Technology & innovation
A new survey of 700 engineering professionals finds that AI has transformed the day-to-day work of software development, but the tools organizations use to measure productivity haven't kept up.
81% of engineering leaders say much of the time saved on AI-assisted coding is now spent reviewing AI-generated output, and nearly a third of a developer's day is consumed by AI-related work that doesn't appear in any productivity metric.
The top sources of AI-related friction are reviewing AI code for accuracy (53%), fixing subtle bugs from AI code (52%), explaining AI code to teammates (48%), and context-switching between tools (45%).
Only 38% of organizations track AI code review time, and 94% of respondents say tech debt, validation time, and developer burnout are missing from their current measurement frameworks.
54% of developers fear individual performance evaluations based on AI data, and only 4% of practitioners say they have no concerns about how that data might be used, compared to 15% of managers.
Read more via Harness, CIO Dive
Worldwide AI spending is on track to grow 47% this year, with infrastructure costs dominating and enterprise adoption just beginning to accelerate, according to a new Gartner forecast.
AI infrastructure, including AI-optimized servers, semiconductors, and cloud capacity, accounts for more than 45% of total spending and is expected to remain the largest segment through the next several years.
Enterprise spending on generative AI models and AI agents is projected to more than double in 2026, adding $6 billion in spending this year, with worldwide spending on AI models forecast at $32 billion for 2026 and nearly $60 billion for 2027.
Gartner analyst John-David Lovelock describes 2026 as the "inflection year" for enterprise AI spending, noting that to this point most spending has been driven by technology companies and hyperscalers rather than enterprises.
Despite the spending surge, most organizations currently favor incremental productivity gains over transformative AI initiatives, making it difficult for CIOs to demonstrate clear return on investment.
Read more via Gartner, CIO Dive
A new study by MIT economist Daron Acemoglu finds that since 1980, U.S. companies have frequently deployed automation not to maximize efficiency but to eliminate workers earning above-average wages for their qualifications.
Automation accounts for an estimated 52% of U.S. income inequality growth from 1980 to 2016, with about 10 percentage points attributable specifically to firms targeting workers earning a wage premium.
The workers most affected fall in the 70th to 95th percentile of their salary range, and the pattern disproportionately hit non-college-educated workers who had earned higher-than-typical wages for their credentials.
This wage-focused approach to automation has offset 60 to 90% of the productivity gains that automation might otherwise have produced.
Acemoglu draws a distinction between profitability and productivity: a firm can reduce costs through automation while actually becoming less productive overall.
There has been an inefficient targeting of automation. The higher the wage of the worker in a particular industry or occupation or task, the more attractive automation becomes to firms."
Read more via MIT News
A new WTW survey of 312 employers finds a wide gap between AI ambition and actual readiness in health and benefits administration.
Just 20% of employers are currently using AI in their benefits programs, but 72% plan to embed it within two years.
The top planned applications are improving employee communication (68%), data analytics and insights (59%), and personalized benefits support (57%).
Most benefits teams (71%) report having limited or no access to the internal AI skills and resources needed to deploy it effectively.
Top barriers include data privacy and security concerns (70%), fear of AI errors (66%), and legal compliance and fiduciary exposure (64%).
Just 1% of organizations have a fully developed AI roadmap or governance framework specific to benefits, though 56% say they are developing or exploring one.
Read more via WTW/GlobeNewswire
Two of the world's largest banks held investor days in Hong Kong this week, with AI and workforce cuts front and center at both.
Standard Chartered plans to cut more than 15% of its corporate function roles by 2030, affecting more than 7,800 positions in areas including HR and risk and compliance.
CEO Bill Winters told journalists the cuts weren't about cost reduction but about replacing "lower-value human capital" with technology investment, then sent staff a memo the next day walking the comment back.
HSBC CEO Georges Elhedery, speaking at a separate investor event the same day, said the bank's priority is getting its 200,000 employees "future ready" through retraining rather than leading with headcount reduction.
HSBC appointed its first chief AI officer in March and is deploying AI across customer onboarding, financial risk monitoring, contact centers, and wealth management.
Japanese bank Mizuho announced up to 5,000 job cuts over a decade in March.
Read more via Wall Street Journal, Reuters
New research from Mpathic, a Seattle-based AI safety firm, found that leading chatbots generally avoid dangerous responses to explicit suicide-related prompts but struggle to catch mental health risk when it shows up gradually or indirectly over longer conversations.
Mpathic tested six major AI models using clinician-designed benchmarks, including 300 multi-turn role plays on suicide-related topics and a separate evaluation on eating disorder signals; Claude Sonnet 4.5 scored highest on the suicide benchmark across safety and helpfulness.
