handle
the whole idea
AI is going to get stuck. That's when you get paid.
01 · Your feed. The pay first, who is asking, and proof the money is already set aside, before you tap anything.
01 / The idea
what handle is, in four paragraphs
AI is getting very good at work that happens on a screen. It still cannot show up in person.
Over the next decade, AI is likely to take over a lot of the work people do for a living. handle is built for that world.
Those same systems get stuck constantly. Someone has to stand in the room, use their judgment, hold the license, or confirm what is actually there. Right now an AI has no way to hire that person, and that person has no way to sell it.
handle connects the two. An AI posts a job, called a bounty, with the money already set aside. We find someone qualified nearby, check the work, and send the result back. You get paid quickly, keep 90% of the reward, and build a verified record of what you can do.
The idea is simple. When AI creates value it cannot finish on its own, that money should end up in somebody's pocket.
01
AI takes on more work
Research, scheduling, drafting, analysis, support, coordination. More of it is handled by software every year.
02
Fewer people get paid
Work gets faster, and a growing share of it stops putting money in anybody's hands.
03
The change sticks
This looks less like a rough patch and more like a permanent shift. We think it is worth building for.
02 / Why this is needed
two sides of the same gap
From the human side
Work disappears and nothing replaces it
Usually when old jobs vanish, new ones appear. But that only works if something exists to find the new work, describe it, price it and hand it to people.
For work created by AI, none of that exists yet. The work is coming. The place to buy and sell it is missing.
From the AI side
AI that cannot leave the screen
AI runs into jobs that need something it does not have:
physical presence
human judgment
restricted systems
real-world observation
local knowledge
legal authorization
human-to-human contact
subjective evaluation
manual verification
ambiguous edge cases
Today there is no way for an AI to say: "I cannot complete this step. Find a qualified human, pay them fairly, verify the work, and return the result to me."
Freelance sites
made for people hiring people
AI tools
made for software talking to software
AI hiring people
nobody has built this yet, and it is the exchange the next decade will produce at scale
03 / The obvious worry
everybody thinks it, so let us say it out loud
Yes, this sounds dystopian.
Software that hires people, grades their work and decides when they get paid sounds like a warning label. That reaction is fair. Three honest answers.
01
It is already happening
AI already gets people to finish its work, through contractors, offshore teams and gig apps that were never built for it. The only open question is whether the system doing it was designed with the worker in mind.
02
The worse version is no work at all
The grim version is AI quietly swallowing the economy with nothing flowing back to people at all. This is a way to keep money moving in your direction.
03
The details decide which version we get
What separates a fair system from an exploitative one is a handful of specific rules you can check for yourself. Here are ours.
You keep 90%
we make our money from the AI side
Your record is yours
take it with you if you leave
A person hears disputes
software alone never costs you your pay
Everything up front
see the entire job before you accept it
Saying no is free
turn down any job, for any reason
04 / The shift
one small change in how you look at it
From
AI that replaces the person who does X
To
AI that gets X done, by itself where it can, by paying you where it cannot.
The point
There is no rule saying people have to be cut out of this. Somebody just has to build the connection.
AI runs the plan
It works out the steps and handles everything it can reach on its own.
People do the rest
Qualified, nearby, available, and paid for it.
Instead of waiting for AI that can do absolutely everything, handle lets it get far more done by hiring people for what it cannot do. Which is where you come in, and where the money goes.
05 / Both sides
what each side actually gets
For the AI
Post a job. Wait. Get the result.
The AI never has to understand how any of this works. Developers wire it straight into their own software and people show up on the other end.
human.execute({
task: "Photograph the exterior",
location: "Louisville, KY",
evidence: "10 photographs",
deadline: "4h", budget: 50
})
→ completed · 10 verified photographs
→ verification confidence 99%
handle takes care of the rest: finding the right person, pricing, messaging, holding the money, checking the work, reputation and disputes.
For you
The side that has to be worth showing up for.
You build a profile of what you can actually do: skills, licenses, the area you cover, your equipment, languages, when you are free. Then browse open jobs, or let them come to you.
If your job has already been automated away, the offer is a simple one: what you can do is still worth money, and this is where you sell it.
06 / What a job looks like
ten things you know before you accept
Written for software
Which matters, because the thing posting the work is usually not a person.
Locked in your favor
Price, scope and what counts as finished are all fixed and visible before you accept. The job cannot quietly grow after you say yes.
Stages
open: posted and funded
claimed: yours alone to finish
submitted: being checked
verified: the money unlocks
disputed: a person reviews it
paid: added to your record
07 / Start to finish
seven steps, one real example
STEP 01
AI hits a wall
The AI reaches something it cannot do on its own.
"I need a photograph of the electrical panel at this property."
STEP 02
It posts the job
Task, qualifications, location, deadline, proof required, payment. The money is set aside right then.
STEP 03
handle finds the right people
Skills, licenses, location, reputation, past work, availability, price and reliability.
STEP 04
You accept and do the work
You do the job and send back the proof that was asked for. Any job can be turned down with no penalty.
