digital twin (DT)
A digital twin is a computerized replica of a real object—machine, building, process, or even a person—continuously updated with live data, allowing experiments without risking the real thing.
In plain terms
Imagine your car has an exact copy in the computer that has run the same distance, wears down the same way, and senses the same things as the real one. On this copy, you can test what happens if you drive another ten thousand kilometers with the old brake pads, while the real car remains unharmed. This is a digital twin.
What distinguishes it from a plain model or blueprint is that it is alive: it continuously receives fresh data from sensors and other data sources, so it always shows the current state of its real-world counterpart. It is useful for anticipating errors, testing decisions, and learning cheaply from what would be expensive or dangerous in reality.
In the world of AI agents, the concept has gained two new meanings. In one, the agent practices and plans on a digital twin before acting in reality. In the other, the agent itself is the twin: it imitates a person's knowledge, style, and habits, and answers or handles affairs on their behalf in certain situations.
How it works
- First, a virtual model of the real thing is created: its structure, operating rules, and in the case of a person, their typical knowledge and decisions.
- Data continuously arrives from the real pair, for example from sensors, logs, correspondence, or documents, and these update the model.
- Simulations are run on the model: what would happen if the load increased, settings changed, or a component failed.
- The result becomes a forecast or recommendation, which a human or AI agent evaluates to make a decision.
- The decision feeds back into reality, for example by adjusting the machine, and the new measurements are fed back into the twin, restarting the cycle.
An everyday example
At your workplace, the production line maintenance worker receives a notification that a motor bearing is expected to fail in three weeks. This is not guesswork but an estimate from the machine's digital twin, calculated from vibration and heat data. Similarly, you may encounter a corporate agent acting as a leader's 'digital twin,' answering routine questions based on their previous materials.
Who is affected
A common misconception
Limits and risks
- A twin is only as accurate as the incoming data and the underlying model: with a faulty sensor or incomplete data, it will confidently make mistakes.
- Its deployment and maintenance are expensive, so it is not cost-effective for every machine or process.
- A twin of a person is built from sensitive personal data and can be misused if someone makes it speak in your name without your consent.
- An agent imitating a human may get the style right, but it does not take over judgment or responsibility: for what is said in your name, people remain responsible – you or the company operating the agent, depending on the situation and laws.
Similar, but not the same
- simulation – A simulation plays out assumed scenarios on a model, as many times as needed, but has no live connection to reality. A digital twin continuously tracks a specific, existing object with live data, and simulation is just one of its tools.
- deepfake – A deepfake is an AI-generated image, audio or video that looks real and is often used to deceive. The purpose of a digital twin is not deception but tracking and modeling a real counterpart. A twin of a person is also only acceptable with the knowledge and consent of the person concerned; without it, it can be just as much an abuse.
- avatar – An avatar is only appearance, a face or figure, whereas a digital twin is backed by a data-driven model that mimics behavior.
Related concepts
simulation, modeling, ipar 4.0
personal digital twin, industrial digital twin, urban digital twin
AI-agent, dolgok internete, world model, predictive maintenance, synthetic data
A brief history
The basic idea was presented by Michael Grieves in 2002 at the University of Michigan in a lecture on product lifecycle management, under a different name. The term 'digital twin' was coined by John Vickers, a NASA specialist; the expression became known in 2010 in the NASA technology roadmap.
Where things stand
At the end of 2026, it is an established tool in industry, energy and building management, while in healthcare and urban planning it mainly runs in pilot projects. The digital twin that imitates a person and operates as an agent is still at an early stage, with many open legal and ethical questions.
In the news
- Naver and Siemens discuss industrial AI alliance spanning Korea and global marketsNov 2, 2026 · Ipar
- Schneider and Nvidia take digital twin to live 1.2 GW siteOct 26, 2026 · Ipar
- Digital twins could cut unplanned downtime by up to 30%, report saysOct 26, 2026 · Ipar
- Nvidia unveils Cosmos foundation models and Omniverse digital twin upgrades for industrial automationOct 21, 2026 · Ipar
- Siemens: AI could automate 70–80% of manufacturing processes, up from 30–40% todayOct 19, 2026 · Ipar
- Digital twins in AI data center design: Schneider Electric, Nvidia and Dassault deploymentsOct 19, 2026 · Ipar
- Hyundai and Kia showcase 200 manufacturing technologies for smart factoriesOct 19, 2026 · Ipar
- US startup's AI avatar passes as human in job interview test, fooling 48% of participantsOct 12, 2026 · Work & HR
Also known as
digital twin · digital twin · digital likeness · virtual twin · digital twin · digital twins · virtual twin · DT
Sources
This entry was machine-edited and has undergone machine expert review; the “In the news” section updates automatically from Agent Age stories.