Written rules
The agents read the same rules before they start: how the software is to be built, what they may do on their own, and what they must ask about first. Some actions are blocked entirely.
Agents
"Agent" has become one of the most used words in artificial intelligence. Here we explain what it means as of October 2026, how we use agents to build software, and what that means for you as a customer.
An AI assistant answers what you ask. An agent is given a goal and works towards it in several steps. It is a language model connected to tools: it can read and write files, search for information, run programs and use other systems — within the access it has been given.
The work runs in a loop:
What sets an agent apart from ordinary automation is that it chooses the steps itself as it goes. An integration or a rule-based script does the same thing every time. An agent can handle variation, but it can also choose wrongly.
As of October 2026
Agents do best when the goal is clear and the result can be verified. Software development is the clearest example: code can be built and tested, and mistakes become visible.
Agents are also used to find and compile information, write drafts and process documents. A person often approves the result before it is used.
At Apalos
Apalos is built to develop software together with agents. Every day we use Claude Code from Anthropic and Codex from OpenAI as development partners. That lets a small company build and operate more, without losing control of what is delivered.
The agents read the same rules before they start: how the software is to be built, what they may do on their own, and what they must ask about first. Some actions are blocked entirely.
Each agent works in its own, separate copy of the code. Changes are collected in a request that must be reviewed before it can be accepted.
For larger work we run two agents from different vendors side by side. One reviews the other's work, so we do not depend on a single model's blind spots.
The most important architecture rules are written as automated tests. Every change is built and tested before it can be accepted. An agent cannot argue its way past a failing test.
We decide what is built and approve the changes. We always review sign-in, invoicing, security, databases and operations ourselves.
Every change gets a note on what was done and how it was checked. Important choices are recorded as decisions, so the work can be verified — including by the next agent.
Customer data does not go to the agents. We use them on code, plans and our own documents, not on what you keep in your solution. Read more under Trust and operations.
What we offer our customers today is software: bespoke solutions and our own products such as Apalos Capture, with operations and support. The agents are how we make the software. The solutions we deliver do not contain AI agents.
That does not mean agents are irrelevant to you. Many wonder what is useful and what is mostly talk, where it makes sense to start, and how to stay in control of data and access. We use agents every day and are happy to share what we have learned.
Questions about agents?
Tell us briefly what you are wondering about, or what you are considering using agents for. Then we will talk it through.