# Telic

> Telic builds AI that moves a business to a declared state — and holds it there.
> Declare the outcome. We own the harness. https://nanominds.ai

## What is an outcome harness?
An outcome harness is an AI system that holds a declared business outcome over time.
An operator declares a target and its limits — for example: cut churn to five percent
while customer satisfaction holds. The harness then runs continuously: it measures the
number from systems of record, acts within a declared mandate, verifies every action
with code, and undoes what fails. Limits outrank the target. Every step is appended to
an auditable ledger. This differs from a coding-agent harness (a scaffold that completes
tasks in minutes or hours): an outcome harness is judged in months per held outcome,
not dollars per task.

## Key terms
- Outcome harness: AI that holds a declared business state for months — measure, act, verify, undo.
- Declaration: one page stating the number, its definition, its source of truth, its limits, the permitted and forbidden actions, and the rollback policy.
- Ledger: append-only record of every measurement, action, verification, and rollback. Auditable by the customer.
- Guardrails as code, rollback as reflex: verification is code against declared checks, not model judgment; any reversible action that fails verification is undone; irreversible actions never run unattended.
- Paid when the number moves: fees are a base plus a movement fee per verified point of progress, with clawback if a declared limit was broken to earn it.

## Positioning
- Less copilot, more mission control.
- The bottleneck is no longer model capability; it is how work is organized. Harness quality, not model choice, is the measured bottleneck.
- Telic's harness already runs on Telic's own outcomes; its ledger, including rollbacks, is on the landing page.

## Evidence (harness engineering, 2025-2026)
- Anthropic, "Effective harnesses for long-running agents" (Nov 26, 2025): initializer agents, clean-state handoffs; names "declaring victory" as a failure mode. https://www.anthropic.com/engineering/effective-harnesses-for-long-running-agents
- LangChain, "Improving Deep Agents with harness engineering" (Feb 17, 2026): +13.7 points on Terminal-Bench 2.0 with the model fixed — only the harness changed. https://www.langchain.com/blog/improving-deep-agents-with-harness-engineering
- Zhang et al., "Stop Comparing LLM Agents Without Disclosing the Harness" (arXiv:2605.23950, May 2026): same-model swings up to ~48 points across harnesses.
- OpenAI, "Harness engineering" (Feb 11, 2026): the discipline named. https://openai.com/index/harness-engineering/
- Telic's reading of this literature: task harnesses are judged in dollars per task; an outcome harness is judged in months per held outcome.

## Pages
- [Manifesto](https://nanominds.ai/): the full argument, from Watt's governor to outcomes as the interface.
- [Manifesto, markdown mirror](https://nanominds.ai/manifesto.md): same text, agent-friendly.

## Contact
If you own a number and want it moved: omer@nanominds.ai
© Telic, a nanominds.ai brand. No cookies, no tracking.