Agentic systems now generate code, reports, analyses and customer responses at volumes once unthinkable. Companies race to deploy them. Yet the humans charged with oversight find themselves buried. One executive put it plainly at a recent gathering of technology leaders: you’ve got so much work that’s been done, so much work to audit, that you can’t truly be accountable.
This tension sits at the heart of today’s artificial intelligence adoption. Fortune reported last week on discussions from the Fortune Brainstorm Tech conference in Aspen. Executives there described a shift from simple productivity tools to autonomous agents that complete entire workflows. The result? Faster output. Greater complexity in verification. And a growing realization that traditional management approaches fall short.
Edwin Olson, founder and CEO of May Mobility, stressed the need to build systems correctly from the start. “How do you build a system that is as right as often as you can possibly make it … how do you create the transparency and introspectability, so you can understand why it made a mistake,” he asked. His firm develops autonomous vehicles. The stakes involve safety and regulation. Similar pressures now appear across knowledge work.
Caitlin Halferty, chief data officer at Thomson Reuters, described efforts to create what she called fiduciary-grade products. These demand transparency, data privacy, subject-matter expertise and reliable content. “Making sure there’s a way in which you can validate the output of any model that you’re using,” she explained. The bar rises when AI touches legal, financial or medical decisions.
But. The volume problem grows faster than solutions. Gregor Stewart, chief AI officer at SentinelOne, pointed to coding as an early warning. AI generates software so rapidly that human review lags. “You’ve got so much work that’s been done, so much work to audit, that you can’t truly be accountable,” he said. Safety-critical techniques from other engineering fields now see renewed interest.
Elena Kvochko, CEO of Trustguard AI, offered one practical approach. Separate the generator from the judge. “You don’t want AI to grade its own work,” she noted. Her firm deploys one agent to draft and another to critique, mimicking newsroom workflows. Multiple systems check each other. The method reduces but does not eliminate risk of shared hallucinations.
These conversations reflect a broader pattern visible in recent data. A Boston Consulting Group analysis forecasts that 50 to 55 percent of U.S. jobs will be reshaped by AI over the next two to three years. Only 10 to 15 percent face outright elimination in the following five years or beyond. The difference hinges on demand. When lower costs unlock more consumption, output expands and humans stay in the loop. BCG calls this demand expandability.
Software engineering provides the clearest case. AI speeds code generation and testing. Engineers shift to system-level judgment and architecture. Organizations respond by building more digital products than before. “As AI reduces the cost and time required to build software, organizations often build more,” the report notes. Total employment in the role can hold steady or grow despite individual productivity jumps.
Contrast that with call centers. AI handles routine inquiries end to end. The number of customer interactions stays tied to the existing base. Fewer representatives suffice. Insurance sales agents see routine tasks automated while higher-value advisory work increases. Financial analysts watch modeling and data aggregation move to machines. The output does not scale proportionally. Headcount shrinks.
Recent surveys reinforce the productivity gains while exposing implementation gaps. BCG’s Global AI at Work report, drawing on nearly 12,000 frontline employees, found 42 percent save the equivalent of one full workday per week through regular AI use. Yet two-thirds receive little or no guidance on what to do with that time. Half fail to redirect it toward strategic priorities. Fortune covered the findings earlier this month.
David Martin, global leader for BCG’s people and organization practice, blamed unclear leadership. “Senior leaders are really struggling to articulate what the vision and strategy is on AI. Consequently, it increases employee fear,” he said. Some workers turn to tokenmaxxing — excessive AI use aimed at internal metrics rather than business value. Companies such as Amazon and Meta have scaled back incentives after observing the behavior.
This workplace-level friction helps explain the wider productivity paradox. Employees work faster. Macro statistics lag. A Fortune article from late May examined the pattern through the lens of the pre-internet era. Workers equipped with better tools deliver the same output in less time. Capital deepening occurs. Yet economy-wide efficiency metrics disappoint until organizations redesign processes. Fortune drew parallels to Robert Solow’s famous 1987 remark that computers appeared everywhere except in productivity statistics.
Newer firm-level evidence points to gains that may soon register more clearly. A survey of corporate executives reported in an Atlanta Fed working paper showed AI-linked labor productivity increases of 1.8 percent in 2025, expected to reach 3.0 percent in 2026. High-skill services and finance lead. Gains come mainly through innovation and demand creation rather than pure cost cutting. Smaller firms anticipate modest headcount growth even as larger ones trim.
PwC’s 2026 Global AI Jobs Barometer, released recently, delivers perhaps the most optimistic corporate picture. Companies with heaviest AI exposure show 40 percent higher productivity growth than laggards. They also post faster wage and headcount increases. Skills in AI-exposed roles change more than twice as quickly. The report credits AI with expanding output and opening new markets rather than simply substituting labor. PwC.
Still, aggregate labor market data remains mixed. The Yale Budget Lab found no major disruption in employment or unemployment trends since ChatGPT’s debut more than three years ago. Routine clerical positions decline. Technical roles in engineering, data analysis and science expand. Jason Furman, former White House economist, noted in recent analysis that revised Bureau of Labor Statistics figures now show productivity running above pre-pandemic forecasts. He joined Erik Brynjolfsson in attributing part of the improvement to AI.
The Stanford HAI AI Index 2026 documented rapid capability growth and adoption. Industry produced over 90 percent of notable frontier models in 2025. Generative AI reached 53 percent population adoption in three years, faster than PCs or the internet in many countries. U.S. consumer value from these tools hit an estimated $172 billion annually by early 2026. Median per-user value tripled in a single year.
Yet verification remains the bottleneck. As agents take on longer task sequences, errors compound. Hallucinations slip through. Rogue behaviors emerge in complex environments. Executives at the Aspen conference described the need for systems that regulate each other. Transparency tools that trace decisions. Human experts who set boundaries and interpret edge cases.
Academic research offers caution. Daron Acemoglu and colleagues have modeled both displacement and reinstatement effects. Automation removes tasks. It also creates new ones. The net employment impact depends on which force dominates. Historical automation often favored higher-skill work while hollowing out middle-tier roles. AI may follow a similar path unless complementary investments in organizational change accelerate.
Consulting firms and banks project meaningful but not explosive growth. Wharton Budget Model researchers estimate AI will lift productivity and GDP by 1.5 percent by 2035, approaching 3 percent by 2055. Goldman Sachs once forecasted up to 1.5 percentage point annual productivity growth over a decade. More recent analyses trim those figures. BCG economists suggest a plausible 0.5 percentage point annual contribution.
The gap between individual gains and organizational results comes down to management. Leaders who treat AI as a simple plug-in tool watch employees save time they then squander. Those who redesign workflows, set clear objectives and invest in upskilling capture more value. The latter group also tends to create the new roles that offset any displacement.
Freelancers already vote with their output. A recent survey highlighted on X showed those using AI earn 40 percent more per hour than non-users on the same platforms for comparable work. Some complete in two hours what once took five, then bill for delivered value rather than time. The pattern suggests markets reward speed and volume when quality holds.
So the question facing executives extends beyond adoption. How do you structure accountability when AI produces more than any team can review? The answers emerging combine technical controls, process redesign and human judgment at critical junctures. Companies that solve this verification challenge first will convert raw speed into sustainable advantage. Those that don’t risk drowning in their own output.
Recent coverage from the World Economic Forum and others highlights the same contradictions. Productivity rises in pockets. Aggregate figures trail. Jobs transform more than they vanish. And the humans remain responsible even as machines generate the bulk of the work. The next phase of AI progress may depend less on model size and more on the systems built around them.
WebProNews is an iEntry Publication