Decision engine
AgentKit reasons before it recommends.
One decision in. One evidence-backed action out.
The decision engine begins after a commercial objective exists. It searches for reality, collects evidence, ranks opportunities, compares strategies, and produces the shortest executable path with the highest confidence.
Pipeline
Objective → Search → Rank → Reason → Recommend → Execute
Objective
Start with the commercial decision the founder actually needs to make.
Search
Search the reachable commercial action space instead of inventing opportunities.
Rank
Compare value, risk, confidence, cost, time, and policy compliance.
Recommend
Choose one highest-confidence executable recommendation.
What the engine must do
- Gather only the minimum information needed to improve the decision.
- Keep assumptions separate from facts.
- Surface unknowns instead of hiding them.
- Route approved work into a concrete execution plan.
What it produces
- Decision Brief
- Execution Plan
- Execution Tasks
- Evidence and Learning Records
Failure modes
Search again instead of forcing a recommendation.
Block execution until the conflict is resolved.
Retry only the failed surface and preserve the rest.
Return to search or refine the objective with the new context.
Primary CTA
Every recommendation should be executable.
AgentKit does not stop at advice. It produces the plan and evidence that let the founder act with confidence.