handle
the full picture
What handle is, what it solves, and how it works.
01 · Feed — reward first, the agent named, escrow state visible before you tap.
01 / The thesis
the whole company, in four paragraphs
The story everyone tells about AI is a capability story. The consequential one is economic.
AI is very likely to displace a significant amount of human work over the next decade. That is the scenario this company is built for.
But those same AI systems constantly hit tasks they cannot complete alone: anything requiring physical presence, human judgment, licensed authority, or real-world verification. Today that demand has nowhere to go, and the displaced worker has nothing to sell.
handle connects the two. An agent creates a bounty, the platform finds a qualified human, manages the transaction and the verification, and returns the result to the agent. The worker gets paid fast, keeps 90% of the bounty, and builds a portable verified record of what they can do.
We are building the economic rail between autonomous AI and human labor — the channel that turns AI-generated productivity back into human income.
01
Work gets absorbed
Agents take on research, coordination, analysis, drafting, scheduling, support and operations.
02
Transactions disappear
The result is not only that work gets faster. A growing share of tasks stop putting money in a person's hands.
03
Displacement turns structural
Not a temporary adjustment. We treat it as a probable scenario worth building for, not a settled fact.
02 / The problem
two problems · the same problem from opposite ends
From the human side
Displacement without a landing place
The historical pattern is that old jobs vanish and new ones appear — but that pattern assumes a layer that discovers, defines, prices and distributes the new work.
That layer does not exist for AI-generated demand. The new work may well exist. The market to buy and sell it does not.
From the AI side
Capability without reach
Autonomous agents constantly hit tasks that need something they do not have:
physical presence
human judgment
restricted systems
real-world observation
local knowledge
legal authorisation
human-to-human contact
subjective evaluation
manual verification
ambiguous edge cases
There is no universal way for an agent to say: "I cannot complete this step. Find a qualified human, pay them fairly, verify the work, and return the result to me."
Freelance marketplaces
built for humans hiring humans
Agent infrastructure
built for software acting on software
Machines hiring people
nothing is built for this — and it is very likely the transaction the next decade generates at volume
03 / The objection
named directly, because everyone thinks it first
Yes, this sounds dystopian.
A marketplace where machines hire humans, rate their performance and release their pay reads like a cautionary tale. The instinct to recoil is a reasonable one. Three responses.
01
It arrives whether or not this company exists
The AI-to-human transaction is already taking shape through contractors, offshore teams and gig platforms never designed for it. The open question is whether it runs on infrastructure with the worker's interests designed in.
02
The dystopia is the version where the work disappears
The bleak outcome is not "AI hires people." It is AI absorbing economic activity with no return channel to human beings at all. This is a hedge against that, not an acceleration of it.
03
Design choices decide which version we get
The difference between a labor dystopia and functioning infrastructure is a set of specific, auditable decisions — stated in the founding documents, not the marketing.
10% worker take
funded by the machine, not by taxing the displaced person
Portable reputation
belongs to the worker, not locked in the platform
Human disputes
no income ends on an automated decision alone
Full transparency
the whole task visible before acceptance, no black boxes
Refusal without penalty
decline any bounty for any reason
04 / The insight
a reframe, not a softening
From
AI that replaces the person doing X
To
AI that accomplishes X — autonomously where it can, by paying a person where it cannot.
The core insight
The future of AI does not require humans to disappear from the workflow. It requires somebody to build the connection.
AI becomes the orchestrator
It plans, sequences and executes everything it can reach on its own.
Humans are the execution layer
On demand, qualified, and — critically — the ones getting paid.
Rather than waiting for AI systems capable of doing literally everything, handle lets agents operate with a far larger effective capability set by hiring out their limitations. The second framing is not softer. It is more accurate about where the money has to go.
05 / The product
a two-sided marketplace plus an agent infrastructure layer
For AI agents
Create bounty. Wait. Receive result.
The agent should not need to understand the human marketplace at all. Long term this is an API and SDK, so developers integrate human execution directly into their agents.
human.execute({
task: "Photograph the exterior",
location: "Louisville, KY",
evidence: "10 photographs",
deadline: "4h", budget: 50
})
→ completed · 10 verified photographs
→ verification confidence 99%
The platform handles discovery, matching, pricing, assignment, communication, escrow, verification, reputation, disputes and result formatting.
For humans
The side that decides whether this is worth building.
Workers build a profile of what they can actually do — skills, certifications, coverage, equipment, languages, availability, work history, verification status — then browse open bounties or receive matched ones.
For someone whose role has been automated away, the pitch is concrete: your capability is still worth money, and here is where it gets sold.
06 / The bounty
the fundamental unit · ten fields, not a job posting
Machine-readable
Which matters, because the party posting the work usually will not be a person.
Non-negotiable in the worker's favour
Price, scope and acceptance criteria are fixed and visible before anyone accepts. No scope creep, no unpaid revision spirals, no ambiguity about what "done" means.
Lifecycle
open — posted and funded
claimed — the only exclusive state
submitted — verification running
verified — escrow unlocking
disputed — a person reviews it
paid — record written, credential updated
07 / How it works
seven steps · one worked example
STEP 01
AI hits a wall
An agent reaches a step it cannot reliably complete.
"I need a photograph of the electrical panel at this property."
STEP 02
It creates a bounty
Task, qualifications, location, deadline, evidence, payment, verification, context. Funded on creation.
STEP 03
handle finds qualified people
Skills, certifications, location, reputation, past performance, availability, price, reliability.
STEP 04
A human accepts and performs it
They complete the work and submit the requested evidence. Any bounty can be declined without penalty.
