Executive summary
A long-running agent takes on obligations faster than any fixed prompt can hold them, and keeping old work valid draws on the same budget as doing new work. The Complete System asks whether an agent can carry that growing load for a month without losing coherence. In the paper’s thirty-day model of five professional fields (legal, M&A, Medicare, grants and government), the system keeps 95% task fidelity while unaided agents fall to 42%. Because staying coherent costs compute, the useful measure is how much productive work an agent sustains once that cost is paid.
Findings
- Fidelity barely moves across the month, from 98% at five days to 95% at thirty, and more compute pushes it towards 99% and 97%. Unaided agents manage 53% at five days.
- Restating every rule at every step, the strongest conventional approach, reaches 77% at five days and 48% at thirty. The Complete System leads it by more than fifteen points at both.
- On a live model, an instruction at risk of being lost held every time with the system in place, against one time in four without it.
- Self-reported compliance proved unreliable: decoy rules fooled the model’s own account of which rules still held.
- Facts with a clear label stay intact deep into a long context. Running totals, by contrast, drift steadily as the work goes on.
Nenya spends that budget so your commitments stay intact for the length of the job and the agent’s effort still goes into new work.