Agreements for the agent economy.

Imagine a founder asking an agent to renew a supplier agreement within an annual budget of $10,000. The founder expects another year. The agent returns a signed renewal with an annual price of $9,000. The price fits, but the new term is three years and renews automatically. This is a hypothetical example, not an OctoDoc product demonstration.

The instruction left an important limit unstated. Checking the amount cannot settle the length of the commitment. A signature may faithfully record an authorized act without establishing that the person understood or intended every consequence.

We believe agents will increasingly coordinate purchases and agreements with other agents. David Marcus describes the shift toward agent-led payments in The Death of the Interface. Visa reported controlled agent-initiated purchases in December 2025. These are foundations for a direction, not evidence of universal adoption. [1] [2]

For agreements, the question is what must survive that change of interface: the parties, their intent, the limits on each agent, and the exact terms each side accepts. Our renewal example follows those questions through the paper.

In Wallet Wars, Terry Angelos separates the ability to complete a payment from preserving the customer relationship behind it. He also distinguishes authorization, correct execution and purchase protection. Those distinctions suggest a useful discipline for agreements: name each problem before claiming to solve it. [3]

For the supplier renewal, three questions matter. Does the agent have permission to sign? Is this the exact file that permission covers? Do its terms meet the founder's needs? An affirmative answer to either of the first two does not answer the third.

The three-year term is visible in the document. Whether it is acceptable depends on the person's intent and circumstances. An amount ceiling alone cannot decide that. The same applies to renewal conditions, confidentiality duties and termination rights.

OctoDoc separates cited reading, authority and execution evidence. A reader can inspect the source passage before deciding. Controls bind an authorized act to the exact transaction. The retained record helps establish what happened; it does not promise that the commercial decision was wise.

OctoDoc accepts an existing PDF, converts a Word file into a PDF for review, or opens a source-checked reusable PDF form from the catalog. It does not generate or rewrite the operative wording of an agreement. The sender supplies the document and decides whether it is suitable.

Only a person approves every converted page and chooses Use this PDF; no machine route can make that choice. The sender then prepares the fields and confirms their placement. The signer reviews the document and makes a separate signing decision.

For the renewal, the founder reviews the supplier's actual terms before signing. After the required signatures, each party receives the same sealed PDF. The file they keep contains the agreement they reviewed and signed.

Answers and summaries remain outside the signed content. They help a reader navigate the agreement; they are not a substitute for the advice of an attorney.

An agent can help fetch a file, locate a passage and prepare an action. Its account credential identifies what it can access. That credential alone cannot authorize sending or signing.

A binding machine act needs authority tied to the exact document and proposed transaction. A person can approve that act through an attended, verified browser approval. Alternatively, an explicit delegation mandate can permit a defined act within stated limits. Existing preparation and confirmation requirements still apply.

Those limits can constrain the operation, counterparty, document class, value and time period. A mandate can be revoked, and an agent cannot expand its own authority. A mandate simulation reports whether a prepared transaction meets the enforced limits and gives reasons for a refusal.

For the renewal, a successful simulation is not a judgment that every clause meets the founder's wishes. If the document changes after authorization, the changed act needs matching authority. That protects the approved version; it does not interpret an unstated preference for a one-year term.

After the required signatures, OctoDoc seals the resulting PDF using PAdES B-T: a cryptographic signature with an external signature timestamp. If the required timestamp cannot be obtained, sealing stops. Answers and summaries do not alter the signed byte range.

For the renewal, the sealed PDF preserves the signed terms. Evidence of identity, consent and authority explains the act. These serve different purposes. Neither establishes that the supplier performed its duties or decides who must remedy a mistake.

Customers can retain the sealed file and export its event record. A matching file digest shows that the bytes match a reference. The ledger verifier checks the linked events and their available external timestamp evidence. A digest check is not certificate validation or proof of signer identity.

