Blog
AI agents that work together: what changed this week
· 2 min read
If you only know AI as a chat window where you type questions, you are missing the biggest shift happening right now. AI assistants are becoming agents: systems that carry out tasks on their own and, as of this week, can also confer with each other while they work.
Agents that send each other messages
Claude Code, Anthropic's programming assistant, recently gained the ability to exchange messages between separate sessions. In practice, this means one agent can track down a bug in your system while a second agent is already building a new feature. As soon as the first finds something that affects the second, it reports it straight away. No human needed any more to copy intermediate results back and forth.
For SMEs this may sound like a toy for programmers, but the underlying movement is broader. Repetitive work that currently moves from screen to screen via an employee could soon be handled by a team of digital assistants that coordinate among themselves. We already build our own systems on this principle and are seeing job turnaround times fall structurally.
A concrete example from our own work: for MEA Beveiliging we built a certification platform that processes installation reports which used to arrive manually by email, from more than fifty dealers at once. Digital assistants now also handle incoming questions there: a report is read, classified and turned into a task, and a person only gets involved where a person is actually needed. That is exactly the pattern now emerging everywhere.
Which software will survive the AI era?
A second debate that flared up this week: now that AI systems can build software themselves, which existing software will remain standing? The tentative conclusion from the industry is surprisingly down to earth. Simple workflow tools, think of forms and lists with a thin layer of logic, are under pressure. These days an AI can build one for you in an afternoon, tailored exactly to your needs.
What does remain standing: systems that record your administration, obligations and risks. Accounting packages, record keeping, compliance. There the value lies not in the screens but in the reliability of the data and the accountability around it. The lesson for business owners: be critical of subscriptions for simple tools, and invest instead in the systems that safeguard your business data.
Not every job deserves the most expensive model
The third development: smart companies do not send every AI request to the heaviest (and most expensive) model. New routing techniques assess how complex each task is, sending simple jobs to a lightweight model and only the real thinking work to the top models. Savings of tens of percent on AI costs are no exception.
We apply this principle ourselves too: routine work runs on cost-effective models, and we only deploy the best where quality makes the difference. Anyone using AI seriously in their business would do well to ask this question as standard: what level does this task really need?
What does this mean for you?
Three developments, one common thread: AI is becoming mature tooling. Not one clever chatbot, but a toolbox from which you pick the right piece for each task, and in which the pieces work together better and better. Curious what that could mean in concrete terms for your business? We are happy to think it through with you.