Rules decide. People make the rules better.
That is why a few experienced people at dynexo carry a large operation, and why it gets better every month.
Guardrails, not gut feeling
Guardrails decide most of the requests: reviewed rules our systems execute in seconds, traceable and logged. Every action can be traced back to the rule that triggered it. Where a rule falls short, a person decides.
80% machine. 20% human.
The 20% is the part that matters.
How a task travels through our agents
A flow in five steps. A task is recorded together with its acceptance criterion and placed in a queue that knows the states open, blocked and finished. An agent takes the task and works within the guardrails. The result is then checked against the acceptance criterion. If it is met, the result goes back with a log. If it is not met, the task returns to the queue. Cases the rules do not cover are decided by a person.
-
Guardrail decides
In seconds, per a reviewed rule.
-
Human reviews the exception
Where no rule applies, experience decides.
-
Evaluation with AI
What happened, and why?
-
New guardrail
Operations get better every month.
You decide how much runs automatically
Manual: Our people work in your environment, following your processes. No automation touches anything. For areas where you want it that way, and as a starting point while trust is still building.
Automated: Recurring work runs as reviewed code. The result is the same every time. This is how we have worked for two decades.
Agents with guardrails: AI-supported agents take on tasks, but only within the rules. Every action is logged.
The levels are not mutually exclusive. You decide which applies where, and can change it at any time.
Our own teams.
No subcontractors. Whoever works at your site works for us.
Logs stay with you.
Whether a person or an agent is working: data and logs reside in your environment. If anything goes to a model, it goes masked and with zero retention.
No lock-in.
What we build belongs to you, as code and with documentation. We hand it over whenever you want.
Concept first, tool second
Network, firewalls, servers, workstations and services belong together. We first define how your environment should be built, and only then which product implements it. Whoever runs five islands separately finds the gap between them only after someone else has found it first.
Topics we are deep in
- Single source of truth No knowledge, no control. The inventory is the source automation and checks read from.
- SASE Replacing the perimeter, evaluated and migrated rather than swapped out.
- Perimeter and gateways What terminates, what inspects, what balances, and what has to stay available.
- Identity and access IAM and IDM: who may do what, who gets it and who loses it again.
- Segmentation Network, identity, application. What actually separates, and what only carries the name.
- OT and control systems Terminating and inspecting protocols instead of handing a network through.
- Detection and response Detection tested against real attack techniques, response running automatically.
- Automation Ansible, Terraform, rule sets from a repository rather than from a GUI.
- Attack analysis Black-, gray- and whitebox, with our own proof-of-concept instead of a scanner report.
- AI in operations Classify, mask, inspect locally, and only then ask a model.
See it for yourself.
30 minutes with real workflows instead of slides. Afterward you will know which level fits you.