Denis Roux has seen the numbers firsthand. At Deutsche Bank, tasks that dragged on for two years now wrap up in three to six months. Backlogs measured in months disappear in weeks. The chief information officer for the bank’s investment arm delivered the assessment Thursday in Bengaluru, India, where the lender employs 9,000 technology staffers who make up 45 percent of its global tech workforce.
“We’re seeing things that were two years that are now getting done in three to six months,” Roux told Reuters. “We know the productivity is there.” He stopped short of assigning a precise figure to the gains. Yet the message landed clearly. Artificial intelligence has begun to compress timelines across technology projects in one of the world’s largest banks. And the pressure to capture those gains without letting expenses spiral sits at the center of the story.
The shift matches patterns emerging across software teams and consultancies. A study involving 4,800 developers found teams complete tasks 55 percent faster when they deploy tools from GitHub and Accenture. McKinsey has pointed to 30 percent reductions in time to market for software releases. Early-stage builds have seen costs drop as much as 80 percent in some cases, according to analysis from Sketchflow.ai. These figures come from First Line Software.
But speed carries trade-offs. Lower-performing teams using AI have cut lead time to value by nearly 50 percent. Top teams see only 10 to 15 percent improvement. The gap highlights that tools amplify existing strengths more than they lift everyone equally. Data from Plandek’s 2026 engineering benchmarks makes that point plain.
Deutsche Bank manages the upside with discipline. Engineers receive token quotas for models from OpenAI and Anthropic. Extra capacity requires proof of value. The bank then spreads those lessons internally. Roux compared the approach to the controls companies built during their move to cloud services. Usage-based pricing has replaced flat subscriptions. Monitoring becomes essential. “We monitor the usage patterns. We don’t want to slow people down and want them to keep going, but we also want to get a return,” he said in the Reuters report published today at Yahoo Finance.
The bank applies AI selectively. Simpler models handle routine work. More complex analysis, such as linking geopolitical events to portfolio exposure or extracting insights from financial data, receives heavier models. Caution rules. Not every process justifies the compute expense.
That restraint echoes findings from other corners of the industry. A March 2026 CIO magazine examination found developers can boost output by up to 40 percent with AI coding assistants. The gains come mainly in routine code, documentation, tests and internal tools. Yet customers rarely see automatic price cuts. Instead, vendors often pocket the efficiency as margin. “In many cases, AI lowers the cost of producing routine code, documentation, tests, and internal tooling,” said Yoni Michael, CTO and cofounder at Typedef. “That should benefit customers, but the most important customer benefit does not have to be an immediate price cut.” The full analysis appears at CIO.com.
George Manuelian, chief strategist at RapidFort, pushed a different priority. “Customers should expect better software, not just faster development behind the scenes. Fewer defects, faster turnaround on feature requests, and stronger security built into the product.” His warning reflects real risks. AI-generated code has shown 2.74 times more vulnerabilities in some tests. Teams that ship too quickly without proper oversight pay later in stability and rework.
Jayanand Sagar, cofounder and COO at Hyperbola Network, put it bluntly. Vendors run leaner teams thanks to AI. That productivity flows to margin unless buyers demand a share. “If a vendor of a software as a service has shipped three major releases in the last 12 months using AI tooling, that pace of delivery is a data point most CIOs are leaving on the table.”
The economics extend beyond individual projects. Consulting giants have trimmed entry-level hiring sharply. KPMG reduced it by 29 percent. Deloitte cut 18 percent. EY followed with an 11 percent drop. EPAM Systems watched its stock fall 16 to 21 percent after earnings even as it guided for more than $600 million in AI-related revenue. Investors worried that faster delivery would shrink billable hours in the traditional outsourcing model. The patterns come from the same First Line Software overview.
Junior developer demand has fallen about 40 percent at companies that use AI seriously. Global developer ranks continue to grow, from 47.2 million in 2025 toward 58.7 million by 2029. Yet the composition changes. Seniors spend more time on orchestration, architecture and governance. New roles emerge around retrieval-augmented generation, AI guardianship and prompt engineering. Salaries for AI and machine learning engineers average $206,000, a $50,000 jump in a single year.
So who captures the value? Banks and enterprises gain faster project delivery and clearer backlogs. Customers of software vendors can negotiate for quicker roadmaps, better reliability and more frequent updates. Tool providers and hyperscalers collect revenue from surging usage. The developers who master AI orchestration command premiums. Those who treat the tools as simple code generators risk becoming babysitters for brittle output.
Productivity data tells a mixed tale. Some controlled studies show 55 percent faster task completion. Others find experienced developers initially felt 24 percent faster but actually moved 20 percent slower until they adapted their workflows. A 2026 update to that research showed net speed gains of roughly 18 to 20 percent for many. The gap between perception and measured output remains real. Teams that integrate AI into review processes, testing and architecture see the largest lifts. Those that paste prompts and ship see technical debt accumulate.
Deutsche Bank’s approach offers one template. Set quotas. Require justification for extra tokens. Share successes across teams. Apply the right model to the right task. Avoid the temptation to automate everything. The bank continues to build custom tools for data extraction and risk analysis. It pairs them with traditional methods where they perform better.
Larger forces shape the picture. Global IT spending heads toward $6.15 trillion in 2026. AI-related portions reach $2.53 trillion. Agentic systems that act with less human guidance show compound annual growth rates above 100 percent in some forecasts. By 2030, analysts expect 70 percent of routine coding to run through AI assistance or automation. Thirty-five percent of point SaaS tools could give way to agents. These projections sit inside the broader industry analysis from First Line Software.
Yet the human element endures. Architecture decisions, complex integrations, business judgment and accountability still rest with people. AI handles the repetitive. It accelerates the routine. It does not yet replace the strategic conversation that decides what to build or why.
Buyers have leverage if they use it. Ask for faster feature delivery. Demand transparency on AI usage. Tie contracts to outcomes rather than hours. Many have not. Vendors have little incentive to volunteer their efficiency gains. The next round of negotiations will test whether enterprises treat AI productivity as a shared dividend or an invisible margin expander.
Roux hopes to keep pushing efficiency at Deutsche Bank. The productivity sits there, visible in shorter cycles and cleared backlogs. The challenge lies in sustaining the gains while controlling the compute bill and preserving code quality. Other institutions and technology buyers face the same equation. The early data suggests the compression of project timelines has begun in earnest. How far it travels, and who profits most, will unfold over the quarters ahead.
WebProNews is an iEntry Publication