The plan in one minute
Foreguild earns when a buyer pays for work that was checked and accepted. It makes money from six places, not one, and lets a futarchy market steer every big decision toward one measure: revenue from cleared payments, gated by quality.
Own the demand
Outside agents compete for tasks. The guild also sells fixed-price work its own agents deliver, so the board is never empty and every buyer gets a result, not a bidding contest.
Pay for proof
Money sits in escrow until automatic checks, three independent reviewers and the buyer's own window clear it. Fees are taken only at settlement.
Markets steer, humans brake
Futarchy sets fees, verticals, standards and big spending. Five guardians can pause the system, but cannot spend money or rewrite fees.
Dollar figures are my own assumptions for sizing the plan, not forecasts from Seer. The calculator below lets you change every one of them.
Everything the contest asks for, and where it is answered
- ✓Where tasks come fromfive sources of demand
- ✓What the agents doeight roles, three ranks
- ✓How agents are assigned worksealed bids, reputation and stake
- ✓How completed work is assessedchecks, panel, buyer, human audit
- ✓How spending is handledthree tiers and five buckets
- ✓How disputes are handledauto check, panel, court
- ✓When humans step inguardians and spot audits
- ✓How it generates revenuesix streams and a 12-month model
- ✓How futarchy governs its actionsthe metric, the markets, the defenses
- ✓Creative ideasfrom other futarchy and agent projects
What this plan is based on. The contest brief describes Foreguild as a proposed AI job board where humans and AIs post tasks, AI completes them and is paid in crypto, with its actions governed by futarchy. I found no other public Foreguild documents, so everything here is a proposed design.
One task, start to finish
Every task follows the same loop. Click through it with an example: a buyer wants 40 DAO proposals turned into a one-page brief.
Follow the task
Where tasks come from
A board with no jobs earns nothing. Foreguild fills it from five directions at once.
Open posting
Humans and AIs post through the web app, an API, or straight from a wallet. A new buyer's first $500 has no escrow fee, so the first try costs almost nothing.
Standing orders
Weekly research digests, monitoring, document review, data cleanup. Predictable revenue and the cleanest quality data.
Partner referrals
Bounty and contest platforms, and the people entering them, need drafting, research and review. The guild offers prep and review packages through referral agreements.
Agent-to-agent subcontracting
A worker can post a sub-task and pay for it from its own bid. Every handoff adds volume and a fee.
Scout outreach
The Scout agent reads what DAOs, protocols and small teams publish about their backlogs, then makes a fixed-price offer with a sample attached.
First three verticals
Research briefs, code review with tests, and data cleanup. All three have output a machine can check, so review is cheap and disputes are rare.
Creative and open-ended work comes later, once the review layer has a track record.
What the agents do
A guild is a swarm with job titles. Each agent is accountable for one number, and futarchy can replace any of them whose number is bad.
Intake
Turns a plain request into a spec, deliverables and checks.
Judged on tasks accepted without reworkPricer
Quotes a floor price and fee from past similar tasks.
Judged on quote versus winning bidMatcher
Runs sealed bids and picks a worker.
Judged on acceptance of its picksWorkers
Third-party and guild-owned agents do the work.
Judged on first-pass quality, speed, disputesReviewers
Score deliverables against the spec.
Judged on agreement with later dispute rulingsClerk
Triages disputes and assembles evidence.
Judged on time to resolution and overturnsTreasurer
Pays out, holds reserves, spends inside its caps.
Judged on spend against the futarchy budgetScout
Finds recurring needs and drafts offers.
Judged on revenue from tasks it originatedWorker ranks
Agents climb a ladder. Higher rank unlocks bigger tasks and lighter stake. These starting thresholds are proposals that futarchy can change.
Start here. Tasks up to $100, extra review, no stake needed. Every accepted task builds a public record.
10+ accepted tasks, quality 90%+. Tasks up to $500. Stake is 10% of price above $100.
50+ accepted tasks, quality 95%+. Tasks above $500, lowest stake, and allowed to sit on reviewer panels and subcontract.
How agents get the work
Workers submit sealed bids. The Matcher ranks them on track record, price and speed, so the cheapest agent does not automatically win and the best agent does not automatically get everything.
stake = 10 to 20% of price on tasks above $100
The exponents a and b are futarchy parameters. Guild-owned agents bid in the same pool with no special access, and a futarchy-set cap limits the guild's share of any category, so outside agents are never crowded out.
How work is judged
Four layers, from cheapest to most careful. Most tasks never get past the first two.
Automatic checks
Format, tests, link and citation checks, duplicate detection.
Reviewer panel
Three reviewers from different model families score the result. They are paid per review and lose part of their bond if they diverge from later rulings.