The record is tamper-evident, not administrator-proof. OctoDoc makes no promise of public trust-list membership, indefinite certificate validation or archival refresh. Both parties can keep the same agreement and inspect its evidence without making their choice of interface the only place the relationship is recorded.

Electronic contracting is an established legal category. Under 15 U.S.C. 7001(a), covered records and signatures cannot be denied effect solely for being electronic. Section 7001(h) addresses records involving electronic agents, subject to legal attribution to the person to be bound. [4]

The Uniform Electronic Transactions Act addresses formation through electronic agents in UETA §14 and attribution in UETA §9. Its application depends on the law adopted in the relevant jurisdiction. [5]

These provisions do not make every agreement valid or every AI action authorized. Intent, capacity, terms, consumer protections and statutory exclusions still matter. The legal foundation makes electronic interaction possible; the facts of a particular transaction remain important.

OctoDoc’s role is to preserve evidence of the act and its authority. Controls contribute to the weight of that evidence. They do not decide a dispute or change which party must prove what. This paper is not a substitute for the advice of an attorney.

Confidential agreements deserve a clear account of what is recorded. OctoDoc serves a read link with a strictly necessary baseline before optional reading telemetry begins. Per-page reading measurements require an affirmative consent decision; a Global Privacy Control signal is treated as declined.

Reading activity is not proof of comprehension or consent. Opening a document, spending time on a page and agreeing to its terms are separate events. Device and transport fingerprinting are not part of the product.

People can download and retain their agreement. Hosted access follows the applicable retention terms. Private storage and access controls protect normal operation, but the service does not claim write-once storage or protection against every privileged administrator action.

Customers should assess these boundaries against their own needs. OctoDoc’s Trust Center publishes the security posture, privacy terms and current subprocessor register at octodoc.org/trust#subprocessors. A clear limit is more useful than an assurance the product cannot support.

The founder and supplier remain the parties whether the renewal starts in a browser or an agent conversation. Each needs its own authority and agreed record.

The browser supports the document lifecycle. The installed command line and agent plugin let Claude Code and Codex help with the same work. A linked Slack workspace provides status and cited document answers under the relevant consent.

Sending and signing require explicit authority. A Slack message, terminal reply or signature image is not a substitute. OctoDoc does not offer autonomous contract negotiation or a shared mandate exchange between organizations.

Our thesis is that agents will coordinate while each party controls its own commitments. More connected tools alone cannot establish that separation.

The immediate customer value is a shorter path from an existing document to a signed record. Preparation, cited reading, signing and evidence belong to the same process. The founder can use that process even if the supplier has no agent.

OctoDoc does not meter sends or workspace membership. Signers are not billed to sign, read or verify. Current allowances, prices and retention windows are published at octodoc.org/about#pricing. The same seal level applies across plans.

A sender can request payment on a file through its connected Stripe account. The named party pays after the file is sealed. Payment does not establish performance of the agreement, and it gives an agent no independent spending authority.

Each agreement introduces another party to the experience. That creates an opportunity to earn repeat use, not an automatic network advantage. Our commercial thesis depends on making the next agreement easier to prepare, understand and authorize. Customers should return for that usefulness while retaining their own signed records.

Return to the founder's renewal. Before any signature, the useful questions are concrete: Which supplier? Which version? What price and term? What authority covers this act? Which questions remain for the person? A simple interaction makes those distinctions visible.

The technical task is to bind authority to the exact act and preserve evidence of the result. The adoption task is to explain those controls in language people can use. A check that permits an action must not be mistaken for a judgment on every term.

After signing, the record must remain intelligible to both parties. An automatic renewal clause still matters after a payment clears or an agent conversation ends. Keeping the agreement is a foundation for that continuing relationship; it does not itself monitor or enforce the obligations.

We believe agent-to-agent interaction will become an ordinary way to do business. OctoDoc's ambition is to make agreements as simple to understand and authorize as payments, with the care lasting obligations require. People set the terms and limits. Software handles more of the work. Both sides keep the record.