Software engineers at Amazon and Google once spent hours crafting product documents, summarizing meetings or debugging code. Now those tasks shrink to minutes. But the time saved rarely translates into lighter workloads. Instead it feeds new demands, higher output targets and a pace that leaves many feeling stretched.
Priyanka Devi Ramesh, a business intelligence engineer at Amazon, relies on the company’s internal AI tools. Pippin helps her draft documents. What once took more than an hour now requires just 15 to 20 minutes. She turns to Kiro for brainstorming sessions and Amazon Quick to assemble simple agents. The minutes add up. Ramesh reinvests every saved slot into the next problem on her list. Business Insider detailed her approach and those of several peers.
Prerit Pathak, a security engineer at Google, uses Gemini to handle meeting notes. Summarizing six months of discussions that would have demanded one to two hours of manual effort now takes him five to 10 minutes. The speed feels liberating at first. Yet Pathak and others report the freed capacity quickly fills with additional assignments or tighter deadlines.
These stories reflect a pattern repeated across tech. Adoption has surged. A Boston Consulting Group study found 74 percent of white-collar workers without managerial duties now use AI regularly, a 23-percentage-point jump from the previous year. Yet many organizations struggle to turn those individual gains into broader business value. Bloomberg reported on the findings in early June.
Sarthak Gupta, a data scientist at Amazon, built automation pipelines with AI assistance. A monthly report that consumed eight to 10 hours of his time now needs only a 45-minute review. The upfront work to set up those pipelines added hours to his week initially. Once running, the tool delivers consistent savings. Gupta still ends up busier. New projects arrive faster than the old ones disappear.
Similar shifts play out at other firms. Tanvi Pisal, a UX designer who has worked at Apple through a contractor, employs AI to transform scattered notes into product requirement documents and brainstorming output. The process that required three or four hours now wraps in about 30 minutes. Udit Mehrotra, a head of product at Amazon, gets a solid first draft of product documents in minutes instead of one or two hours spent on initial scaffolding.
At smaller shops the effects feel even more pronounced. Iren Azra Zou, a software engineer at Double Nickel, leans on Claude Code for coding tasks. Work that once stretched across a week now finishes in a day. The same tool reviews code and trims hours from what used to be laborious manual checks. Zou appreciates the acceleration. She also worries that reduced human oversight on reviews could introduce subtle flaws.
Controlled studies back up the time compression. Researchers from MIT, Princeton and others examined nearly 5,000 developers at Microsoft, Accenture and a Fortune 100 company who gained access to GitHub Copilot. Those using the tool completed 26 percent more tasks per week on average. Less experienced developers posted the largest gains. Google has reported roughly 10 percent higher engineering velocity tied to its AI tools, with 25 to 30 percent of new code generated by AI though every line still receives human review.
Yet the bigger picture remains mixed. A longitudinal analysis by DX across more than 400 engineering organizations showed that a 65 percent rise in AI tool usage produced only an 8 percent increase in median pull-request throughput. Most teams land between 5 and 15 percent gains. Expectations often run much higher. Some tech leaders push for three to five times productivity improvements in individual OKRs, tying them directly to performance reviews.
The pressure shows in other ways. Microsoft reportedly canceled thousands of Claude Code licenses after engineers racked up token bills reaching $2,000 per person each month. Heavy usage delivered results but at unexpected cost. Tokenmaxxing has emerged as an internal competition at certain firms. Employees chase leaderboards that measure AI consumption, sometimes burning through budgets in the process. The New York Times described the phenomenon in March.
Layoffs add another layer. Meta, Coinbase and Block each cut at least 10 percent of staff in recent rounds, eliminating around 13,000 jobs combined. Executives pointed to AI as a reason they could do more with fewer people. Analysts remain skeptical. Mark Mahaney of Evercore called AI a “nice excuse” for companies that may have overhired during the pandemic or face other business challenges. More than 115,000 tech jobs disappeared across 150 companies this year through early June, according to Layoffs.fyi data cited in the same report. The New York Times examined the trend on June 1.
Productivity data at the macroeconomic level has yet to reflect the gains engineers describe. Federal Reserve researchers and others note that workers report saving time. The economy as a whole has not accelerated in the way many forecasts predicted. A London School of Economics study last year suggested AI could equate to one extra workday of output per week for some professionals. But that extra effort often flows into more tasks rather than shorter hours.
Harvard Business Review examined 200 employees at a U.S. technology company. Those who adopted AI tools saved time on individual assignments yet took fewer breaks and logged more overall hours. The pattern raised burnout concerns. A separate study linked heavy AI use to higher cognitive load and mental fatigue.
Developers themselves express mixed views. Many enjoy escaping tedious boilerplate code, repetitive documentation or long email threads. They prototype faster. They iterate more quickly. Yet context switching multiplies when AI lets them spin up multiple tasks in parallel. Flow states can vanish under the new pressure to keep several threads moving at once.
Tools have evolved. Cursor, an AI-native code editor, earns frequent praise for shaving two to five hours a day off engineering workloads according to recent surveys. Claude Code stands out for large codebases and review work. Amazon Q Developer targets cloud-specific tasks. GitHub Copilot remains the enterprise standard for many. Each delivers measurable speed on narrow activities. None removes the need for judgment, architecture decisions or coordination with other humans.
Companies respond by tightening expectations. Performance metrics now track AI usage in some organizations. Managers factor token consumption or percentage of AI-assisted code into reviews. The message is clear. The technology exists. Workers must demonstrate they apply it aggressively.
But. The gains come with friction. Code quality can suffer if reviews slacken. Security engineers worry about subtle vulnerabilities slipping through AI-generated suggestions. Data scientists spend extra time validating automated pipelines. The saved hours rarely stay saved. They get consumed.
So the paradox persists. Engineers work faster. They produce more. Many feel no less busy. Some feel more so. The tools compress routine work. Management expands the definition of what counts as routine. The result is a tighter cycle of output and expectation that shows little sign of easing.
Tech giants continue to invest heavily. Amazon’s CEO Andy Jassy highlighted generative AI’s role in productivity and cost avoidance in his shareholder letter. Google, Microsoft and others report measurable velocity improvements in internal metrics. Yet translating those internal wins into economy-wide productivity growth remains elusive for now. The gap between individual experience and aggregate data defines the current moment.
Engineers like Ramesh, Pathak and Gupta keep adapting. They master new prompts. They integrate agents into daily flows. They accept that the first draft arrives instantly and the real work begins afterward. The technology delivers on speed. The workplace has not yet figured out what to do with all that extra capacity except ask for more.
WebProNews is an iEntry Publication