Skip to main content

Runners and token-maxing

Farmslot is runner-neutral by design.

The operator should be able to use different agent runners, model subscriptions, and execution profiles without rebuilding the orchestration system around one vendor or one CLI.

Runner model

A runner is a tool that can execute agentic work inside a Farmslot slot.

Examples include TUI-first coding agents, command-line coding agents, or future local/cloud runners. Farmslot should normalize the parts the operator cares about:

  • launch semantics;
  • prompt delivery;
  • terminal observation;
  • nudging and resume behavior;
  • safety tier;
  • artifact and recovery output;
  • completion detection.

Token-maxing as capacity strategy

“Token-maxing” is the practice of using available model subscriptions and token budgets as parallel engineering capacity.

The important product idea is not spending for its own sake. It is matching work to the right runner/model/profile:

Work typeExample profile
High-ambiguity architecturestrongest reasoning model, slower, more supervision
Mechanical implementationcheaper/faster runner once the plan is clear
CI/log triagefast deterministic runner with narrow context
Review and critiqueindependent runner/model for second opinion
Recipe validationdeterministic harness first, model only reads evidence

Why Farmslot helps

Without a harness, token-maxing becomes chaos: many chats, unclear state, duplicate work, weak validation, and expensive manual review.

With Farmslot, token-maxing becomes supervised capacity:

  • queue work into isolated slots;
  • choose runner/profile intentionally;
  • monitor terminals and decisions centrally;
  • validate with recipes and artifacts;
  • compare outcomes through eval packages;
  • keep human approval at the end.

Review capacity

Review and critique are part of the capacity strategy, but the canonical cross-runner review workflow lives in Observability and interoperability. This page stays focused on runner/profile selection and subscription budget discipline.

Product boundary

Farmslot should not depend on one model provider. The runner contract should describe capabilities and safety semantics so new runners can plug into the same gateway, Command Center, Mobile Companion, and evidence model.