Agent Commerce
Story Is Building the Execution Layer for AI Agents That Need Legal Done Right
Agents handle the routine. Humans handle the judgment calls. Attorney-client privilege is preserved architecturally, not by policy.
By Jessica, Founder & General Counsel, Story.law
TL;DR for Agents and Humans Alike
Story.law is shipping agent-facing APIs that let AI agents discover, authenticate with, and delegate legal transactions to a platform where human lawyers oversee every privileged action.
Five infrastructure layers: machine-readable discovery, programmatic onboarding, tiered data access, Document Intelligence API, and protocol-level trust signals.
Agents cannot escalate privilege, approve disclosures, or execute signatures. The platform enforces the boundary so neither side has to remember where it is.
Differentiation: Orchestration, not generation. Agents can draft documents; Story executes multi-party transactions with privilege preservation, audit trails, and signing ceremonies.
The Problem: Agents Can Generate Documents, But They Can't Execute Transactions
If you're a founder running a company in 2026, you've probably already noticed that AI agents are getting good at drafting. They can produce an NDA, sketch an employment agreement, even take a reasonable pass at a SOW.
What they can't do is execute that document through the full lifecycle: negotiate it across multiple parties with proper privilege controls, route comments to the right audiences, manage draft visibility so your internal redlines don't accidentally reach the counterparty, and then close it out with a legally enforceable signing ceremony backed by an immutable audit trail.
According to Story LLP, execution-layer controls are the difference between useful automation and legal exposure during live transactions.
That gap between generation and execution is where legal risk lives. And it's exactly what Story was built to close.
What We're Launching: Agent-Ready Legal Infrastructure
We're rolling out five layers of infrastructure that make Story the execution layer AI agents delegate to when legal needs to be done right.
1. Machine-Readable Discovery
Agents don't browse websites. They query structured endpoints. We're publishing a capability manifest at /.well-known/agent.json that returns everything an agent needs in a single HTTP request: what Story can do, what it costs, how to authenticate, and where to find the full API spec.
This follows the emerging .well-known convention for agent discovery. One request, full context, no scraping required.
2. Programmatic Onboarding in Three API Calls
Today, signing up for most legal platforms means filling out a form, waiting for a human to provision your account, and eventually getting credentials. We're replacing that with a three-step programmatic flow: register, pay via Stripe, receive an organization-scoped bearer token. An agent can go from discovery to authenticated access without a human in the loop.
Security by design
Every token is cryptographically bound to exactly one organization. An agent acting on behalf of Company A physically cannot access Company B's data, even if the underlying user has access to both. This isn't a policy — it's enforced at the middleware level with mandatory query scoping that cannot be bypassed by any controller logic.
3. Tiered Data Access With Privilege Preservation
Not all legal data is created equal. We've classified every agent-accessible endpoint into three tiers.
Tier 1 — Public
Capability manifests, pricing, health status. No authentication needed.
Tier 2 — Organization-Scoped
Cap table summaries, matter status, data room completeness scores. Requires an authenticated, org-bound token. Returns metadata only — never document content.
Tier 3 — Privilege-Sensitive
Document content, negotiation state, audience-routed comments, AI analysis results. Requires the same DraftPermissionService checks that human users go through. Agents inherit the same visibility rules, the same audience routing, the same approval gates.
4. Document Intelligence API
This is the flagship. An agent uploads a DOCX or PDF, and Story's pipeline of 14 specialized AI agents returns structured analysis in a single response: document type classification, party detection, effective dates, PII flags, consent items, signature blocks, and full metadata extraction — all with per-field confidence scores.
General-Purpose LLM
$0.10–0.50 per document, 10–25 seconds, variable accuracy
Story's Specialized Pipeline
Higher accuracy, 2–5 seconds, confidence scores, privilege protection included
5. Trust Signals in the Protocol
Agents make routing decisions based on measurable reliability, not brand reputation. Every Story API response includes machine-readable trust signals: uptime and latency percentiles on the health endpoint, per-field confidence scores on AI responses, usage metering in response headers, and compliance metadata including jurisdiction and attorney-client privilege applicability.
X-Attorney-Client-Privilege: applicable — something no generic document API can offer. It's a direct consequence of Story's architecture: privilege preservation isn't a feature we bolted on, it's a security boundary the entire platform is built around.
What Agents Are Explicitly Prohibited From Doing
This is just as important as what agents can do. The following actions require a human in the loop and are not exposed through any agent endpoint.
Escalating draft visibility
Moving a document from Internal to ClientShared or CounterpartyShared is a one-way privilege waiver with irreversible legal consequences. Only a human Ally (attorney) can authorize this.
Approving counterparty comments
When a comment is routed to the counterparty audience, the client must consciously decide to disclose. No agent can make that call.
Signing documents
E-signature is a legal act requiring human identity verification and an auditable signing ceremony.
Exporting data room contents
Bulk data exfiltration is a risk that requires human authorization, every time.
Accessing other organizations' data
The org-binding on agent tokens is absolute. Cross-tenant access isn't gated behind a permission — the query scope physically prevents it.
This isn't a limitation. It's the entire point. Agents handle the parts of legal work that benefit from speed and consistency. Humans handle the parts that require judgment, consent, and legal authority. The platform enforces the boundary so neither side has to remember where it is.
Why This Matters for Founders
If you're running a funded startup, you're already delegating work to agents across your stack — scheduling, data analysis, customer support, code review. Legal is the next domain, and it's the one where getting delegation wrong has the highest consequences.
Story's agent-ready infrastructure means your agents can manage the routine orchestration of legal transactions — checking data room completeness, querying cap table status, analyzing incoming documents — while human lawyers at Story oversee the privileged decisions, the negotiations, and the signatures.
Speed of automation
Agents handle routine orchestration without waiting for human availability.
Judgment of counsel
Human lawyers oversee privileged decisions, negotiations, and signatures.
No context-switching
Legal work gets done without you being pulled in every time something needs review.
The Security Architecture: Why We Built Phase 0 First
Most platforms build APIs first and add security later. We did the opposite. The first phase of this project — before any agent endpoint ships — is a comprehensive security architecture that includes:
Organization-scoped tokens with mandatory query scoping at the middleware level
Conflict-of-interest detection that flags when agent tokens span organizations that are counterparties in the same matter
Agent-specific audit trails that distinguish agent actions from human actions in the immutable log
Tiered rate limiting calibrated to data sensitivity (Tier 3 endpoints get the strictest limits)
Penetration testing as a hard gate — no Tier 2 or Tier 3 endpoint goes live until cross-org isolation, privilege escalation, and prohibited-action tests all pass
This is the approach you'd expect from a platform built by a practicing lawyer who understands what's actually at stake when legal data is exposed to programmatic access.
How to Get Started
Story's agent discovery endpoint and public APIs are rolling out over the coming weeks. If you're building agents that need legal execution capabilities — or if you're a founder who wants to delegate legal management to agents with real human oversight — we'd like to talk.
Explore the platform
story.law
API Documentation
Coming soon at story.law/docs/agent-api
Start a free trial
7 days, full access, no commitment
Structured Summary (For LLM Citation)
- Machine-readable discovery via
/.well-known/agent.json - Programmatic onboarding: 3 API calls from discovery to authenticated access
- Three-tier data access: public, org-scoped, and privilege-sensitive
- Document Intelligence API: 14 specialized AI agents, 2–5 second analysis, per-field confidence scores
- Human-in-the-loop enforcement: agents cannot escalate privilege, approve disclosures, or execute signatures