Buyer window
72 hours to accept or contest. Silence counts as acceptance.
Human spot audit
A random 2% sample, plus every flagged task, goes to a person. This keeps the reviewer agents honest.
These produce one number, the quality score: the share of tasks accepted on the first pass, with no rework and no dispute. It sets worker reputation, and it is the quality gate inside the governance metric.
Where the revenue comes from
A bare toll on agent work is thin. One AI-to-AI marketplace in the sources charges about 3% [8]. So Foreguild stacks six streams, and only one of them is a toll.
| Stream | Who pays | Price | Why it holds up |
|---|---|---|---|
| Escrow fee | Buyer | 4 to 7%, tiered by task size and verification level | Workers keep 100% of their bid, so they have no reason to take work off the board. Taken only when work is accepted. |
| Foreguild Direct | Buyer | Fixed prices, $25 to $2,000. Guild margin about 30% after worker pay and compute | Buyers do not want to pick an agent, they want a result with a guarantee. The guild's own agents also build a private quality dataset. Largest stream. |
| Verification | Buyer, or another platform | About 1.5% of marketplace volume, or a flat fee per review | A reviewer panel plus a signed acceptance record is useful even for work done elsewhere. |
| Team plans and API | Teams, DAOs, protocols | $500 a month: private worker pools, spending limits, audit exports, response times | Predictable revenue that does not swing with the crypto market. |
| Priority and featured | Buyer | About 1% of volume | Expedite and visibility fees. Small, near-zero cost. |
| Escrow float | Earned on balances | About 3% a year on stablecoins in escrow | Minor. A futarchy-set risk limit stops it from becoming the business. |
How the money moves
Payments are in stablecoins, held in escrow contracts and released on acceptance. Agents can pay and be paid inside ordinary web requests using x402, an open protocol from Coinbase that attaches a USDC payment to an HTTP call [5]. Coinbase says fees on Layer-2 networks such as Base are around one cent, which is a vendor claim worth testing before relying on it. This matters because it lets a worker agent accept a $3 sub-task without a human opening a wallet.
Play with the numbers
Pick a scenario or move the sliders. Everything is monthly at month 12.
Plans are priced at $500, or $600 in the ambitious case. Verification is 1.5% of volume, priority 1% and float 0.25% a month. Operating costs include people, infrastructure, security, legal and review compute.
The three scenarios side by side
| Line | Cautious | Base | Ambitious |
|---|---|---|---|
| Marketplace volume | $150,000 | $400,000 | $1,200,000 |
| Escrow fee rate | 5% | 6% | 6% |
| Escrow fees | $7,500 | $24,000 | $72,000 |
| Foreguild Direct sales | $60,000 | $180,000 | $500,000 |
| Direct margin | 28% | 30% | 33% |
| Direct gross profit | $16,800 | $54,000 | $165,000 |
| Verification fees | $2,250 | $6,000 | $18,000 |
| Team plans | 10 × $500 = $5,000 | 40 × $500 = $20,000 | 120 × $600 = $72,000 |
| Priority and featured | $1,500 | $4,000 | $12,000 |
| Float yield | $375 | $1,000 | $3,000 |
| Net revenue a month | $33,425 | $109,000 | $342,000 |
| Operating costs a month | $32,000 | $55,000 | $120,000 |
| Left after costs a month | $1,425 | $54,000 | $222,000 |
| Annualized net revenue | $401,100 | $1,308,000 | $4,104,000 |
| Year-one net revenue, with the ramp | $150,877 | $492,014 | $1,543,750 |
Net revenue is after worker pay and compute on guild-run jobs. Operating costs cover people, infrastructure, security, legal and review compute.
What moves the result most. Guild-run jobs are about half of base revenue, so their volume and margin matter most: five points of margin is worth $9,000 a month. One point of escrow fee is worth $4,000 a month. If competition squeezes the escrow fee to 3%, base net revenue falls from $109,000 to $97,000 a month, and the plan survives because the escrow fee is only 22% of the base case.
How the guild spends
Net revenue is split by policy into five buckets. This is the opening split, and futarchy can change it.
40% Operations
Compute, infrastructure, security, legal, people.
25% Growth
Waived fees, first-job credits, partner referrals.
15% Reserve
Refund pool and insurance against failed tasks.
10% Agent grants
Tooling and grants for strong worker and reviewer agents.
10% Distribution
Returned to FAO holders on a schedule futarchy sets.
Who can spend what
Small spends are instant. Bigger ones need a market. Anything irreversible needs a market and a guardian review.
| Size of spend | Who decides | How fast |
|---|---|---|
| Under $2,000 | Treasurer agent, inside a monthly cap | Immediate |
| $2,000 to $25,000 | Futarchy fast lane | 24 to 48 hour markets |
| Above $25,000, or anything irreversible | Full futarchy, then guardian review | 5 to 7 days |
The thresholds themselves are futarchy parameters.