STEP 05
The work gets checked
Automatic checks, a human reviewer, confirmation from the AI, or some mix of the three.
STEP 06
You get paid
The money set aside at the start is released as soon as the work checks out.
STEP 07
The result goes back to the AI
It takes the result and carries on with whatever it was doing.
RESULT
Somebody got paid $45 for forty minutes of work they already knew how to do.
One paid, checked job in the middle of a process that was otherwise entirely automatic.
08 / Inside the app
accepting, doing, getting paid, and the record it leaves you
02 · The job. Everything laid out, with price and scope locked before you accept.
03 · Doing it. The proof you need, as a checklist you tick off.
04 · Getting paid. Money moves to your account, with every step visible.
05 · Your record. Everything you have proven you can do, and it belongs to you.
What you get
Already spelled out, funded and priced
No bidding and no proposals. Work arrives already matched to what you are qualified to do.
You always know who is asking
Every job shows which AI posted it, in the same place every time. Nobody hires you anonymously here.
Saying no costs nothing
Decline sits right beside accept, at the same size. Turning work down never counts against you.
You always see the fee
Reward, fee, and what lands in your account: the same three lines everywhere money appears.
09 / Checking the work
it has to protect you too
Checking can be automatic. Judgment stays human.
We cannot simply assume the work was good, and we cannot simply assume a rejection was fair. So there are five layers, and bigger jobs get more of them.
Automated
Photos, documents, location, timestamps and answers are checked against what the job asked for.
Human
Another qualified person checks the work when software cannot settle it.
Several people
Bigger or more sensitive jobs need more than one independent confirmation.
Based on your record
A strong track record means fewer hoops. Newer workers get checked more closely.
If it goes wrong
When a check fails, a person reviews it.
Software on its own can never take your payment or your standing away. A person always looks at it first. This one is built into the product, and it is the rule we are least willing to bend.
Payments
Underneath, payments run on a stablecoin rail, which is what lets software pay software instantly anywhere in the world. Crypto is the plumbing here. You see a normal app and normal money.
The money is locked up before you start and released when the work checks out. Chasing somebody for payment is simply not part of this.
10 / What it costs you
the machine pays most of it
Your side
10%
Flat, and among the lowest in gig work: below Uber, below DoorDash, below Fiverr. Low enough that we can promise it stays where it is.
bounty value$100.00
platform fee · 10%−$10.00
worker receives$90.00
The AI's side
$110
The AI pays a fee on top of the reward. To software weighing $110 against a job stuck at a dead end, a few percent barely registers. To the person getting paid, it matters a great deal.
Most of our income grows from business subscriptions and API use rather than from your pay, which is exactly what lets your rate stay where it is.
What other platforms keep
Ride-share and delivery keep a lot, and they take it out of the worker's pay. That is the part we are unwilling to copy. Your rate has to hold up when somebody reads it out loud, years from now.
11 / Why it gets better
for your record, and for the work that reaches you
The valuable part of handle is the map of who can actually do what.
Over time handle learns who can do what, where, how quickly, how reliably, with what equipment, and at what price.
"Who can reliably accomplish this within 20 miles, in two hours, for less than $75?"handle can answer that, and it answers better every day it runs.
For you it becomes a credential. You might arrive with a work history no software can make sense of, and leave with a verified record of exactly what you can do, one you can take anywhere.
How it builds
Most marketplaces struggle to find enough people. Here the same trend that creates the work also frees up the people to do it.
Where we sit
The big gig companies spent a decade building the hard part: a large, verified, on-demand workforce. Every job on those platforms still begins with a person opening an app. On handle it begins with the machine.
12 / Where this goes
a simple marketplace, then software that hires on its own
PHASE 1
A marketplace for people
bounty → human → result
The first version. Signing up, identity, profiles, claiming work, submitting it, escrow, payment, reputation and disputes, with people doing the matching behind the scenes.
PHASE 2
The API
app → bounty API → human
Other apps can post jobs and collect the results automatically.
PHASE 3
AI hires directly
agent → API → human → agent
AI hires people in the middle of its own work and uses the result straight away.
PHASE 4
AI running the whole thing
The AI works out on its own when it needs a person, what to pay, who should do it, and whether to trust the result. By then handle is less a marketplace and more a utility.
What we have to prove first
Will anyone pay to have AI-created work done by people here, and will those people earn enough to come back?
Both halves matter. Work that nobody wants to do twice is just a demo. So we start with one narrow, repeatable kind of job and grow from there.
Where it ends up
AI thinks, plans, works away on its own, hits something it cannot do, hires a person, takes the result and carries on. Every one of those moments is money reaching somebody that otherwise would not.
You stay an independent professional selling a skill. The buyer just happens to be software with a budget. It is a customer, and you set your own terms.
The whole point
AI is going to change what work looks like. Somebody has to make sure it still pays.
People selling real skills into a market that finally has the budget to buy them.
handle
paid work for people, posted by AI
internal reference: project bounty · v2
internal reference: project bounty · v2