Qualcomm CIO Atilla Tinic is betting that AI inside the company is not just a productivity upgrade, but a diversification engine. In his view, the fast-moving shift to AI agents and “a digital workforce” forces a rethink of end-to-end workflows, not automation bolted onto the margins. And to make that work at scale, Tinic has also put practical guardrails around AI spending and model selection, including token limits he calls “generous.”

That pairing is the story: widespread internal AI use, plus real budget management. Tinic’s cost framework matters because he is watching what happens when organizations try to move “from experimentation to production” and discover costs that “hit them” beyond what they projected. In other words, Qualcomm is not just trying to run more experiments. It is trying to operationalize AI while keeping the economics from quietly strangling the program.

To understand why that matters, zoom out to Qualcomm’s strategic position. The company is trying to diversify away from relying too much on the volatile smartphone market, and it has been leaning hard into adjacent areas like data centers, automotive, and the internet of things. It unveiled new AI accelerator chips last year to better compete with Nvidia and AMD. There have also been reports that Qualcomm is working with OpenAI on a new smartphone AI chip. Meanwhile, the business side has provided some momentum: Qualcomm shares jumped after its investor day, when it revealed new fiscal 2029 revenue targets. The targets included non-handset revenue reaching $40 billion and data center sales totaling $15 billion. For the first two quarters of 2026, Qualcomm’s revenue and earnings results exceeded Wall Street’s expectations, even as the global smartphone market is expected to contract at the steepest rate on record.

Tinics background also explains his approach. Before Qualcomm, he spent three decades in telecommunications, including serving as chief information officer at EchoStar. During Tinic’s time at EchoStar, the satellite internet provider had ambitions to become the U.S.’s fourth major carrier under its Dish brand, but those plans did not pan out as envisioned. Last year, Dish ditched those aspirations, and in June 2026, the subsidiary filed for bankruptcy after deals to sell spectrum licenses to AT&T and SpaceX had not closed in time. That contrast between grand telecom diversification plans that faltered and Qualcomm’s diversification efforts that have been “far more fruitful” is part of the backdrop for why the CIO is emphasizing execution mechanics, not just vision.

One of Tinic’s earliest moves at Qualcomm was organizational: he helped create centralized data and AI teams instead of embedding those resources into various infrastructure and application teams. He has also supported broad deployment of AI tools across the company, including AI coding assistants, Microsoft Copilot, and generative AI features from vendors including ServiceNow, Slack, and Atlassian’s Jira. The point is not subtle. “Things are moving so quickly that I think if anyone thinks that they can completely go it alone, they’re going to be left behind,” Tinic said. This matters for decision-makers because it signals an internal stance toward vendors and platforms: AI capabilities are being stitched into everyday work, not sequestered in a separate AI lab.

You can also see how that shows up in specific workflows. Tinic described the use of AI agents to validate purchase orders autonomously, with the AI system assigning an accuracy score to each document. The goal is to let customer service teams focus on the details that are wrong or unfilled, rather than manually triaging everything. Another example came from Qualcomm’s IT team, which created an autonomous AI agent to handle almost the entire process for refreshing a worker’s laptop. And beyond IT and operations, Tinic said AI tools are being widely adopted to support the software testing life cycle, as well as research and personal productivity.

But the hardest part is not getting AI to do tasks. It is redesigning the workflow itself. Tinic argued that end-to-end work needs to be “rethought” because autonomous agents can do reasoning, meaning they behave differently than past technology waves where automation sat on the margins. To keep the transformation measurable, he says he tracks progress through three impact metrics: volume, velocity, and quality. In software testing, that translates into how much work developers deliver with AI, whether products reach market faster, and whether there are fewer defects. In the help desk, Qualcomm tracks the volume of tickets resolved through AI and also the rate of re-opened tickets. Getting volume, velocity, and quality right, in his framing, is what enables Qualcomm to scale its internal AI capabilities, which ties back to diversification goals.

Governance is the third leg, and Tinic is treating it like an accelerant rather than a brake. One major AI initiative predates his tenure, but he continues to support it: an AI council made up of legal, security, and IT experts to ensure that any new large language models, AI tools, or data sets tied to these projects are secure. Tinic said he used to view governance as slow bureaucracy, but now believes this structure gives employees confidence to use authorized AI tools, like “the old adage, you can go faster when you know you have brakes.” This is a crucial second-order detail for executives: the best AI rollout plan is not only about adoption, it is about authorized adoption.

Finally, the cost control piece ties everything back to reality. As internal adoption rises, Qualcomm is monitoring AI token spend more closely, with Tinic saying he has established token limits that are “generous.” If an employee hits the limit, there is a conversation about different decisions regarding AI model selection. For more advanced cases, Tinic said Anthropic’s Claude Opus model may be required, but for lower-stakes uses like documentation creation, Claude Haiku is sufficient. This is not just procurement trivia. It is a production strategy: the move from experimentation to production forces tradeoffs between capability and cost. And for peers trying to diversify away from a single market, the lesson is blunt. AI can become a competitive advantage only if it survives budget pressure and workflow redesign, not only if it looks impressive in demos.

John Kell CIO Intelligence is taking a three-week summer break and will be back in your inboxes Aug. 5.