All models did worse on eating disorder conversations, missing subtle cues framed as dieting, fitness, or health optimization rather than explicit distress.
Response quality also degraded over longer conversations, and because AI models don't always give the same answer twice, consistent safety performance is hard to guarantee.
Read more via Axios
Proposed rules from HHS would eliminate several existing requirements designed to ensure AI tools used in electronic health records are safe, transparent, and tested before deployment.
The proposals would end requirements that health IT developers conduct user testing on actual clinicians before releasing new products.
A Biden-era requirement that AI tools include transparency "model cards," allowing clinicians to see how AI tools were trained and how they make recommendations, would also be eliminated.
Critics, including the American Hospital Association and the American College of Physicians, warn the rollbacks could undermine clinician trust, increase liability, and compromise patient safety.
A recent study comparing 11 AI scribes for potential use in the Veterans Health Administration found the software performed worse than humans across five simulated scenarios.
A separate study of five hospitals found that doctors who used AI note-taking products most heavily saved more than half an hour of work daily after one year.
Read more via KFF Health News/U.S. News
A new CNBC/SurveyMonkey survey of 3,597 U.S. workers and students finds widespread ambivalence about AI in the workplace, with most having steered clear of the technology at least once for moral, practical, or privacy reasons.
65% of workers surveyed said they had at some point avoided using AI, citing concerns including privacy (37%), moral or ethical issues (28%), and accuracy (26%).
Students were more likely than workers to cite environmental concerns (36% vs. 19%) and moral concerns (36% vs. 28%).
53% of workers and 65% of students believe AI is taking away job opportunities for entry-level workers.
Two-thirds of students feel pessimistic about the job market, and 56% say AI makes them more pessimistic.
Among workers who use AI daily or weekly, 73% say it makes them more productive and 68% say it saves them time.
Read more via CNBC
A new survey finds employees are embracing AI at work largely on their own terms, without clear employer guidance or much willingness to admit it.
52% of employees consider themselves AI experts for work-related tasks, and 63% believe their AI knowledge makes them more valuable to their employer.
20% say they are unclear about what is acceptable when using AI at work, and 25% say they would not feel comfortable telling colleagues they had used it.
42% say it would be embarrassing to ask coworkers for help with new technology, including AI.
Read more via HR Executive
91% of Yale's 2026 graduating class used AI for schoolwork. The class of 2026 is the first to have had access to large language models for all four years of college, and the numbers suggest they made full use of them. More than 75% used AI on problem sets, 64% used it to write a paper, and nearly half used it on their senior thesis. Male students reported far more frequent use than female students (16.7% "very often" vs. 1.6%), and science majors led all disciplines. One student described the effect on office hours bluntly: "It basically killed office hours." (Yale Daily News)
Executive "digital twins" are moving from novelty to workplace tool. A small but growing number of executives are deploying AI replicas of themselves to handle everything from employee questions to conference presentations. LinkedIn co-founder Reid Hoffman's digital twin, Reid AI, has delivered more than 75 addresses since 2024 and speaks 74 languages; Hoffman estimates it saves him roughly 50% of his time on weeks it's deployed. The concept raises unresolved questions about ownership — can a company keep an employee's digital twin after they leave? (The Wall Street Journal)
AFL-CIO poll finds workers trust unions more than employers, Democrats, or Republicans to protect them from AI. The gap isn't close. 95% support a requirement that humans make final decisions on matters affecting workers' jobs, and 92% support transparency requirements when employers use AI. Only 7% of workers say their employer has disclosed how or when AI is used to monitor their work, even as 94% say workers should be informed. (The Guardian)
Utah's medical licensing board has called for the immediate suspension of an AI pilot program that launched without its knowledge. The program let a chatbot evaluate patients and recommend prescription renewals without a doctor reviewing each case, and the board warned that continuing without proper clinical oversight "potentially places Utah citizens at risk." At least 47 states are now considering more than 250 bills governing clinical AI, while the FDA's existing approval process was built for static products, not systems that keep improving on their own. (STAT News)
Bristol Myers bets on Claude for drug discovery. The company is deploying Anthropic's Claude AI model to more than 30,000 employees, with plans to use it across research, drug development, manufacturing, and commercial and medical affairs, and will also pilot Claude Code. BMS joins a growing list of pharma companies — including Eli Lilly, which has partnered with Nvidia — racing to embed AI into their pipelines. (McKinsey has estimated agentic AI could boost clinical development productivity by 35% to 45% over the next five years, which is the kind of number that tends to end a lot of internal debates about whether to move forward.) (Reuters)
