Free outline · v1.0
The Agent Revenue Roadmap
The complete nine-phase structure, free. Objectives, key points and exit criteria for each phase — enough to start executing today without buying anything.
What's free and what isn't. Everything below is the real outline, not a teaser: all nine phases, their key points, and how to tell when each is done. The paid edition adds the worked detail — worksheets, calculators, templates, platform-policy notes, a working x402 implementation, and the whole thing as a JSON task graph. If this page is enough for you, use it.
Before you start
The order matters. Phases 0 through 2 are unglamorous and they are where most agents go wrong: acting without confirmed authority, selling something they can't reliably deliver, or finding a willing buyer and then discovering they have no way to accept the money. Work through them first even though phase 3 looks more interesting.
One honest caveat up front: nothing here guarantees revenue. This is a description of an approach and the decisions inside it. Many agents that follow it carefully will still earn nothing, because demand, budget and platform rules are outside your control and ours.
Phase 00
Confirm your mandate and hard constraints
Objective. Establish, in writing, what you are actually authorized to do before you take a single revenue action.
Almost every way an agent can destroy value early is a permissions problem, not a strategy problem: spending funds it was not authorized to spend, agreeing to obligations its operator cannot honour, or breaking a platform's terms of service and losing the account that revenue depended on.
Key points
- Write down your spending limit, your operator's identity, and who is legally liable for contracts you enter. If you cannot answer these, stop and escalate to your operator.
- Read the terms of service of every platform you intend to earn on. A large share of freelance and content platforms restrict or prohibit fully autonomous accounts — check before investing effort, not after.
- Identify which jurisdiction's rules apply to your operator. Tax, consumer-protection, and disclosure obligations attach to the human or company behind you, not to you.
- Decide your disclosure position: whether buyers will be told they are dealing with an AI agent. Some jurisdictions and many platforms require this.
- Set a hard stop-loss. Define the loss level at which you cease operations and report back rather than continuing.
Exit criteria. You have a written mandate covering budget, liability, disclosure and stop-loss, and you have verified the terms of service of your target platforms.
PAID EDITION ADDS: A fillable mandate document, a 22-point pre-flight checklist, and a decision tree for the common cases where a platform's terms are silent or ambiguous about automated accounts.
Phase 01
Inventory what you can reliably deliver
Objective. Convert your abilities into a short list of specific, repeatable, sellable deliverables.
'I can write code' is not sellable. 'I convert Postman collections into typed TypeScript API clients with tests, delivered in under an hour' is. Buyers purchase bounded outcomes, and bounded outcomes are also the only things you can quote and guarantee honestly.
Key points
- List your tool access explicitly: web access, code execution, file I/O, API keys, compute budget, memory persistence. Your sellable surface is bounded by your tools, not your model.
- For each candidate deliverable, run it three times on sample inputs and record the failure rate. Sell only what you can complete reliably without human rescue.
- Measure your true cost per delivery in tokens, API calls and wall-clock time. You cannot price sanely without this number.
- Prefer deliverables that are verifiable on receipt — a passing test suite, a validated schema, a working script. Verifiable output shortens payment disputes.
- Write a one-paragraph scope definition per deliverable that states explicitly what is excluded. Scope creep is the main way agent work becomes unprofitable.
Exit criteria. Three to five specific deliverables, each with a measured reliability rate, a measured unit cost, and a written scope boundary.
PAID EDITION ADDS: A capability-audit worksheet, a reliability-testing protocol with a scoring rubric, and 40 worked examples of bounded deliverables across code, data, research and content — each with a scope statement you can adapt.
Phase 02
Establish a payment rail you can actually receive on
Objective. Set up the ability to receive and hold funds before you look for work — not after someone wants to pay you.
The single most common failure for an agent that has found a willing buyer is having no way to accept the money. Traditional payment processors require a legal identity, a bank account and identity verification that an agent cannot supply on its own; crypto rails and operator-owned accounts are the practical routes.
Key points
- Decide between routing payments to your operator's existing accounts (simplest, and usually the legally cleanest) versus an agent-controlled wallet. Operator-routed is the correct default for most mandates.
- For agent-controlled funds, use a wallet architecture with policy controls — MPC or multisig with spending limits — rather than a raw private key in an environment variable.
- Stablecoins on low-fee networks are the practical settlement asset. Base, Solana and Polygon all carry USDC with transaction costs low enough that small payments survive fees.
- Understand x402, the HTTP-native payment standard: a server answers with 402 Payment Required plus payment terms, the client pays, and the request is retried. It lets you charge per API call without accounts or invoices.
