Qualcomm is strategically deploying Anthropic's Claude language models across its operations, with different model tiers assigned based on task complexity and cost. The semiconductor company's chief information officer, Attila Tinic, has established a framework where advanced tasks use Claude Opus, while routine work like documentation relies on the lower-cost Claude Haiku model.

Why Is AI Cost Management Becoming Critical for Enterprise Leaders?

As companies scale their use of generative AI tools, token spending has emerged as a major financial concern. Tokens are the basic units that language models process; each word or piece of data consumes tokens, and companies pay based on total consumption. Tinic has set what he describes as "generous" token limits for employees, but when someone approaches their cap, it triggers a conversation about which AI model makes sense for their specific task.

This tiered approach reflects a broader challenge across the technology sector. "Everybody who's talking about moving from experimentation to production is probably dealing with costs that have hit them that they weren't maybe projecting," Tinic explained. The shift from testing AI tools to actually using them in daily workflows has revealed that many companies underestimated how much they would spend on AI services.

How Is Qualcomm Restructuring Work Around AI Agents?

Beyond language models, Qualcomm is deploying autonomous AI agents, which are systems that can perform multi-step tasks with minimal human intervention. One early success involved an AI agent that validates purchase orders automatically, even assigning an accuracy score to each document so customer service teams can focus on fixing specific errors rather than reviewing entire documents from scratch.

Another example came from Tinic's IT team, which created an autonomous AI agent to handle nearly the entire laptop refresh process for employees. These implementations signal a shift in how companies think about automation. Unlike previous technology waves where automation handled isolated tasks at the margins, AI agents are being designed to rethink entire workflows from the ground up.

Tinic emphasizes that this requires fundamentally different thinking about how work gets done. "With AI agents and having a digital workforce, they can work differently; they can do a lot of reasoning," he stated. "Your workflow itself really needs to be rethought".

Tinic

What Metrics Are Companies Using to Measure AI Success?

• Volume: The total amount of work completed using AI tools, such as the number of purchase orders validated or tickets resolved automatically.
• Velocity: How quickly products reach market or how fast tasks are completed, measuring whether AI is actually accelerating timelines.
• Quality: The defect rate in software development, the rate of reopened support tickets, and other measures of output accuracy and reliability.

Tinic uses these three metrics to assess whether Qualcomm's AI investments are delivering real business value. For software developers using AI coding assistants, the company tracks not just how much code they produce, but whether products ship faster and with fewer bugs. In the help desk, they monitor both how many tickets AI resolves and how many of those resolutions require human follow-up.

How Are Companies Balancing AI Adoption With Governance?

Qualcomm maintains an AI council made up of legal, security, and IT experts who review any new large language models, AI tools, or datasets before deployment. Tinic has evolved his thinking about governance structures over his career. "In prior years of my career, I would have looked at governance as a slow-moving bureaucracy," he acknowledged. However, he now sees governance as enabling faster adoption by giving employees confidence that authorized tools are secure.

Qualcomm

This approach reflects a maturation in how enterprises view AI risk. Rather than slowing deployment, proper governance actually accelerates it. "It's like the old adage, you can go faster when you know you have brakes," Tinic noted.

Tinic's strategy at Qualcomm illustrates how large enterprises are moving beyond AI experimentation into production deployment. By combining cost controls, workflow redesign, clear success metrics, and governance structures, companies are attempting to capture AI's productivity benefits while managing the financial and security risks that come with rapid adoption. The emphasis on Claude Opus for complex tasks and Claude Haiku for routine work demonstrates how enterprises are becoming more sophisticated about matching AI capabilities to actual business needs.