log

basicwidespread

A log is a program's automatically kept record: it records, line by line and with timestamps, what happened during operation so it can be reviewed later.

In plain terms

A log is like a ship's logbook or an aircraft's black box. The program continuously records what it is doing while running: it started, someone logged in, it sent an email, it encountered an error. It also writes the exact time next to every line.

It is good for finding out afterwards what happened. If a system goes down at 3 a.m., no one was sitting next to it, but the log shows the last steps. The developer investigates from this, the security specialist sees suspicious logins from this, and the operator notices something slowing down from this.

With AI agents, the log is even more important because the agent works independently in multiple steps. The log reveals what instruction it received, which tools it called, what it read, and what it modified. Without it, you would only see the final result, not the path leading to it.

How it works

The developer writes logging instructions at key points in the program: note here that you started, and here that an error occurred.
While running, the program prints a line at each such point: timestamp, severity level (for example, information, warning, error), and a short message.
The lines go into a file or a central logging system where logs from multiple machines and programs are searchable in one place.
The system compresses or deletes old logs after a certain period, otherwise the storage would fill up.
If something goes wrong, a human or an analysis program retrieves the relevant period and reconstructs the cause of the error from the logs. An alert can also be linked to this, for example if the number of error lines suddenly increases.

An everyday example

You cannot log in to your online banking and call customer service. The representative checks the system log and sees that you tried a wrong password three times at 9:42, which is why your account was locked. Similarly, you can review a coding agent's log to see which files it modified while you were having coffee.

Who is affected

developerThe first step in debugging is almost always reviewing the logs, and this is also how agents' work can be monitored. → Code

entrepreneur, executiveThe log can prove who did what in corporate systems, which can be decisive in disputes, audits, or data protection incidents.

lawyer, citizenLogs often contain personal data (such as IP addresses, login times), so data protection rules also apply to them. → Law

everyoneMost applications and websites log much of what you do within them, even if you see none of it.

A common misconception

MythThe log records everything that happened in the system, so if something is not in it, it did not happen.

In realityOnly what developers or operators have configured to be logged ends up in the log. About anything that was left out, the log is silent. Moreover, the log may be incomplete, and it may have been deleted or modified afterward.

Limits and risks

Sensitive data (passwords, personal data, trade secrets) can easily enter the log, so log files must also be protected.
Too many logs are as useless as too few: the essence gets lost among millions of lines.
The log shows what the program did, not necessarily why. The reason behind an AI agent's decisions is only partially reflected in its log.
Storing logs costs money, and they are deleted after the retention period expires, so you may not be able to retrieve an old event.

Similar, but not the same

logaritmus – In mathematics, 'log' is the abbreviation for logarithm, which has nothing to do with program logs.
backup – Backups preserve the actual data for restoration, while the log only records what happened to them.
trace (tracing) – A trace follows the complete path of a single request or task as a coherent chain. A log, by contrast, is a collection of separate entries.

Related concepts

Broader

observability, operations, software development

Narrower

error log, Audit log, access log, system log

On the edge

debugging, tracing, monitoring, timestamp, metrika

A brief history

The word originates from shipping: a ship's speed was measured with a piece of wood tied to a rope and thrown into the water (called a log in English), and the result was written into the ship's logbook. Computing adopted this for program event logs; syslog, the logging solution still used in Unix systems, was created in the 1980s.

Where things stand

Logging has been standard equipment in all software for decades. At the end of 2026, what is new is that the steps of AI agents are also logged in detail, and AI itself increasingly helps in reviewing the logs.

Status: 2026 November

In the news

Appears in 63 Agent Age stories; the curve shows weekly occurrence in the news magazine.

Bug report alleges Claude Code writes user's email address into User-Agent headerNov 5, 2026 · Code
AWS launches new package to help bring physical AI into manufacturingNov 5, 2026 · Ipar
Penske Logistics selects Pallet AI Solutions to handle freight invoicesNov 4, 2026 · Ipar
JPMorgan, Amazon and Bank of America restrict external AI agents' accessNov 2, 2026 · Money
Audi brings AI-controlled humanoid robots to its Ingolstadt plant to automate a tough logistics taskNov 2, 2026 · Ipar
Dachser: AI agents are transforming logistics, but risks must be weighed tooOct 30, 2026 · Ipar
FDA and CMS scramble to avoid regulatory logjam as more states look to test AI in clinical settingsOct 30, 2026 · Health
Audi plans humanoid robots to automate parts logisticsOct 29, 2026 · Ipar

Also known as

log · log · log file · log file · log file · event log · logging · logging

Sources

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