- Product
- Team of Rivals
- Delivers
- Verified answers
- Proof
- arXiv 2601.14351
If you want coherence, orchestrate a team of rivals.
One model grading its own work is how wrong numbers reach the board. aidnn splits the job across agents with opposing incentives, and one of them can say no.
- 01PlanA planner turns the question into steps, using your definitions.
- 02RunCode runs in a sealed sandbox next to your data. Only a summary comes back.
- 03CritiqueA separate critic can veto the plan and send it back with the reason.
- 04VerifyA verifier checks the result against the source before anyone sees it.
- 05AnswerYou get the answer and the trail of how every number was reached.
Source: “If You Want Coherence, Orchestrate a Team of Rivals,” arXiv 2601.14351.
What it means for you
Fewer surprises. Every step on record.
Errors stop at the critic.
A separate agent can veto any plan, query or result on logic, math or code.
Every number has receipts.
Plans, queries and checks are logged, so any answer can be traced and reproduced.
Data stays where it is.
Code runs in a sealed sandbox beside your data. Agents only ever see summaries.
The roles
Strict boundaries. Opposing incentives.
Like a well-run org chart, each role owns one part of the work and answers to the others.
Planners
Turn the question into a plan, using NeoCortex for your definitions.
Executors
Write code that runs in a remote sandbox, never in the agent's head.
Critics
Review plans and results, with veto power.
Experts
Bring domain knowledge such as revenue recognition, with citations.
Security
Your data stays yours.
aidnn sits on top of your systems. It doesn't take the data away from them.
Never trains on your data.
You delete it, we forget it.
Isolated, encrypted, and able to run where your data lives.
Next step
See aidnn on your messiest data.
A live walkthrough on your workflows: reconciliation, NRR, cohorts, forecasting. Connect in minutes, not months.