Maintenance is changing too.

We choose the model and agent environment around your task. Claude with Cowork capabilities, ChatGPT Work or Codex, Microsoft Copilot, Grok Bot and Mistral Vibe represent different approaches. Local models, for example through LM Studio Bionic or connections described by Unsloth, can also be considered.

Your authoritative data, rules and work status stay outside a single proprietary chat. Documented connections and a tested alternative make a later change practical. Switching the whole agent environment can require different connections and permissions. The new combination is checked against your real requirements.

Choose the data boundary →

Test a prototype against outcomes described in advance. Include missing sources, incorrect assignments, outdated records and unauthorised actions, rather than only normal cases.

A bounded shadow run can compare the new approach with the current procedure. Assess quality, human checking time and full costs together.

After a model change, interface update or new rule, test the affected workflow again. Documentation and test cases therefore belong to the handover.

Business rules, technical updates and bugs are part of running software. Agents can investigate affected code, prepare changes, run tests and update documentation. An expert can coordinate substantially more of this work.

Markus Kirchmair expects individuals to coordinate development work that previously required teams of ten, twenty or thirty people. For suitable tasks, he considers gains up to a hundredfold possible. This is a forecast, rather than a general replacement rate or cost promise.

The comparison separates human steering and review time, elapsed time and total cost. Established software vendors also use AI. The right choice depends on your actual requirements and the complete workflow.