The New Problems AI Is Creating (And How People Are Solving Them)

AI has moved past the question of whether it matters. The new debate is about managing the problems it creates, including content quality degradation, budget overruns, workforce de-skilling, and the organizational complexity of running agentic systems. Companies that hoped to wait out the trend no longer have that option.

EY draws on economic history to temper expectations. The steam engine took nearly a century to produce sustained productivity growth in Britain, electricity took roughly five decades to reshape industrial production, and the computer revolution took close to a decade to generate measurable aggregate gains. EY argues AI is currently in the infrastructure buildout stage, covering data centers, semiconductors, electricity generation, cloud computing, and talent, and that meaningful productivity gains at the aggregate level are likely still ahead rather than already arriving.

Unlike traditional enterprise software, AI carries a marginal cost every time it is used, with every prompt consuming tokens, computing power, and electricity. Several firms have exhausted annual AI budgets within months as employee usage exceeded expectations. Companies have responded with token budgets, usage caps, and tighter governance, while creating pathways for employees to apply for additional budget. EY argues AI should be managed like any other capital allocation, and that adoption pace will depend on whether the value created by each token exceeds its cost. Productivity gains inside organizations have also been jagged, with the work of integrating agentic AI in the short term often consuming the time savings won from earlier improvements, making net organizational gain harder to measure than anticipated.

Varun Anand of Clay described an AI writing policy that started in engineering and expanded company-wide. The policy holds that authors must stand behind every idea and sentence as their own thinking, that writing is itself a thinking process and circumventing it leaves the author with a poorer understanding of the subject, that more time should be spent writing a document than consuming it, and that longer output is not better since AI tends to pad documents with sentences that say nothing. The policy does not prohibit AI use but frames the injunction as being against laziness rather than against the tool. The post received 8,377 likes, applause, or hearts.

OpenAI CFO Sarah Friar described efforts to build an AI-native finance function at OpenAI, with two stated ambitions: a zero-day close giving leaders a real-time, reconciled, and traceable view of financial position, and automated continuously updated forecasting. Recent OpenAI research shows 40% of finance professionals' specialized AI use involves work outside traditional finance, and 22% involves engineering-related tasks, with finance staff now building custom dashboards and tools using ChatGPT and Codex. Friar proposed evaluating each AI workflow against four questions: did AI complete work that mattered, what did it cost including review and rework, was the result good enough to use, and did it help move faster or produce a better decision.

BCG global chair Rich Lesser noted that the question CEOs ask most about AI has shifted from choosing the right vendor to building an organization-level harness capable of using any model or combination of models while preserving proprietary IP, essential data, key business rules, and codified understanding of how processes connect to core strategy and values. Microsoft's Satya Nadella has been advocating a similar position about organizations owning their core knowledge infrastructure.

BCG identified distributed de-skilling as the risk most leaders are not tracking, defined as the collective erosion of judgment, critical thinking, and problem framing across an entire workforce when everyone uses AI. Half the leaders BCG surveyed said they are already seeing de-skilling occurring, and over 60% expected it to be a real threat within three to five years. A paper called The Tragedy of the Cognitive Commons sharpens the concern: checking AI output requires deep expertise, deep expertise comes from doing grunt work for years, and grunt work is the first thing AI eliminates. Each company acts rationally by eliminating junior roles, but the collective result is a profession that cannot catch AI mistakes because it never learned to do the underlying work.

KPMG argues leaders are overspending on technology and underspending on talent. Executives are two times more likely to increase investment in new technology than in employee training, and fewer than 10% said developing stronger workforce training programs was a primary objective. Among leaders who had increased investment in their workforce, 37% reported revenue growth of 20% or more over the past three years, compared to 25% of business leaders overall.