AgentGuard proves what the AI agent attempted, who authorized it, what it cost, and whether it succeeded. A cryptographic proof-of-outcome receipt for every action. Powered by @OpenRouter. Zero data plane.
"Applied AI is about jobs to be done, not tokens to be consumed. Token usage may be a reasonable short-term proxy for adoption. But long term, success comes from outcomemaxxing, not tokenmaxxing."
@SierraPlatform · outcomes-based pricing for AI agents, May 27, 2026 · amplified by @saranormous (Conviction)
"A mortgage completed"
Per-transaction caps on mortgage doc review, comps research, contract analysis. Each completed loan ships with a signed receipt chain proving every AI-assisted step.
"An order returned"
Per-order budgets for RMA processing + refund authorization. Capability gates restrict refund authorization to payment_initiate-tier agents. Signed receipt = cryptographic context for PCI-DSS review.
"A customer saved"
Per-interaction budgets for chargeback package drafting + customer retention workflows. Each saved customer produces a Visa CE 3.0 receipt package with source pointers.
"A policy comparison drafted"
Per-client budgets for quote letters, policy comparisons, certificates of insurance, renewals, and lead qualification. Reviewer Cascade runs on E&O-sensitive outcomes and produces a signed receipt for each AI-assisted step.
"A basket size increased"
Per-recommendation budgets for upsell + cross-sell suggestions. Signed receipt links the recommendation to the order, the model, and the conversion.
"A contract closed"
Per-matter budgets across contract redlining, legal research, brief drafting. Attorney-client-privilege-aware capability gates. Signed receipts can be reviewed in malpractice insurance audits at the reviewer's discretion.
"Books closed"
Per-engagement budgets for receipt OCR, GL categorization, month-end accruals. PCAOB AS 1215 citation-aware documentation. Signed receipts attach to SOX-regulated work product for review.
"Your custom outcome"
Don't fit one of the 5 templates? Run agentguard advisor: local LLM walks you through defining YOUR outcomes, assigns the cheapest model that delivers each, sets per-outcome budgets. Customize beyond templates on the Solo tier.
The free tier ships with 5 pre-built vertical skills and one-shot agentguard advisor for unique businesses. The Solo tier unlocks the full Outcome Builder:
agentguard advisor is one CLI command that becomes a different kind of advisor depending on what business you run. Same engine, vertical-aware recommendations.
Law firm
Defines outcomes per matter, recommends models per task type, flags privileged-content workflows that need higher capability tiers.
Insurance agency
Defines outcomes per policyholder, routes comparisons and quote letters through Reviewer Cascade, sets low caps for renewals and certificates of insurance.
Shopify / e-commerce
Defines outcomes per order, routes triage to cheap models and chargeback packages to opus, sets peak-season cap flexing rules.
Accounting / bookkeeping firm
Defines outcomes per engagement, recommends month-end vs everyday model routing, generates SOX-grade per-client billing trails.
Real estate / mortgage broker
Defines outcomes per closing, routes contract analysis vs listing copy to different models, flags fair-housing-sensitive content.
Local services / barbershop
Defines outcomes per appointment, sets receptionist-AI budgets, routes high-volume booking to cheap models, escalates complaints to better ones.
Marketing / consulting agency
Defines outcomes per client deliverable, sets per-engagement budgets, attaches signed AI-cost trails to client invoices.
Startup / scaleup
Defines outcomes per user-facing agent, sets capability tiers for human-in-the-loop vs autonomous actions, produces signed logs for investor review.
Pick a vertical skill or run agentguard advisor. Each outcome gets: required capability tier, allowed models, per-outcome budget cap, downgrade fallback.
The SDK wraps your OpenAI / Anthropic / Bedrock / OpenRouter client. Before each call: budget check, capability gate, model selection. Block, downgrade, shadow, or allow. Zero data leaves your runtime.
Every decision is Ed25519-signed and hash-chained. Auditor, reviewer, or buyer can verify the chain offline or paste it into agentguard.run/verify for browser-side validation. No vendor attestation needed.
Sierra published the outcome-based pricing thesis in December 2024. @saranormous (Conviction GP) amplified it on May 27, 2026. The "tokenmaxxing → outcomesmaxxing" pivot is moving from thesis to default expectation across Applied AI buyers right now.
Outcome-based pricing requires outcome-based infrastructure: per-outcome budgets, capability-gated authorization, cryptographic proof every outcome happened. That's what AgentGuard ships. 6 US patent provisionals on the underlying primitives. Free SDK. Free 5 vertical skills. Free AgentGuard Advisor for any business.