Editor’s note: This post is part of the AI On blog series, which explores the latest techniques and real-world applications of agentic AI, chatbots and copilots. The series also highlights the NVIDIA software and hardware powering advanced AI agents, which form the foundation of AI query engines that gather insights and perform tasks to transform everyday experiences and reshape industries.
As agentic AI adoption continues to grow, with open-source models and tools maturing, companies across industries are increasingly asking: what AI agents should we build to solve our unique business challenges?
Although faster outcomes are a core benefit of using AI, organizations are finding that specialization is the key to business impact and long-term AI adoption. Rather than relying on one-size-fits-all models and services, leading companies are developing specialized AI agents designed to understand and act within the needs of a specific use case.
CrowdStrike, PayPal and Synopsys are examples of companies combining NVIDIA Nemotron open foundation models with their proprietary data and institutional knowledge to create specialized applications. The results are intelligent agents that have the level of expertise required to work alongside human colleagues and boost business operations.
In cybersecurity, speed and precision are essential, especially as cyber threats become more advanced and grow to larger scales.
To meet these rapidly evolving digital threats, CrowdStrike is building specialized AI agents that can work alongside security teams through Charlotte AI AgentWorks. These agents, powered by NVIDIA Nemotron open models and NVIDIA NIM microservices, automate high-volume tasks such as alert triage and remediation, allowing human analysts to focus on higher-order decision-making.
Built on open models and continuously trained by incident responders, CrowdStrike’s Falcon agentic security platform increases accuracy of alert triage from 80% to 98.5%, reducing security analyst teams’ manual effort tenfold. The platform can adapt to new risks and collaborates across the security operations center.
PayPal, a leader in payments and e-commerce, is building agent-driven infrastructure to accelerate intelligent commerce .
The company’s specialized AI agents, developed on Nemotron open models, will enable the first wave of conversational commerce experiences, where agents can shop, buy and pay on a user’s behalf.
With this approach, PayPal built a fine-tuning pipeline in two weeks and reduced latency by nearly 50% while maintaining the high accuracy required to serve its 430 million customers and 30 million merchants.
PayPal’s agents rely on open, modular models that are fine-tuned specifically for payments and commerce, giving the company the control to balance performance, accuracy and cost at a massive scale.
The complexity of modern semiconductor design and manufacturing calls for expertise, precision and speed. Synopsys addresses this with its Agent Engineer , AI agents deployed across the entire chip development workflow, from verification to implementation.
These agents dramatically boost productivity in research and development, identifying critical design bugs that traditional techniques can miss to reduce costly delays.
Running on NVIDIA accelerated infrastructure, Synopsys’ agent-driven workflows deliver up to 15x faster digital design verification performance.
Using open models fine-tuned for each engineering task, as well as software like the NVIDIA NeMo Agent Toolkit and NVIDIA Blueprints, Synopsys quickly moved its chip designs from prototyping to production.
Companies across industries are taking the following steps to transform their proprietary knowledge into specialized AI agents:
Evaluate open models, like NVIDIA Nemotron , that provide a powerful building block to create specialized models for any domain.
Curate, generate and secure domain data using NVIDIA NeMo for agent lifecycle management.
Create specialized agents using customized models that have access to proprietary data.
Continue to fine-tune agents over time with a data flywheel .
Key considerations
- Investor positioning can change fast
- Volatility remains possible near catalysts
- Macro rates and liquidity can dominate flows
Reference reading
- https://blogs.nvidia.com/blog/specialized-ai-agents/#content
- https://www.nvidia.com/en-us/
- https://blogs.nvidia.com/?s=
- ASML allegedly offered to spy on China for the US — company proposed being 'Washington’s eyes and ears in China' after breaking gentlemen’s agreement on limitin
- The upcoming Steam Machine won't be 'subsidized' like consoles to hit a more attractive price target, suggesting high relative pricing — Valve engineer confirms
- Frustrated users crowdfund a $2,000 fix for Lenovo Legion ‘speakers not working properly’ error — bug bounty posted, coder wins the cash by fixing complex audio
- The industry skipped from IPv4 to IPv6, leaving IPv5 and the Internet Stream Protocol to the annals of history — a data streaming experiment rendered unnecessar
- Employee quits job over an Nvidia RTX 5060 — intern refused to hand in GPU won on an all-expense-paid business trip
Informational only. No financial advice. Do your own research.