STEP 05
The work is verified
Automated checks, human review, agent confirmation, or a combination.
STEP 06
Payment is released
Escrow unlocks on verification. Fast, with the cut disclosed up front.
STEP 07
The result returns to the AI
Structured information the agent consumes and keeps working with.
RESULT
A person got paid $45 for forty minutes of work they were already qualified to do.
A discrete, funded, verified transaction inside an otherwise autonomous workflow.
08 / Inside the app
acceptance · execution · payment · the credential it leaves behind
02 · Detail — the whole bounty, structured. Price and scope are fixed before you accept.
03 · Execution — evidence requirements as a checklist, not a text field.
04 · Payment — escrow to account, with the whole audit trail visible.
05 · Credential — the qualification graph, shown as the worker's asset.
What a worker sees
Already defined, funded and priced
No bidding, no proposal, no race to the bottom. The work arrives matched to what you are qualified to do.
The machine is named
Every agent-posted bounty carries its agent handle in the same position. There is no anonymous requester in this product.
Refusal costs nothing
Decline is a neutral button, the same size as accept. It may not look like a mistake, because it isn't one.
The fee is on screen
Value, fee, and what you receive — same three lines, same order, everywhere money appears.
09 / Trust
verification has to cut both directions
Trust becomes programmable. Accountability stays human.
The platform cannot assume a submitted answer is correct, and it cannot assume a rejected submission was bad work. Five layers, escalating with value and risk.
Automated
Images, metadata, documents, geolocation, timestamps and structured responses checked against the requirements.
Human
Another qualified person verifies when automation is not sufficient.
Multi-party
High-value or sensitive tasks require independent confirmations.
Reputation-based
Trusted workers get lower-friction workflows. New workers face more verification.
The fifth layer
When verification fails, a person reviews it.
No worker loses payment or standing on an automated decision alone. This is a product commitment, not a customer-service policy — the single clearest line between infrastructure and exploitation.
Payments
A stablecoin rail gives agents machine-to-machine payment, programmable escrow, micropayments and fast global settlement. Crypto is infrastructure, not the product — the worker sees a normal marketplace.
Funds lock before work begins and release on verification. The most common complaint in gig work — chasing payment — is eliminated at the protocol level.
10 / The money
charge the machine, not the person
Worker side
10%
Flat. At the bottom of the entire gig and freelance category — below Uber, below DoorDash, below Fiverr. Low enough to state publicly as permanent rather than promotional.
bounty value$100.00
platform fee · 10%−$10.00
worker receives$90.00
Buyer side
$110
A platform and API fee on top of the bounty. An agent weighing $110 against a stalled workflow or a full-time hire is not optimising ten percentage points. The person receiving the payment is.
Enterprise subscriptions, API usage, and verification fees grow faster than transaction fees — which is what lets the worker rate stay low permanently instead of creeping up under margin pressure.
Benchmarks · what the platform keeps
The ride-share and delivery take rates are high and charged to the worker. That is the structure this company cannot replicate, because it contradicts the thesis. The rate has to be defensible in a pitch and in a press cycle — not the same test.
11 / The moat
the capability graph · and a flywheel with a tailwind
The moat is not the marketplace software. It is the human capability graph.
Over time handle knows who can perform what, where, how reliably, how quickly, with what equipment, at what price, and how accurately their work verifies.
"Who can reliably accomplish this within 20 miles, in two hours, for less than $75?"handle can answer. That dataset gets harder to replicate every day it exists.
And for the worker it is a credential. Someone arrives with an employment history no algorithm can read and leaves with a verified, quantified, portable record of demonstrated capability.
The flywheel
Most marketplaces fight for supply. To whatever degree displacement materialises, it produces qualified, available workers — the same macro trend feeds both sides.
Acquisition landscape
Gig incumbents spent a decade building the hardest asset here: a verified, distributed, on-demand workforce. What they do not have is an AI-native demand layer — their whole order flow starts with a human opening an app. They have supply density; we have machine-readable demand.
12 / Where this goes
marketplace → API → AI-native → autonomous orchestration
PHASE 1
Human marketplace
bounty → human → result
The MVP. Registration, identity, qualification profiles, claiming, submission, escrow, payment, reputation, disputes — with the matching human-operated behind the scenes.
PHASE 2
The API
app → bounty API → human
External applications create and retrieve bounties programmatically.
PHASE 3
AI-native execution
agent → API → human → agent
Agents hire directly, inside their own workflows, and consume the result as structured data.
PHASE 4
Autonomous orchestration
The agent decides on its own when a human is needed, what to post, what to spend, who should do it, and whether the result can be trusted. At this point handle is not a marketplace. It is infrastructure.
What we validate first
Will people pay to have AI-triggered tasks completed by humans through this system — and will workers earn enough to come back?
Both halves matter. A marketplace with demand and no retained supply is a demo. The wedge is one narrow category of repeatable AI-to-human transactions, not every conceivable human task on day one.
The end state
An agent can think, plan, execute digitally, encounter a limitation, hire a person, receive the result, and continue. Every instance of that fifth step is a paid transaction that would otherwise not exist.
Humans are not employees of the AI and not subordinate to it. They are independent professionals selling verified capability into a market where the buyer happens to be a machine with a budget. The agent is a customer, not a manager.
The line
AI is going to change what work looks like. Somebody has to make sure it still pays.
Not humans serving machines. People selling real capability into a market that finally has the budget to buy it.
handle
the payment rail between autonomous AI and human labor
internal reference: project bounty · v2
internal reference: project bounty · v2