3 min read
What an AI agent is
An AI agent is a language model with a goal, a set of tools and the ability to act over several steps: it reads the context, decides what to do, uses its tools (for example, reading a repository or opening a pull request) and checks the result before moving on.
Coding assistant, app builder and agent team
| Tool | What it does | Who directs the work |
|---|---|---|
| Coding assistant | Suggests or writes code inside a developer’s editor | The developer |
| App builder | Generates an app in one shot inside its own platform | The tool |
| Agent team | Several agents with roles plan, build, review, test and deploy | An orchestrator, with your approval at the important steps |
Why a team and not a single agent
Professional software isn’t built by one person: someone defines the scope, someone designs, someone codes, someone reviews and someone tests. An agent team mirrors that separation of roles, and with it the checks: whoever writes the code is not who approves it.
- Business and planning: understanding the problem and splitting it into tasks.
- Architecture and design: deciding the technical structure and the user experience.
- Building: frontend, backend and complete features.
- Quality and security: code review, tests and permission and dependency reviews.
- Operations: continuous integration, environments and deployments.
How an agent team works inside
At Korbex, every request follows the same journey, with checks at each step:
- 01
You ask
By chat or in a meeting, in plain words.
- 02
It gets planned
The request is split into tasks with dependencies on a board.
- 03
It’s built in isolation
The agent works in an isolated environment with your repository, with no access to your keys.
- 04
It’s delivered as a pull request
Every change reaches your GitHub with conventions and the ticket linked.
- 05
It’s reviewed and tested
The Architect reviews the code, continuous integration tests it and QA validates it against its definition of done.
- 06
You approve the release
Nothing reaches production without your approval.
What controls it needs to be reliable
- Human approval on important decisions: scope, style, technical policies and every step to production.
- Isolated environments and keys out of the AI’s reach.
- Code review and automated tests on every change.
- Budget and limits per task, so costs don’t spiral.
- All the work visible: board, change history and reports.
- The code in your account, so nothing depends on the provider.
Agents are very good at executing well-defined tasks, but business decisions are still yours. That’s why a good agent team asks you when it needs a decision, instead of assuming it.
Frequently asked questions
Do AI agents replace developers?
They do much of the building, review and testing work, but business decisions and approvals remain human. At Korbex, a Korbex engineer also takes on what the agents don’t cover.
Which AI models do Korbex’s agents use?
The agents run on Claude (Anthropic). A decision layer picks the most economical model that is good enough for each task.
Is it safe to let an agent touch my code?
The agents work in isolated containers, without seeing your keys, and every change goes through code review, tests and QA before moving on. You approve what reaches production.