serverless architecture

in English: serverless computing

advancedwidespread

A cloud solution where the program runs and costs money only when it has work to do, while the provider fully manages the machines.

In plain terms

The name is misleading: servers are still present, but you do not need to manage them. It is like a taxi compared to owning a car: you do not buy a vehicle, do not service it, and do not pay for parking. You only pay when you actually travel, and only for the distance covered.

In the traditional solution, a company rents or buys a server that runs day and night, even if no one uses it. In the serverless model, the developer hands over only the program code to the cloud provider. The provider starts it when a request arrives and shuts it down afterward.

If many users arrive suddenly, the system automatically launches more instances. If no one comes, none run, and the bill is close to zero. This is why startups and those with fluctuating traffic favor it.

This is common in AI too: many individual steps of an AI agent (for example, processing an email or querying data) run as short-lived programs that start on demand.

How it works

The developer writes a small program designed for a single task (a function) and uploads it to the cloud provider.
It specifies what triggers the start: for example, a website click, an uploaded file, an incoming message, or a schedule.
When the event occurs, the provider prepares a runtime environment and executes the code within it. This usually takes a fraction of a second, but if the function has not run recently, it can take several seconds (cold start).
If many requests come at once, the system automatically launches many parallel instances, then shuts them down after the peak.
Billing is based on the number of calls, runtime, and memory allocated to the function, often with millisecond precision.

An everyday example

You upload a photo to an online store or classifieds site, and within seconds a thumbnail preview appears. Behind the scenes, your upload most likely triggered a serverless function, which resized the image and then shut down immediately.

Who is affected

developerYou do not need to install and maintain a server, but you must break your code into small, event-triggered units. → Code

entrepreneur, executiveWith low traffic, a service can be launched very cheaply using it, as there is no need to pay for machine capacity in advance.

everyoneA significant portion of everyday applications, chatbots, and AI agents run partly on such backend systems; this is simply not visible from the outside.

A common misconception

MythServerless means there is no server; the program runs somewhere 'in the air'.

In realityIt runs on data center machines just the same, but these are operated by the provider, and the developer sees nothing of them. 'Serverless' indicates that you do not have to deal with it.

Limits and risks

Cold start: if a function has not run for a while, serving the first request may be noticeably slower because the environment must be set up first.
With large and constant traffic, it can be more expensive than a continuously running own or rented server.
There is strong provider lock-in: a solution written for one cloud is difficult and costly to move to another.
Function runtime and memory are limited, so this is not the right tool for long, complex calculations (such as training large AI models).

Similar, but not the same

cloud computing – Cloud is the broader concept in which you can rent a continuously running machine; serverless is the form where you do not rent a machine at all and pay only for actual usage.
function as a service (FaaS) – FaaS is the best-known form of the serverless model, but it is not identical to it: in a serverless manner, one can use not only functions but also, for example, databases, storage, or message queues.
container – A container is a packaged program, meaning a packaging method, not an operational model. You can run it on your own machine, where you are responsible for scaling, but nowadays also in a serverless service, where the provider handles this.
virtual machine – A virtual machine is a rented full computer for which you pay even when it is powered on but not working; serverless code starts only on request.

Related concepts

Broader

cloud computing, IT infrastructure

Narrower

function as a service, cold start, edge network function

On the edge

container, microservice, API, automatic scaling, event-driven architecture

A brief history

Amazon popularized the approach widely when it introduced the AWS Lambda service in 2014. Following this, other major cloud providers launched their own versions.

Where things stand

At the end of 2026, it is part of the basic offering of every major cloud provider and an established solution for web back-end systems and data processing. AI agents' steps and tool calls increasingly run on it as well.

Status: 2026 November

Also known as

serverless · serverless computing · serverless computing · serverless architecture · serverless architecture · FaaS · Function as a Service · function as a service

Sources

This entry was machine-edited and has undergone machine expert review; the “In the news” section updates automatically from Agent Age stories.