July 7, 2026, (Inside AI) — As organizations pour billions into artificial intelligence, journalist and author Josh Tyrangiel warns that most are skipping the most critical step: defining the actual problem. In a recent Harvard Business Review IdeaCast, he argued that successful AI adoption hinges not on model selection but on disciplined problem identification and execution.
Tyrangiel, a staff writer at The Atlantic and author of AI for Good: How Real People Are Using Artificial Intelligence to Fix Things That Matter, urged executives to resist the pressure to become “AI-native” overnight. Instead, he advocates for targeted, high-impact applications where AI can deliver measurable value, drawing on examples from healthcare and beyond.
“The number one thing that comes back for the successful implementations is, ‘Hey, did you know what problem you were trying to solve to begin with. And did you have realistic expectations of how you were going to solve it?’” Tyrangiel said.
He emphasized that AI software, while powerful, requires a scalpel, not a sledgehammer. It demands a rare breed of talent—people who can bridge organizational systems and the technology. “The software engineers and makers are not concerned about your problem. You actually need to take a deep breath and kind of stick to fundamentals,” he noted.
Tyrangiel dismissed the notion that AI lab leaders possess superior insight into practical applications. He pointed out that these “labs” are for-profit companies under immense pressure to justify massive investments. “If there’s one thing that the labs have been masterful about, it is projecting themselves as holders of wisdom,” he said. Their urgent calls to adopt AI often serve their own financial needs rather than genuine business transformation.
He illustrated his point with a biotech CEO who met Sam Altman and was told, “You do not want to miss the boat.” The CEO’s reaction: “What boat?” Tyrangiel explained, “It’s their boat. The boat that we took out major loans to buy. Can you get on that boat please because we need to pay them back?”
The Cleveland Clinic Blueprint
Tyrangiel offered a concrete case study from the Cleveland Clinic, where CEO Tomislav Mihaljevic set a clear mandate: technology must serve medicine, not the other way around. The clinic’s margins hover at a razor-thin 2.2%, making operational efficiency a matter of survival.
One project tackled patient flow. A hospitalist, who began her career as a nurse, led an initiative with Palantir to build a scheduling tool that predicts patient discharges. By integrating health data and doctor’s audio notes, the system allowed staff to “play the hospital like a video game,” Tyrangiel said. Results included a 90-minute reduction in emergency room wait times and a significant boost in transfer rates, directly improving the bottom line.
Another project targeted sepsis, a condition that kills 350,000 Americans annually—more than prostate cancer, breast cancer, and opioid addiction combined. The clinic partnered with Bayesian Health to deploy predictive software. Early versions failed because doctors distrusted unexplained flags. After adding explainability, the system cut sepsis mortality by 41% over a year, saving roughly 1,000 lives. Yet, Tyrangiel noted, it still missed about 10% of cases that clinicians caught through intuition—proof that human expertise remains irreplaceable.
From Hype to Practical Management
Tyrangiel stressed that AI conversations must shift from technology to management. He advised leaders to stop treating AI as a monolith. “It’s actually a hairball of lots of different scientific techniques,” he said. Specificity reduces fear and focuses teams on real problems.
He recommended transparent communication, even when job impacts are uncertain. “Our job is to solve problems for our customers in the best and most efficient way possible, and we need you as part of that effort,” he suggested telling employees. He also urged an R&D mindset, with budget controls to avoid runaway costs. “The last year has kind of been like this wild-eyed craziness at casinos,” he observed.
For traditional companies, Tyrangiel advised looking for “code” in their operations—legal rules, mathematical models, or policy manuals—where large language models excel at translation. But he cautioned that dirty data pipelines will derail any AI effort. “You have a massive infrastructure project ahead of you,” he said.
Ultimately, Tyrangiel’s message is one of sober pragmatism. The technology is dazzling, but its rollout demands leadership, domain expertise, and a relentless focus on the problem—not the tool. As he put it, “You can’t hide from it. You just can’t.”
