Why Tools Matter: From 3.3-Million-Year-Old Stones to AI — What the Evidence Says

Start hereIn a 2025 trial, experienced developers expected early-2025 AI tools to cut their time by 24% and afterwards believed they had saved 20%. In fact the tools made them 19% slower. So how do you know whether a tool helps?

Stone tools go back millions of years, and people still argue about whether tools save time. This page starts with the words, then looks at the oldest stone tools and the debate about them. Then it asks the modern question: do tools, from washing machines to AI, actually make work faster?

011. A simple example

Say you buy a new app that promises to save you an hour a week. A month later, do you have that hour? Or did the app add new jobs: settings to manage, messages to answer, things to check?

That gap between what a tool promises and what it does runs through the evidence below.

022. What the words mean

The word technology goes back to Greek. In English it comes from the Greek tekhnologia, “systematic treatment of an art or craft”.

Its root is tekhnē. In the standard dictionary of ancient Greek, its first meaning is “art, skill, cunning of hand”, especially in metalworking. A later sense is a set of rules, a system or method of making or doing. The plainer word tool comes from Old English.

Our reading: a tool is a method as much as an object. That is worth remembering when an app changes and the method has to survive it.

033. The oldest tools, and the argument about them

Lomekwi: 3.3 million years old, if the layer holds

In 2015 a team reported a 3.3-million-year-old site at Lomekwi, in West Turkana, Kenya. They described stone artefacts in place, alongside fossils of early human relatives, in a wooded setting. The toolmakers combined flaking stone with battering.

• They named the industry the Lomekwian and said it predates the Oldowan by 700,000 years.
• In a follow-up, the same team noted that the tools date to before the earliest known fossils attributed to Homo. They argued that toolmaking was not limited to our genus.

Other archaeologists are not convinced. A 2016 paper argued that the report fails to convincingly show the tools were found in place. It said the possibility that they came from mixed or later layers had not been ruled out. A 2020 reply by another team put the question in its title: “What is ‘in situ’?”

Both sides keep the door open. The 2016 critics wrote that they “do not reject the hypothesis” of stone tool use that early.

Still argued Stone tools are 3.3 million years old. The discoverers say yes; critics say the layer is not secure. We can’t settle it, and the evidence does not clearly lean either way.

The Oldowan: about 2.6 million years ago

The Oldowan is a style of systematically flaked, sharp-edged stone tools.

• Before a 2019 report, the oldest known were from Gona, Ethiopia, at 2.58 to 2.55 million years ago. A find at Ledi-Geraru, also in Ethiopia, is bracketed between 2.61 and 2.58 million years ago.
• Sites at Nyayanga, Kenya, are dated to between 3.032 and 2.581 million years ago. They extend the known range by over 1,300 kilometres beyond Ethiopia’s Afar region.
• Wear on the tools and marks on bones show they were used on plants and animals, including hippopotamus butchery.

The Nyayanga team describes the earliest Oldowan as associated with Paranthropus, whose teeth were found at the site. Associated means found together; it does not say who made the tools.

Not only humans

Chimpanzees have an impressive range of tool-aided ways of finding food. Wild capuchin monkeys and long-tailed macaques also crack food with stones. But early human relatives showed skill at making and using sharp flakes that has not been seen in other primates.

Doesn't hold up Only humans use tools. Other primates do. What stands out in our ancestors is making sharp flakes on purpose.

044. Do tools save time?

The long argument about home appliances

Since 1974, researchers have argued about whether “labour-saving” appliances save labour time.

• One researcher, Vanek, argued that housework time had barely changed since 1926.
• Gershuny and Robinson argued that between 1965 and 1985 appliances cut women’s weekly hours of routine housework.
• A 2004 study used Australia’s 1997 time-use survey, which also recorded which appliances each household owned. It found that domestic technology rarely reduced women’s unpaid work time, and sometimes increased it.

The argument did not end there. In the same journal issue, Gershuny replied with a paper titled “Domestic equipment does not increase domestic work”.

Still argued Labour-saving appliances save labour. Earlier work said they cut routine housework; the one study with ownership data says rarely. Its critic’s reply is known to us only by its title, so we can’t say which way the evidence leans. Our reading: a tool can free time, and new jobs can fill it again.

AI tools: three studies, three answers

Three studies point in different directions. Each tested a different kind of person on a different kind of work.

• Writing tasks. In an online experiment, 453 college-educated professionals did writing tasks from their own fields. Half were given ChatGPT. Their average time fell by 40% and the quality of their work rose by 18%.
• Customer support. A study of 5,172 support agents found an AI assistant raised issues resolved per hour by 15% on average. Less experienced workers gained in speed and quality; the most experienced saw small gains in speed and small declines in quality.
• Expert coding. In a randomised trial with early-2025 AI tools, 16 developers with moderate AI experience did 246 tasks in projects they knew well. They expected AI tools to cut their time by 24%, and afterwards believed they had saved 20%. In fact the tools increased their completion time by 19%. The authors say experimental artefacts cannot be entirely ruled out, but that the slowdown held up across their analyses. In February 2026 they published an update, and the study page now marks these results out of date. They believe developers are likely more sped up by AI tools in early 2026 than in early 2025. But they say their newer data are only very weak evidence of how much. The original paper is a preprint.

Partly AI tools make work faster. For some people on some tasks, yes. Our reading: in these studies the gains were largest for less experienced workers on clear tasks. People’s own estimates of the gain could be wrong, and results can date quickly as the tools change.

For how these systems work, see our plain guide to large language models and ten AI myths, checked.

055. How to test a tool yourself

• Time it. Do a real task with and without the tool, and write down the minutes. In the coding study, people’s own sense of the gain was wrong.
• Give it a few weeks. Judge after you have learned it, not on the first day.
• Use it where you are new. The support study found the biggest gains for less experienced workers.
• Count the new jobs. Ask what the tool adds, not only what it removes.
• Write down the method. Keep the steps you use a tool for, so they survive when the app changes. Our page on image and file format myths shows the same habit with file formats.

For a tool that changed everything slowly, see the printing press.

06Quick answers

07Use this

08Quick quiz

How we checked this · Last checked October 2026

• We read the archaeology papers' abstracts, the full 2016 critique, and the title of the 2020 reply, whose body we did not open.
• For the appliance dispute we read the 2004 study's abstract; Gershuny's reply had no abstract we could read, so we give only its title.
• For the AI studies we read the published abstract of each; the customer-support figures are from the 2025 journal version, which differ slightly from the 2023 working paper.
• The coding study's authors published an update in February 2026 and now mark the early-2025 results out of date; we read both pages and give their update.
• The AI section is dated October 2026 and will be rechecked, because new studies appear often.

How we check every article →

09Sources