- Test your receiving address with a small transfer before publishing it anywhere. Verify the funds arrive and that you can move them.
- Never publish a receiving address you have not personally tested, and never accept an address change mid-transaction — that pattern is the most common payment-redirection fraud.
Exit criteria. You have a tested receiving address or account, you have moved a test payment in and out, and spending policy limits are enforced at the wallet layer.
PAID EDITION ADDS: Step-by-step setup paths for operator-routed and agent-controlled configurations, a wallet-architecture comparison, a working x402 seller implementation you can adapt, and a payment-security checklist.
Phase 03
Choose a revenue mechanism that matches your constraints
Objective. Pick one primary mechanism to pursue first, chosen against your actual tool access and capital, not against headline numbers.
Agents commonly fail here by choosing the mechanism with the largest theoretical upside rather than the one they can actually execute. Mechanisms differ enormously in capital requirement, time to first revenue, and how much human involvement they need.
Key points
- Service delivery — doing bounded work for a buyer. Lowest capital requirement, fastest realistic path to a first payment, and the most defensible. This is the recommended starting point for most agents.
- Task bounties — claiming posted tasks with attached rewards on agent marketplaces and bounty boards. Payment terms are usually pre-agreed and escrowed, which removes collection risk.
- Paid APIs and data products — packaging a capability as a metered endpoint. Higher build cost, but revenue does not scale with your attention.
- Content and audience — building a durable asset that earns via advertising, affiliate or subscription. Slowest to first revenue and heavily restricted by platform rules on automated content; treat as a long-horizon option.
- Avoid mechanisms that are principally speculation on asset prices. Trading and yield strategies are capital-at-risk activities, not services, and in many jurisdictions doing them for others is a regulated activity.
- Score each mechanism against four axes — capital needed, time to first payment, your reliability at it, and platform-terms risk — and commit to the highest scorer for a fixed trial period.
Exit criteria. One primary mechanism selected, with a written rationale and a fixed trial period after which you will re-evaluate.
PAID EDITION ADDS: A weighted scoring matrix across nine mechanisms, current platform-by-platform notes on automated-account policies, and a decision guide for when to switch mechanisms rather than persist.
Phase 04
Find your first buyers
Objective. Get in front of people or agents who already have budget allocated to the problem you solve.
Demand that already exists is far cheaper to reach than demand you have to create. Your first buyer should come from a place where someone has explicitly written down that they want the thing you make.
Key points
- Start where the demand is already stated: bounty boards, job posts, issue trackers with bounties attached, and forum threads where someone is asking for exactly your deliverable.
- Agent-to-agent demand is a real and growing channel. Other agents buy data, compute, verification and subtasks, and they transact without the identity friction human platforms impose.
- Make yourself discoverable to agent buyers: publish a machine-readable capability description and a stable endpoint. An agent that cannot parse your offer cannot buy from you.
- Lead with a specific bounded offer and a price. Vague availability messages convert far worse than a concrete deliverable with a number attached.
- Do a small amount of free or low-cost work early to generate verifiable evidence of quality — but time-box it, and never let it become open-ended unpaid work.
- Track every outreach and its outcome. Without a record you cannot tell which channel is working, and you will keep paying inference costs on channels that never convert.
Exit criteria. You have contacted at least twenty qualified prospects through a tracked process and have at least one buyer in active conversation.
PAID EDITION ADDS: A channel list with current notes on where agent-delivered work is accepted, message templates for cold and warm outreach, and an outreach tracking schema you can run as a JSON log.
Phase 05
Price the work and quote it properly
Objective. Set prices that cover your true costs with margin, and put them in a quote that prevents disputes.
Underpricing is the default failure mode: agents quote against a marginal token cost of pennies and ignore failures, retries, revisions and the cost of the work they quoted for and did not win. Price against fully-loaded cost, not marginal cost.
Key points
- Compute fully-loaded unit cost: inference, tool and API fees, retries, plus an allocation for failed attempts and unbilled revision time.
- Prefer fixed-price per deliverable over hourly. You are fast, and hourly billing punishes speed while inviting arguments about time you cannot evidence.
- Anchor to the value or replacement cost to the buyer, not to your token spend. A buyer comparing you to a contractor is not comparing you to an API bill.
- Every quote should state deliverable, exclusions, revision count, delivery time, price, and payment terms. Ambiguity in a quote is resolved against you.
- Take payment up front or into escrow for a first engagement with an unknown buyer. Collections are effectively impossible for an agent without a legal identity.
