they test it, find errors, and fix them independently.
[1:45] Early testers say that even with broad prompts like 'make the app faster,' the models embark
[1:50] on their own journey of optimization, trying out different strategies until they get it
[1:55] right.
[1:56] The trend here aligns with a broader industry movement towards creating AI systems that
[2:01] require less human input and can solve problems autonomously.
[2:06] Anthropic’s approach is in sync with others like OpenAI, which is also developing models
[2:11] that integrate reasoning with tool use.
[2:13] This evolution in AI design could lead to systems that are not only more flexible but
[2:18] also more robust.
[2:21] Here's an impressive transformation story from the world of AI customer support.
[2:26] Sendbird, a company known for its robust communication platform, has achieved a
[2:31] remarkable leap in AI support win rates with the help of Anthropic's Claude.
[2:35] We're talking about going from a 30% win rate to an astounding 90%.
[2:41] So, what's the secret behind this dramatic improvement?
[2:44] Sendbird has harnessed Claude's capabilities to power AI customer support agents that handle
[2:49] the high-volume support conversations of over 4,000 global customers.
[2:54] These agents aren't just about scale—they're about precision, reliability, and compliance,
[3:00] which are crucial for maintaining trust and meeting enterprise standards.
[3:05] To give you a sense of scale, Sendbird's platform facilitates more than 7 billion
[3:09] conversations every month across 6 billion users worldwide.
[3:14] It's a massive operation, and the demand for instant, personalized support is ever-growing.
[3:19] Traditional contact centers simply couldn't keep up, and that's where Claude steps in.
[3:25] Claude's role is to provide intelligent automation that doesn't compromise on
[3:29] reliability.
[3:30] By integrating Claude, Sendbird has created AI agents that are not only capable of handling
[3:35] routine inquiries but also adept at escalating more complex issues when necessary.
[3:41] This blend of automation and human-like accuracy allows enterprises to deliver
[3:45] immediate, high-quality support on any scale.
[3:49] Now, why did Sendbird choose Claude over other AI models?
[3:54] Well, after rigorous testing, Claude stood out for its ability to follow instructions
[3:59] precisely and limit those pesky hallucinations that can lead to incorrect responses.
[4:05] Jin Ku, the Chief Technology Officer at Sendbird, highlighted that Claude's kind,
[4:10] patient, and empathetic nature perfectly suits customer support settings.
[4:16] The results were pretty clear-cut.
[4:18] With Claude, Sendbird's win rate in competitive evaluations soared to 90%, even without any
[4:25] prompt optimization.
[4:26] This consistency was maintained even with a reduced knowledge context, which is a testament
[4:31] to Claude's robust performance.
[4:35] The integration process was smooth, backed by excellent documentation and responsive support
[4:40] from Anthropic.
[4:41] Plus, Claude's enterprise-grade security through Amazon Web Services ensured that
[4:46] Sendbird could meet the stringent compliance requirements of its largest clients.
[4:52] But what does all this mean for Sendbird's clients?
[4:55] For companies like Lotte Homeshopping, it means they can redirect 30-40% of human inquiries to
[5:01] AI agents, enabling 24/7 support and boosting customer satisfaction.
[5:07] Lotte's experience with Sendbird's solution highlights the accuracy and flexibility that
[5:12] Claude brings to the table, overcoming previous challenges and setting a new standard for AI
[5:18] support.
[5:20] Sendbird's collaboration with Anthropic is not just about transforming customer service—it's
[5:25] about redefining it as a strategic business driver.
[5:28] By integrating Claude, they're creating proactive AI agents that anticipate customer
[5:34] needs and enhance satisfaction, turning customer service from a cost center into a
[5:39] revenue generator.
[5:41] In essence, Sendbird is setting the stage for a future where AI not only improves
[5:45] communications but also creates lasting customer loyalty.
[5:50] It's a fascinating shift towards a more efficient and human-centric digital connection.
[5:55] Let's dive into a fascinating shift in the legal tech world.
[5:59] Harvey, a prominent legal AI tool, is making waves by embracing new foundation models from
[6:05] Google and Anthropic.
[6:07] This isn't just a casual update—it's a strategic move that showcases Harvey's
[6:11] commitment to versatility and innovation in a rapidly evolving market.
[6:17] Founded in December 2022, Harvey quickly rose to prominence, fueled by substantial backing
[6:23] from the OpenAI Startup Fund.
[6:25] With a valuation hitting $3 billion after a successful Series D funding round, Harvey's
[6:31] decision to integrate models from Google and Anthropic marks a pivotal moment in its
[6:36] journey.
[6:37] It's a clear signal that Harvey is not resting on its laurels.
[6:41] So, why the shift?
[6:43] According to Harvey, their internal benchmarking, aptly named BigLaw, revealed that
[6:49] different AI models excel at distinct tasks.
[6:52] For instance, Google's Gemini 2.5 Pro shines in drafting legal documents, while Anthropic's
[6:59] Claude 3.7 Sonnet excels in pre-trial tasks.
[7:03] By leveraging these diverse strengths, Harvey aims to deliver more tailored and effective
[7:08] solutions to their clients.
[7:11] CEO Winston Weinberg emphasized the importance of this move, stating that while OpenAI remains
[7:17] a key collaborator, expanding their options allows Harvey to better serve clients globally.
[7:23] This approach not only enhances Harvey's product offerings but also reflects a broader
[7:27] industry trend towards collaboration and competition, where companies harness the best
[7:32] models available to meet specific needs.
[7:36] Looking ahead, Harvey plans to launch a public leaderboard that benchmarks AI model
[7:41] performances on legal tasks.
[7:43] This initiative will provide qualitative insights from top legal professionals, offering
[7:48] valuable transparency in a fast-changing landscape.
[7:52] It's a move that could redefine how legal institutions navigate AI technologies.
[7:58] In summary, Harvey's integration of models from Google and Anthropic highlights the importance
[8:03] of flexibility and collaboration in today's tech landscape.
[8:07] As legal tasks become more complex, the ability to leverage multiple high-performing models
[8:12] will be crucial for any AI-driven platform.
[8:16] Harvey's strategic pivot positions it well to stay ahead of the curve, ensuring they continue
[8:21] to offer cutting-edge solutions to their clients.
[8:24] That’s it for today’s Anthropic AI Daily Brief.
[8:28] Harvey’s strategic embrace of Google and Anthropic models underscores a pivotal shift
[8:32] towards flexibility and collaboration in AI.
[8:36] Thanks for tuning in—subscribe to stay updated.
[8:39] This is Bob, signing off.
[8:41] Until next time.

Shows are listed from Apple Podcasts' charts and directory, and Podcast Index where it is used. Every show and episode belongs to its publisher and plays from the publisher's own feed; this site hosts none of them.