Eighty-two percent of U.S. workers say their productivity increases when they use AI tools, according to PYMNTS Intelligence data. Those same developers estimated they had been sped up by 20%.
That gap — between perceived productivity and actual productivity — is one of the most important findings in recent workplace AI research. It suggests that using AI effectively isn’t just about adopting the tools. It requires understanding where they help, where they hurt, and how to structure your workflow around their strengths and weaknesses. In this article, Hassan Taher makes recommendations about how workers in any role can use AI to improve their daily workflows.
Start With What AI Actually Does Well
A widely cited Stanford and MIT study tracked more than 5,000 customer support agents at a Fortune 500 software company over the course of a year. Workers who used AI-generated conversational scripts resolved issues 14% faster on average. Newer and lower-skilled workers saw even larger gains — up to 35% — because the AI essentially distilled best practices from top performers and distributed them to everyone else.
A separate set of randomized controlled trials at Microsoft, Accenture, and a Fortune 100 electronics manufacturer found that developers given access to GitHub Copilot completed 26% more weekly tasks on average. Less experienced developers saw gains of 27% to 39%. Senior developers, by contrast, saw gains of only 8% to 13%.
The pattern across both studies is consistent: AI tools provide the most help with routine, structured tasks and deliver the biggest benefits to less experienced workers. For highly experienced professionals working on complex problems, the gains shrink considerably — and can sometimes turn negative.
Understand the 90% Trap
A Fast Company analysis in October 2025 captured the core difficulty: “AI often gets you 90% of the way there. But that last 10% — checking for errors, refining details, making sure it actually works — can eat up as much time as you saved”.
The piece cited an MIT study finding that 95% of generative AI pilot programs in companies produced little to no measurable impact on profit and loss, despite $30 to $40 billion in enterprise investment. The reason: “most GenAI systems do not retain feedback, adapt to context, or improve over time.” McKinsey research added that while AI helps with repetitive or “shallow work,” the productivity boost shrinks when tasks are complex or require sustained, deep attention.
Hassan Taher, who founded the AI consulting firm Taher AI Solutions in 2019, has advised organizations across healthcare, finance, and manufacturing on integrating AI into their operations. His firm emphasizes a practical, process-oriented approach — identifying appropriate technologies, maximizing benefits while minimizing risks — rather than adopting tools for their own sake. Taher’s consistent message has been that AI should serve as a tool to augment human judgment, not replace it.
Five Practical Applications That Deliver Measurable Results
Automating meeting documentation. Tools like Otter.ai, Fireflies.ai, and Microsoft Copilot can record, transcribe, and summarize meetings in real time. Rather than manually taking notes, participants can stay engaged in the conversation and receive organized summaries with action items afterward. For teams that spend significant portions of their week in meetings, this alone can reclaim several hours.
Summarizing long documents. ChatGPT, Claude, and Perplexity can condense financial reports, research papers, legal filings, and other dense materials into brief overviews. For professionals who need to quickly assess whether a document is relevant before committing time to a full read, this application is among the most consistently useful.
Drafting and editing written communications. Grammarly, Jasper, and similar tools check grammar, adjust tone, and improve clarity. For employees producing high volumes of written content — email, proposals, reports — AI writing assistants can reduce revision cycles and catch errors that human proofreaders miss on first pass.
Data analysis and visualization. Microsoft Copilot and ChatGPT can process spreadsheets, identify patterns, and generate charts without requiring users to write formulas or code. This lowers the barrier for non-technical employees who need insights from data but lack formal training in analytics.
Research and competitive intelligence. AI search tools can synthesize information from multiple sources, track industry trends, and compile competitor profiles more quickly than manual research. Platforms like AlphaSense, Crayon, and Perplexity specialize in turning raw information into structured, usable intelligence.
What Experienced Workers Should Know
The METR study’s finding that experienced developers were slowed by AI tools deserves attention. The researchers noted that developers who already understood their codebases deeply didn’t benefit from AI suggestions because those suggestions often lacked the contextual knowledge the developers already possessed. Reviewing, correcting, and integrating AI-generated code took more time than simply writing it themselves.
MIT Sloan professor Danielle Li offered guidance for organizations deploying AI: treat these tools the way you would treat a talented but inexperienced new hire. “How do we move from an AI that is a genius to one that’s capable within an organization?” she asked at the World Economic Forum in January 2025. She recommended building robust data infrastructure, providing clear examples of what good work looks like, and compensating employees who train and guide the AI systems.
Hassan Taher has written about a related challenge: the tension between AI capability and user trust. “Users are more likely to trust AI in well-defined, low-risk environments where the technology has a proven track record,” Taher has noted. “For example, AI-powered spell checkers and grammar tools are widely trusted because their potential for harm is minimal, and they consistently improve user experience”. His advice for higher-stakes applications: build trust incrementally, verify outputs rigorously, and never deploy AI in a context where the cost of error is high and the system hasn’t been thoroughly tested.
The Bigger Picture
A Harvard Business School study involving 758 consultants at Boston Consulting Group found that those who used GPT-4 within the boundary of its capabilities saw a 38% increase in performance. But those who used it for tasks outside that boundary — where the AI was unreliable — saw performance drop by 19 percentage points. Knowing where AI’s capabilities end is just as important as knowing where they begin.
Manufacturing firms that adopt AI experience a measurable short-term productivity decline before eventually outperforming peers who don’t adopt, following what researchers describe as a “J-curve” trajectory. Older, established firms struggled most during the transition period.
Hassan Taher, who regularly writes on AI applications across industries on Medium, explore how businesses can restructure processes around AI rather than simply layering it onto existing workflows.
The workers who benefit most from AI aren’t necessarily the ones using the most tools. They’re the ones who understand what each tool does well, where it falls short, and how to structure their work to take advantage of both.
