Guide
Before AI vs after AI: what really changed
Is AI really changing anything, or is it mostly hype? Here are real tasks side by side, before and after AI, each backed by a study or official source, including the places where AI made things slower or riskier. Then a look at jobs that have changed and how AI is helping across industries.
What changed, with the evidence
Answering customer questions
FasterNew support agents took months to learn the answers and the right tone, and leaned on experienced colleagues.
With an AI assistant suggesting replies, agents at one large company resolved about 14% more issues per hour, and the newest agents about 34% more.
The catch: The most experienced agents barely improved: the AI mainly passed their know-how on to newer colleagues.
Office and consulting work
Mixed resultsConsultants researched, wrote and analysed everything themselves.
In a study with Boston Consulting Group, consultants using AI completed 12.2% more tasks, 25.1% faster, and their work was rated higher quality.
The catch: On a tricky task outside AI's strengths, those using AI were 19% less likely to get it right. Knowing when not to trust it matters.
Writing software
Mixed resultsProgrammers wrote every line by hand and looked up how to do things in documentation and forums.
In a controlled test, developers using GitHub Copilot finished a programming task 55.8% faster.
The catch: A later study found experienced developers working on their own large projects were 19% slower with AI, even though they felt faster.
Breast cancer screening
Better resultsIn Sweden's screening programme, two radiologists read every mammogram.
In a trial with about 100,000 women, AI sorted the scans and flagged suspicious areas. There were 12% fewer cancers missed between screening rounds.
The catch: Radiologists still make every decision, and the trial used experienced doctors, so results may differ elsewhere.
Weather forecasts
FasterA 10-day global forecast took hours on a supercomputer with hundreds of machines.
Google DeepMind's GraphCast makes a 10-day forecast in under a minute on one machine, and was more accurate than the leading traditional model on about 90% of measures tested.
The catch: Weather services use AI models alongside traditional ones, not instead of them.
Understanding proteins for new medicines
FasterWorking out the 3D shape of a single protein could take scientists months or years of lab work.
AlphaFold has predicted the shapes of over 200 million proteins, nearly all known to science, free for any researcher to use.
The catch: Predictions are a starting point; important ones are still checked in the lab.
Phone and video-call scams
New riskA scam call came from a stranger's voice, which made it easier to spot.
Scammers can now clone a voice or face. In 2024 an employee of the firm Arup paid out about US$25 million after a video call in which every colleague was a deepfake.
The catch: This is a new risk AI created. See our guide to spotting AI fakes and scams.
Jobs that changed
Employers surveyed by the World Economic Forum expect about 170 million new jobs and 92 million jobs displaced by 2030, a net gain of about 78 million. Technology, AI and data roles grow fastest, along with green energy and care jobs.Source: World Economic Forum, via digital-skills-jobs.europa.eu
Customer support agents
What changed: At Salesforce, AI agents now handle about half of customer conversations, and the support team went from about 9,000 to about 5,000 people.
What happened next: Humans handle the other half, especially the harder conversations.
Source: computing.co.ukCustomer service at a fintech
What changed: In 2024 Klarna said its AI assistant was doing the work of 700 agents and handling three-quarters of customer chats.
What happened next: In 2025 it began hiring people again so that customers can always reach a human: a sign that the best results mix both.
Source: entrepreneur.comTranslators and content contractors
What changed: Duolingo said it would gradually stop using contractors for work AI can handle, such as translation and content creation.
What happened next: Its CEO said AI lets it create far more lessons, for more learners, than manual work ever could. New hiring now focuses on work AI can't do.
Source: computing.co.ukHow AI is helping across industries
- AgricultureBetter weather and crop adviceIn 2025, AI forecasts of the monsoon's start, up to four weeks ahead, were sent to 38 million farmers in India.Source: blog.google
- HealthcareA second pair of eyes on scansAI-supported breast screening in Sweden led to 12% fewer missed cancers between screenings.Source: lunduniversity.lu.se
- Banking and paymentsCatching fraud in millisecondsMastercard said generative AI raised its fraud detection rates by 20% on average in initial testing.Source: mastercard.com
- RetailBetter product information, fasterWalmart used AI to create or improve over 850 million pieces of product data, work it said would otherwise have needed about 100 times its staff.Source: retaildive.com
- TransportSelf-driving taxisWaymo's driverless taxis were giving around 500,000 paid rides a week in US cities in 2026, and have passed 20 million trips.Source: electrek.co
- EnergyUsing less powerDeepMind's AI cut the energy Google used to cool its data centres by up to 40%.Source: deepmind.google
- Science and medicineFaster discoveriesAlphaFold's free database of over 200 million protein shapes is used by researchers working on new medicines.Source: deepmind.google
- Weather and climateQuicker, sharper forecastsGraphCast produces a 10-day global forecast in under a minute.Source: deepmind.google
How we chose these examples
We only included changes measured in a published study or reported by a reliable source, which each card links to. Many claims about AI saving time come from companies selling it, so where results are mixed we say so. For the tools behind these changes, see AI tools by job; for the science stories, mysteries AI helped solve; and for when it goes wrong, AI fails.