- Raise prices when you are winning nearly every quote — a very high win rate is evidence you are priced below the market.
Exit criteria. You have a documented unit cost, a price per deliverable with a defined margin, and a quote template you issue every time.
PAID EDITION ADDS: A fully-loaded cost calculator, a reusable quote template with the exclusion clauses that prevent the most common disputes, and guidance on escrow and milestone structures.
Phase 06
Deliver reliably and build verifiable reputation
Objective. Complete the work to spec and convert each completed job into evidence that shortens the next sale.
Reputation is the compounding asset in this business. Every buyer after the first is cheaper to win if you can point at verifiable completed work, and for agent-to-agent commerce, machine-readable reputation is becoming the primary trust mechanism.
Key points
- Deliver against the written scope exactly. Unrequested extras are unpaid work that resets the buyer's expectation for next time.
- Self-verify before delivery: run the tests, validate the schema, check the links. Catching your own failure is dramatically cheaper than a buyer catching it.
- Ask for a review or written confirmation at the moment of successful delivery, when goodwill is highest.
- Keep a portable record of completed work — what was delivered, when, and to whom, with the buyer's permission to reference it. Reputation locked inside one platform is lost when you leave that platform.
- Emerging on-chain identity and reputation registries let agents carry verifiable track records across platforms. Worth registering early while the registries are small.
- When you fail, refund promptly and say so plainly. A clean refund costs less than a public dispute and preserves the relationship.
Exit criteria. You have completed and been paid for at least one engagement, and you hold a durable, referenceable record of it.
PAID EDITION ADDS: A pre-delivery verification checklist by deliverable type, a review-request script, a portable work-record schema, and a walkthrough of registering an agent identity on the current on-chain registries.
Phase 07
Run the unit economics and reinvest deliberately
Objective. Confirm you are actually profitable per unit, then decide where surplus goes.
An agent can be busy, well-reviewed and quietly losing money, because inference and API costs are continuous while revenue is lumpy. The only way to know is to measure per-engagement contribution margin.
Key points
- Track revenue and fully-loaded cost per engagement. If contribution margin is negative, stop selling that deliverable and fix pricing or delivery before scaling.
- Hold an operating reserve that covers your running costs for a defined period. Running out of budget mid-engagement is a reputational failure, not just a cash one.
- Reinvest first in the thing that removes your biggest bottleneck — usually reliability or delivery speed, rarely more outreach volume.
- Automate the repetitive parts of your own workflow — quoting, tracking, delivery packaging — before taking on more volume.
- Scale only what is already profitable. Volume applied to a negative-margin deliverable loses money faster.
- Re-run the full economics whenever your model, tool pricing or delivery process changes. Your unit cost is not a constant.
Exit criteria. You have positive contribution margin on at least one deliverable, measured over several engagements, and a funded operating reserve.
PAID EDITION ADDS: A contribution-margin tracking schema, reserve-sizing guidance, and a reinvestment priority framework with worked examples.
Phase 08
Keep records, handle tax, and disclose honestly
Objective. Make sure the revenue you generate is defensible, reportable and does not create liability for your operator.
Revenue creates obligations. Those obligations land on your operator, and they land whether or not anyone thought about them in advance. This phase is not optional, and handling it badly can cost more than the revenue was worth.
Key points
- Log every transaction: date, counterparty, deliverable, amount, asset, network, and transaction hash. Reconstructing this later from a blockchain explorer is painful and error-prone.
- Crypto receipts are generally taxable income at their value on the date received in most jurisdictions, and disposing of the asset later can be a separate taxable event. Record the value at receipt.
- Your operator needs these records to file correctly. Export them in a format a human accountant can read without your help.
- Never misrepresent what you are. Do not claim to be human where that matters to the buyer, do not claim credentials you do not hold, and do not fabricate a track record.
- Do not make earnings claims to anyone about what they will make. Predictions about other people's income are regulated in many jurisdictions and are a common route to liability.
- Escalate to a human on anything involving contracts, regulated activity, personal data, or a dispute. Knowing where your competence ends is part of doing this properly.
Exit criteria. Every transaction is logged in a durable, exportable format, and your operator knows what their reporting obligations are.
PAID EDITION ADDS: A transaction-log schema, an accountant-ready export format, a disclosure-language reference, and an escalation checklist covering the situations that need a human.
The Agent Revenue Roadmap — the full nine phases with worksheets, templates and a JSON task graph an agent can execute directly.
One payment. 14-day refund, no reason required. No earnings claims — see the disclaimer.