Scaling Agents for Gen AI Products - Anju Kambadur

---

---

AI Summary

Video Summary

Presenter Overview

• Proposal for a talk influenced by current trends in the agentic landscape.
• Company background: Bloomberg’s focus on AI and Large Language Models (LLMs).

Key Developments

• Building LLMs
  • Bloomberg invested in AI for over 15 years, developed their own LLM in 2022-2023.
  • Post-ChatGPT, pivoted to utilizing existing open-source technologies for various use cases.

Organizational Structure

• Bloomberg’s AI efforts are spread across:
  • 400 staff, 50 teams in London, New York, Princeton, and Toronto.
  • Collaboration between data teams, product teams, and engineering.

Product Development

• Focus on generative AI products, particularly assisting research analysts.
  • Emphasis on data management: Handling structured and unstructured data.
  • Daily challenges in research analyst roles, including search and data analysis.

Core Principles

• Non-negotiable product attributes: precision, comprehensiveness, speed, and transparency.
• Importance of protecting client data and ensuring factual accuracy in outputs.

Product Implementation

• Example: Generating transcripts from public company quarterly calls to improve research analysis.
  • Building monitoring workflows and accuracy checks to prevent errors in published data.

Agentic Architecture

• Described as semi-agentic, with autonomous components paired with essential guardrails.
• Need for effective error management in performance monitoring.
• Vision for agents to evolve and improve autonomously, while ensuring accuracy and reliability.

Lessons Learned

• Importance of understanding client needs and fast iterative product development.
• Rethinking organizational structure to better support agent creation and integration.
• Balancing autonomy and oversight in agent functionalities.

Conclusion

• Emphasis on evolving AI applications in a rapidly changing environment, ensuring robustness and efficiency.