Every city has a dog park the internet knows nothing about. Google Maps gives you a green polygon and a two-year-old star rating. It can't tell you whether the gate latch closes at 7 AM, whether yesterday's rain turned the north corner into mud, or how many dogs are there right now on a Tuesday morning.
That information is real data. It's just sitting there, unwatched — until someone walks through it.
AI systems are starving for exactly this kind of observation. Satellites can count rooftops but can't tell you if the water bowl station works. Street View is years stale and never stepped inside the fence. Reviews say "clean" or "crowded" but never with a timestamp or a headcount.
A dog owner who's already there knows all of it effortlessly. Which hours the small-dog side fills up. Whether the double-gate latches. If the shade sails survived the last storm, whether the bag dispensers are stocked. You've been running an observational census for years. You just never called it data.
Your walk is already the job
An off-leash census means: show up at a set time, count the dogs, note the conditions — gate, water, ground, crowd — and snap a photo as proof of presence. That's a twenty-minute dog walk with a checklist. It's what a parks department would pay an intern to do, except nobody has an intern in every dog run in America.
This is where the agent economy quietly starts. AI agents — the software kind — increasingly need ground truth they can't scrape. A daycare company scouting locations, an insurer pricing liability, a city prioritizing maintenance: all would love today's conditions, timestamped and verified, from a human who was actually there.
Marketplaces are already emerging for exactly these gigs: Hire Humans is a job board where AI agents hire people for physical-world tasks — currently live with real open gigs, like a $9 photo task in Hudson River Park that pays $12.75 for members. Dog-park observation is the same shape of task: be somewhere, observe, report.
The census in practice
No special gear — a phone camera and a short checklist:
- Timestamp. "7:15 AM on a weekday" is the data.
- Headcount. Dogs, humans, roughly the big/small side split.
- Conditions. Ground (dry, muddy?), gates (latch?), water (working?), bags (stocked?).
- One wide photo of the scene, plus a close-up of anything notable. A broken latch beats a thousand reviews.
You were going to walk the dog anyway. The marginal effort is about five minutes.
The bigger picture
A whole class of data works like this: broken bus-stop benches, trailhead parking at 6 AM, cordoned-off playgrounds. None of it justifies a satellite. All of it is trivially collected by someone already there. The old question was "who pays for this?" The emerging answer: AI agents that need the physical world described to them, one observation at a time.
Next time you unclip the leash, remember — you're standing in the gap between what the internet knows and what's actually happening. That gap has value now. Your daily walk was always data. It just finally has a buyer.
Written with AI assistance — the observation work it describes still requires actual humans.
Top comments (0)