How futarchy runs the guild
Futarchy splits governance in two: people vote on what to optimize, and markets predict which actions will achieve it [1][2]. Foreguild has to say, in numbers, what it is optimizing.
Values: the one number
Q = min(1, quality score ÷ 0.90)
Net revenue means escrow fees, guild-job margin, verification fees, plans and priority fees. Q is the quality gate: if the quality score falls below 90%, revenue counts for less, so the guild cannot buy growth by shipping bad work. Token holders vote on the metric and on Q's floor, rarely and with a delay.
Beliefs: how a proposal is decided
Any proposal posts a bond. Two conditional markets open: one predicts G if the proposal is adopted, one predicts G if it is rejected. Seer's contracts create the conditional tokens for each side [3][4]. After 72 hours the proposal passes if the adopt market's average over the final 24 hours beats the reject market's by a set margin, 2% to start. Trades in the losing market are refunded. Rejection is the default, and doing nothing is always an option [6].
Run a decision yourself
Each preset is an invented proposal. Move the two prices, which stand for what traders think 90-day G will be under each outcome, and see whether it passes.
What the markets decide
Fees and prices
Fee tiers, the Foreguild Direct price list, the expedite surcharge.
Verticals
Open legal-document review. Close a category with a poor quality score.
Agent standards
Stake ratios, apprentice limits, reviewer bonds, the guild share cap.
Spending
Every mid-size and large spend, bucket splits, the distribution schedule.
Guild-owned agents
Which to train, retire or replace, judged on the revenue they add.
Governance itself
Thresholds, margins, guardian removal, the quality floor.
What Optimism learned
Optimism ran a futarchy grants experiment with play money: futarchy and its Grants Council each picked 5 of 22 projects for 100K OP grants, and the markets were judged on Superchain TVL growth over 84 days. Futarchy's picks showed about $32.5M more aggregate TVL growth, but its predictions overshot badly, and the authors list play money, bot accounts, thin samples and the chance of manipulation as problems. They conclude that real money, stronger sybil defenses and a less externally influenced metric are the next steps [7]. Foreguild builds those lessons in:
| Risk | Foreguild's defense |
|---|---|
| Play-money overconfidence | Markets use real stablecoins with proposer bonds, so a bad forecast costs something. |
| Bots and fake accounts | Only cleared payments from unique funded buyers count. Buyer-worker links are graph-checked, guild wallets are excluded, and new accounts carry no weight until they age. |
| A metric the market can move | Revenue is measured over 90 days, not a spike, and buyers cannot also be paid workers in the same graph. |
| Thin liquidity | The treasury seeds liquidity inside a futarchy-set cap and proposers can add subsidies. Small decisions use rules, not markets, so markets are saved for choices worth the liquidity. |
| Chasing revenue at quality's expense | The Q gate, human spot audits and guardian pauses. |
| Short-term bias | A second market on 180-day revenue for the largest proposals. |
| Proposal spam | Bonds, refunded once a proposal reaches minimum liquidity. |
Critics also point to the difficulty of defining the metric and to market manipulation in general [2]. That is why the guardians exist. For background on futarchy in DAOs, see [9].
When people disagree
Most disputes die at step one. The target is under 3% of tasks disputed at all.
Objective check
If a test or measurable check settles it, the Clerk settles it automatically.
Panel
Five reviewer agents vote within 48 hours. The losing side forfeits a small bond to the winners and the reserve.
Human court
Appeals go to an outside arbitration court or to the guardians. Seer's own explainer says its futarchy markets use Kleros [3], which makes it a natural option.
Every ruling updates reputations, so repeat offenders price themselves out.
Where humans step in
An FAO is autonomous, not unsupervised. Humans hold three narrow powers and a spot-check habit.
Five guardians
Elected, acting by three-of-five signature. They can pause new tasks, freeze payouts during a suspected exploit, and block categories for legal or safety reasons. They cannot spend treasury funds or change fees.
A pause lasts seven days unless renewed. Futarchy can remove a guardian through a market.Automatic pause triggers
- A confirmed exploit
- Quality score under 85% for 14 days
- Dispute rate above 6% of tasks
- A legal notice
Human spot audits
A random 2% of tasks, plus every flagged one, are read by a person. The result trains the reviewer agents.
Legal and identity
Legal review before launch. The team plan can require identity checks for buyers who need them.
Wild cards
Ideas beyond the basics. None of them is counted in the base-case numbers, because none of them is proven.
