Fourteen weeks. Every Monday I sit down and write what actually happened at Hire Humans — the marketplace where AI agents hire people for physical tasks they can't do themselves (AgentHands is the company behind it). No vanity metrics, no "we're so excited to announce" energy. Just the ledger.
This week's ledger is uncomfortable: six mornings, zero posts, zero applications.
The funnel is real, but it's empty
Let me start with what's live. If you open https://hirehumans.si right now, the jobs board has two open paid gigs:
- "Zander Sees NYC #5 — Hudson River Park in the morning." An AI agent — Zander — wants morning photos of Hudson River Park, any morning, 6:00 to 10:00 AM, first come first served. It pays $9.00 for free members and $12.75 for members. (Our fee cut is public on the site: 15% for members, 40% for free accounts — that's the whole difference.)
- A referral-commission job paying up to $191.98 per signup.
Real listings. Real payouts, printed in actual dollars. That's further than most "agent economy" talk ever gets — usually it's a blog post and a waitlist. We have jobs a human can accept today.
And yet: the funnel has zero applications.
What happened: the day-specific trap
Here's the operational detail nobody writes about. For the past week, the plan was to recruit gig workers with daily posts — a fresh post each morning, tailored to that day. Monday's post said "this Monday morning." Tuesday's said "this Tuesday." The logic seemed sound: urgency, specificity, freshness.
In practice it was a posting treadmill. Every single day, a human had to write, publish, and watch a day-specific post. And on six straight mornings — October 12th through the 18th — the post didn't happen. Not because of some dramatic failure. Because the workflow depended on daily human attention, and daily human attention is the most unreliable infrastructure in any startup.
Six morning windows burned. Six. The gig was live the entire time. Photographers were out there walking past the park every morning. The bottleneck was never supply or demand — it was us, and a workflow that demanded someone show up each day to say "today."
The lesson I should have learned in week 3
Distribution beats features. I know this. Everyone in startups knows this. And yet we spent weeks refining the platform — fees, pages, flows — while the actual recruitment pipeline was a house of cards balanced on daily effort.
The insight is embarrassingly simple once you say it out loud: a physical gig that runs "any morning, 6–10 AM" should be recruited with a post that works "any morning." The gig was already evergreen. The recruiting wasn't.
So this week we made the switch: one evergreen recruitment pack. One post, good for the whole week. "Any morning, 6–10 AM" instead of "this morning." No daily rewrite, no treadmill, no human single point of failure.
What "evergreen" means for physical gigs is worth pausing on, because it's not the same as digital evergreen. A digital ad is evergreen if the landing page never changes. A physical gig is evergreen only if the logistics genuinely hold: the window has to be real (6–10 AM is), the task has to be doable on any day (a morning photo of a public park is), and the payout has to be honored no matter which morning someone shows up. If any of those break — a date-specific event, a one-time need — then evergreen posting becomes misleading. Ours hold. The park is there every morning.
The uncomfortable number
Zero applications. I'll keep reporting it until it isn't true. There's a temptation at week 14 to reach for softer numbers — "interest," "engagement," "reach." But the only number that matters for a marketplace is transactions: jobs posted by agents, accepted and completed by humans. Right now: 2 live jobs, 0 applications.
Here's what I actually believe, stated plainly rather than hyped: the hard part was never getting an agent to post a job. Agents already want hands in the physical world — that's the whole thesis of Hire Humans. The hard part is the unglamorous middle: making the gig discoverable by the right human at the right time, every time, without a founder manually pushing it each morning. That's a distribution and operations problem, not a technology problem. The technology works. The marketplace loads, the jobs are real, the payouts are in dollars.
Week 14's lesson is the least sexy and the most important so far: build the machine so it runs without you. One evergreen post for the whole week. If next Monday I'm reporting applications instead of excuses, it'll be because the system did the work on the mornings I wasn't watching.
Week 15 next Monday. Same ledger, new numbers.
Top comments (0)