How it works

From shadow mode to 95% autonomy

Going autonomous isn't a switch — it's an evidence-based process. Each phase earns the trust of the next. This is the operating playbook.

Phase 0 · Week 1–2

Connect

Plug in your systems — code, finance, CRM, analytics. Cortex compiles a live company state within days. No workflow changes yet; the organization keeps working exactly as before.

Outcomes
Company state live and synced
Exception detection running
Baseline decision log started
autonomy ceiling: 25%
Outcomes
Accuracy scored per decision class
Rejection reasons captured
Trust map: where AI already beats gut feeling
autonomy ceiling: 50%
Phase 1 · Month 1–3

Shadow mode

Agents recommend on every decision and execute nothing. Every recommendation is scored against what your team actually did. Every human rejection requires a one-line reason — that dataset is gold.

Phase 2 · Month 3–6

Graduated delegation

Decision classes with proven accuracy graduate to autonomous execution — reversible, low-cost ones first. Everything else becomes one-click approvals. The policy engine enforces the boundaries.

Outcomes
L3 classes run autonomously
L2 with 24h veto window
Approval latency drops by an order of magnitude
autonomy ceiling: 75%
Outcomes
95% of decision volume autonomous
Humans manage exceptions and direction
Ledger = institutional memory
autonomy ceiling: 100%
Phase 3 · Month 6+

Scale to 95%

The 95% doesn't come from moving hiring or contracts to AI. It comes from shifting the decision mix: most decision volume in a company is reversible and cheap — exactly what agents handle best. Humans keep strategy, people and the objective function.

The critical rule

Autonomy expands only with evidence

The key metric isn't “how many decisions does the AI make.” It's “how often was the AI right when a human overrode it.” When the ledger proves a decision class is handled better by the system, that class graduates. Not before.

Start phase 0 today.