The principle
AI drafts; the business authorizes. This is the load-bearing rule for every AI-assisted feature in Kordless:- AI may extract, classify, organize, explain, and propose.
- AI may not invent a price, publish a rule, override policy, or silently mutate the Live Book.
- A suggested value, a calculated floor, an owner target, and an evidence-derived value are structurally distinct. No suggested value becomes an authorized rule without explicit owner verification.
What AI does in the Book
Extraction
When you provide pricing artifacts (spreadsheets, documents, screenshots), AI extracts claims from them: services, prices, modifiers, conditions, and constraints. Extracted claims carry provenance — you can see where each one came from.Guided first-pricing
If you do not have a price book, AI conducts a consequence-driven interview. It asks about your offerings, units, costs, time, capacity, minimum charges, and return constraints. From your answers, it proposes a first pricing model with visible derivation for every calculated or suggested value.Classification and normalization
AI classifies and normalizes customer requirements — mapping unstructured descriptions (“my check engine light came on and the car is shaking”) to typed factors the evaluator can use.Conflict detection
AI identifies conflicts between sources (for example, a price in your spreadsheet that disagrees with a price in your PDF) and missing information that would improve coverage. Conflicts are surfaced as suggestions, never silently resolved.Scenario generation
AI generates representative calibration scenarios — including boundary cases — so you can test the model against edge cases before publishing.Change proposals
When you give a natural-language instruction (“raise labor rate to $120/hour”) or when the system observes a pattern in your corrections, AI drafts a structured proposed diff. You review the affected services, statements, scenarios, and historical quotes before accepting.What AI does in quoting
Interpretation
AI interprets unstructured customer descriptions and media (photos, voice notes) to extract the services and selections the evaluator needs.Deterministic evaluation
The evaluator itself is deterministic — not AI. Given the same Book version and the same inputs, it always produces the same result. AI prepares the inputs; the evaluator computes the price. This means the same request produces the same result across HQ, hosted pages, embeds, and APIs. There is no variability from model temperature or prompt timing in the price itself.What AI may not do
- Invent a price. AI may not present an inferred, calculated, template-derived, or suggested value as the business’s existing or authorized price.
- Claim optimality without evidence. AI may not claim a candidate price is optimal without sufficient identified evidence and a defined objective.
- Personalize pricing using hidden factors. AI may not personalize a price using undisclosed benchmarks, missing cost assumptions, or protected customer characteristics.
- Publish an unverified rule. Every material Live rule requires owner verification before it reaches a published Book version.
- Change a Live Book directly. AI proposes changes; the owner reviews and publishes them through the compile-and-publish flow.
- Override approval policy. AI may not bypass the configured autonomy mode — if the Book is in approval-first mode, every result is held for owner review.
- Treat confidence as authorization. A high-confidence suggestion is still a suggestion. Confidence does not grant authority.
- Hide source conflict. When sources disagree, AI surfaces the conflict rather than picking one silently.
- Turn unsupported work into a plausible answer. When the evaluator cannot produce a reliable price, it returns
unavailableorfailed— never a guess.
The autonomy ladder
How quotes are handled after evaluation depends on the autonomy mode you choose. The ladder progresses from fully supervised to fully autonomous:Trust is earned in Shadow mode, or approval-first when there is no baseline. Kordless never manufactures an agreement score: when there is no human or historical baseline to compare, it runs approval-first, and the owner’s decisions become the evidence for later autonomy.
The agent runtime
On any active paid plan, the agent runtime is live. This means:- Quoting — Customer inquiries are evaluated against your published Live Book automatically.
- Follow-up — The agent follows up on sent quotes and pending commitments according to your configured policies.
- Outreach — Automated outreach runs according to your configuration.
Product boundary
Kordless stops cleanly at customer commitment. The following are explicitly out of scope — they are not features Kordless will build, and they do not affect the evaluator or commercial record:- Full scheduling, dispatch, and production management
- General CRM and customer database
- Invoicing, balances, payouts, and accounting
- Recall and reactivation campaigns
- Review management
- Website builder
- Forms builder
- Marketing automation
- SEO, GEO, visibility, and growth tools
- General-purpose inbox and messaging
- General automation builder
- AI employee, agent persona, or AI cockpit
- Inventory, payroll, ERP, fulfillment, and field operations
What “stops at commitment” means
Kordless handles the journey from pricing knowledge through customer commitment:- Make the Book — Turn artifacts and explanations into a verified pricing model.
- Verify the Book — Approve material rules through sentences and scenarios.
- Publish the Book — Make one immutable version available to every surface.
- Commit the offer — Collect requirements, evaluate, apply conditions, and support approval, hold, booking, or deposit through commitment.