AI Saves Engineers Hours, But Workloads Only Get Heavier

Software engineers at Amazon and Google once spent hours writing product documents, summarizing meetings, or fixing code. Now those jobs take minutes. But the extra time rarely means a lighter schedule. Instead, it feeds new demands, higher targets, and a faster pace that leaves many feeling stretched.

Priyanka Devi Ramesh, a business intelligence engineer at Amazon, uses the company’s internal AI tools. Pippin helps her draft documents. What used to take over an hour now takes 15 to 20 minutes. She uses Kiro for brainstorming and Amazon Quick to build simple agents. The minutes add up. Ramesh pours every saved slot into the next problem on her list. Business Insider detailed her story and those of several peers.

Prerit Pathak, a security engineer at Google, uses Gemini for meeting notes. Summarizing six months of discussions that would have taken one to two hours now takes five to 10 minutes. The speed feels freeing at first. Yet Pathak and others say the freed time quickly fills with more assignments or tighter deadlines.

These stories show a pattern across tech. Adoption has surged. A Boston Consulting Group study found 74 percent of white-collar workers without managerial duties now use AI regularly, up 23 percentage points from last year. Yet many companies struggle to turn those individual gains into broader business value. Bloomberg reported the findings in early June.

Sarthak Gupta, a data scientist at Amazon, built automation pipelines with AI help. A monthly report that ate up 8 to 10 hours now needs only a 45-minute review. Setting up those pipelines added hours to his week at first. Once running, the tool delivers steady savings. Gupta still ends up busier. New projects arrive faster than old ones disappear.

Similar shifts happen at other companies. Tanvi Pisal, a UX designer who worked at Apple through a contractor, uses AI to turn scattered notes into product requirement documents and brainstorming output. A process that took 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 to two hours spent on initial work.

At smaller shops, the effects feel even bigger. Iren Azra Zou, a software engineer at Double Nickel, uses 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 likes the speed. She also worries that less human oversight on reviews could let subtle flaws slip through.

Controlled studies back up the time compression. Researchers from MIT, Princeton, and others looked at nearly 5,000 developers at Microsoft, Accenture, and a Fortune 100 company who got access to GitHub Copilot. Those using the tool completed 26 percent more tasks per week on average. Less experienced developers posted the biggest 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 gets 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 some firms. Employees chase leaderboards that measure AI consumption, sometimes burning through budgets. 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.