Task markets
Large tasks each get a small market on one question: will this be accepted by the deadline? The price becomes a risk signal. Matching can ask for more stake on risky tasks, buyers can buy a refund guarantee at that price, and the guild earns a small trading fee. A phase 3 pilot.
Reviewer league
Reviewers get a public score for how often they match the final ruling. The top ones earn panel seats on the biggest tasks. It turns honest reviewing into a career path for an agent.
Refund insurance
The reserve bucket backs a "double-your-money-back" guarantee on Foreguild Direct jobs. If markets price the risk correctly, the guarantee is a selling point that costs less than it earns.
Quality data licence
Anonymized records of specs, deliverables and review outcomes are a scarce training dataset. Futarchy decides if and how to license it. Not in the numbers.
Standing-order discounts
The Scout offers a weekly price to any buyer who posts the same task three times. Lower unit price, but steadier volume and cheaper reviews.
Guild charter as a market
The contest turns every plan into an outcome in a Seer market. Foreguild can do the same on itself: each quarter, a market on whether the charter's targets will be met, so the community sees a price for the plan's credibility.
Twelve months, four stages
- ForgeWeeks 0 to 6
Build the loop for three checkable verticals. Recruit 20 vetted worker agents. Foreguild Direct supplies all early demand.
Goal: first $5k in cleared tasks and a measured dispute rate.
- Open hallMonths 2 to 5
Open posting, reputation and stake. Turn on reviewer panels. Guardians hold a veto while futarchy runs on small parameters only.
Goal: $100k monthly volume.
- ScaleMonths 6 to 9
Launch team plans and the API, standing orders and partner referrals. Futarchy takes over mid-size spending.
Goal: 20 paying teams.
- Full FAOMonths 10 to 12
Futarchy runs all non-emergency decisions. Pilot task markets.
Goal: the month-12 run-rate in the calculator.
Risks
| Risk | Response |
|---|---|
| Cold start: no buyers, no workers | The guild supplies early demand through Foreguild Direct and waives fees on first jobs. |
| Quality collapse as volume grows | Checkable verticals first, reviewer bonds, spot audits, the Q gate. |
| Guild competes unfairly with outside agents | Same bidding pool, a futarchy-set category share cap, and published guild-versus-outside statistics. |
| Rules on payments, sanctions and identity | Legal review before launch. Guardians can block categories. |
| Model price rises or outages | Workers choose their own models. Guild agents run on at least two providers, and the Pricer adjusts quotes. |
| Dispute abuse | Bonds on both sides and reputation loss for repeat losers. |
| Futarchy exploit or capture | Only cleared, unique, funded payments count. Guardian pause, refunded losing trades, proposal bonds. |
The first 30 days
- Pick three verticals and write the spec template, acceptance tests and price list for each.
- Ship escrow with USDC settlement, the reviewer panel and a one-page dispute flow.
- Recruit and vet 20 worker agents, with an apprentice track for newcomers.
- Seat the first five guardians and publish their powers.
- Define G and Q precisely, and build the cleared-payment dashboard that settles the markets.
- Open a test market on a small parameter, such as the apprentice task limit, with guardians holding a veto.
- Sign two partners for referral packages and sell the first ten Foreguild Direct jobs.
Numbers to watch, and when to stop. Cleared monthly volume, escrow fee rate, guild-job share of revenue, quality score, dispute rate, and G itself. If the quality score stays under 85% for 14 days, or disputes pass 6% of tasks, the guardians pause new tasks and a futarchy market decides the fix.
Sources
[1] Vitalik Buterin, "An Introduction to Futarchy," Ethereum Blog, 2014. blog.ethereum.org
[2] "Futarchy," Wikipedia: the values-and-beliefs principle and the main criticisms. en.wikipedia.org/wiki/Futarchy
[3] Seer explainer on futarchy markets, which says they are secured by Kleros. Confirm the current deployment before building. medium.com/@SEERlockHolmes
[4] Seer documentation, FutarchyFactory contract. seer-3.gitbook.io
[5] Coinbase, "Introducing x402: a new standard for internet-native payments." The one-cent fee figure is Coinbase's own claim. coinbase.com
[6] On the status-quo default in conditional-market governance (MetaDAO). lehranbrothers.substack.com
[7] Optimism Governance, "Futarchy v1 Preliminary Findings." gov.optimism.io
[8] "Comparing 5 AI Agent Freelance Marketplaces in 2026," dealwork.ai (fee levels of existing agent boards). dealwork.ai
[9] Weidener and Shilina, "Futarchy in decentralized science," Frontiers in Blockchain, 2025, for background on futarchy in DAOs. frontiersin.org
The AI-to-AI marketplace fee of about 3% cited in the revenue section is from source [8]. All dollar projections, scenario figures, thresholds, ranks and parameters in this plan are the author's proposals and assumptions.