AI agents can now do meaningful parts of consulting's research, analysis and drafting. The hard part is directing that work so the result is something you can stand behind: evidence traceable, assumptions explicit, judgement and final responsibility staying human. That discipline is what I call the agentic consultant's way of working.
The Consulting AI Operating System is a workflow system for exactly that. It is currently in development and being tested around one workflow: turning mixed client evidence into an accountable recommendation brief — with an evidence register, claim-status labelling (sourced, calculated, estimated or uncertain), defined human review gates, and a final sign-off that stays with the consultant.
The system uses four levels of control: instructional (checklists and status labels), workflow (mandatory checkpoints), technical (source capture), and professional (human approval and stated limitations). It does not guarantee accuracy, confidentiality or compliance — no AI system can — and it is designed on precisely that understanding.
Independent consultants who personally produce recommendation-led client work from mixed evidence, use Claude or ChatGPT at least weekly, and remain personally accountable for what they deliver.
A small founding design-partner cohort is being prepared to test the first workflow on anonymised, synthetic or explicitly approved material. If this is the kind of work you do, you can express interest below. Places are limited and qualification matters more than speed.