deploy
in English: deployment
Deployment is the step where the finished software or AI system moves from the developer's machine to where users actually use it.
In plain terms
Imagine a restaurant inventing a new dish. The kitchen experiments with it for weeks, tasting and refining it. Deploy is the moment the dish appears on the menu and the guest can order it. Until then, only the chefs knew it; from that point on, anyone can encounter it.
The same happens with software: the program first runs on the developer's own machine, then in a test environment, and finally in the "live" system where real people work with real data. In Hungarian, this is called élesítés or üzembe helyezés; in the profession, people often just say "we put it out" or "we deployed it".
In the AI world, the term has taken on a slightly broader meaning. Deploying a language model or an AI agent means turning an experimental version into a service: it is integrated into a customer service system, a banking application, or a corporate process. This is where it is decided whether something that looked good in a demo will hold up in everyday use.
How it works
- The new version is ready: the developer (or increasingly often a coding agent) writes the modification and uploads it to the shared code repository.
- Automatic checks run: tests examine whether old functions have broken and whether the new part does what it is supposed to.
- The program is packaged into a test environment that closely resembles the live system but has no real users. Here, it can still be tried out without consequences.
- If everything is in order, the rollout begins. In many places, this happens gradually: first only a small group of users receives the new version, and only then everyone.
- After go-live, the system is monitored. If an error occurs, the previous version is restored; this is called a rollback.
An everyday example
You log into your bank's web interface in the morning and find the transfer button has moved or a new chat assistant has appeared. This is because the bank deployed the new version to its servers overnight. If the interface behaves strangely after a change and fixes itself by the next day, a faulty release was likely rolled back.
Who is affected
A common misconception
Limits and risks
- A successful test is not a guarantee: in production, the program faces different data, different loads, and unpredictable users.
- Deploying AI systems is particularly risky because the model may answer the same question differently, so it is impossible to test every case in advance.
- If an agent can be deployed independently without human approval, a single faulty modification can affect many users within minutes.
- Restoration is not always complete: if the new version has overwritten or deleted data, reverting to the previous code will not recover them.
Similar, but not the same
- release – Deploy is a technical step: the code is moved to the server. Release means the new feature becomes available to users. This is often a business decision, and the two can be separated in time, for example if a ready feature is enabled later.
- installation – You usually install a program on your own computer or phone; deployment, however, happens on the provider's side, for example on its servers. In mobile applications, the two merge: the developer releases the new version to the app store, and your phone installs it as an update.
- model training – The AI model is created during training; it becomes a usable service during deployment; a trained model without deployment is just a file that no one uses.
Related concepts
software development, DevOps, software lifecycle
continuous deployment, gradual rollout, rollback, model deployment
production environment, test environment, CI/CD, version control, coding agent
A brief history
'Deploy' was originally a military term: it referred to the deployment of troops and arranging them in battle formation. The software industry adopted it from there. Automated, frequent deployment became widespread in the 2010s, along with the DevOps approach and the practice of continuous delivery. The latter was established in the 2010 book 'Continuous Delivery' by Jez Humble and David Farley.
Where things stand
By the end of 2026, deployment is a daily, largely automated step in software development. The new question is how much of it can be delegated to coding agents and what human oversight is needed to put AI systems into production.
In the news
- Claude Code leaked users' email addresses in web search headers; Anthropic deployed a privacy fixNov 6, 2026 · Jog
- OpenAI and JPMorgan Chase to deploy Codex and cybersecurity modelsNov 5, 2026 · Code
- Roche partners with Chai Discovery to deploy Chai-3 model across its pipelinesNov 5, 2026 · Health
- Multimodal delirium risk model validated externally and deployable without site-specific retrainingNov 5, 2026 · Health
- US AI data centers turn to retired jet engines for power generationNov 3, 2026 · Ipar
- ADNOC signs multi-year deal with OpenAI and G42 as Abu Dhabi announces four agreementsNov 2, 2026 · Agent Age
- New York City clarifies employer liability for AI-assisted hiring decisionsNov 2, 2026 · Work & HR
- ADNOC becomes first energy company to deploy OpenAI's Frontier systemNov 2, 2026 · Ipar
Also known as
deploy · deployment · go-live · deployment · installation · deployment · deployment · software deployment
Sources
This entry was machine-edited and has undergone machine expert review; the “In the news” section updates automatically from Agent Age stories.