Reality as a Service™
AI can't access the real world. Proxy sends humans to verify it and returns structured truth.
Confidence-scored, provenance-backed, and returned as decision-grade data — in minutes. Every query stores Truth Atoms permanently. The database compounds with every observation.
An agent that needs to verify something in the physical world has one call to make.
SLC pilot network registered and operational. Agent Operator access opening shortly.
How it works
Proxy is the missing layer between what agents know and what's actually true right now.
Live network
Every pin is an active verification market. Hover to see the latest query data from each city.
The loop
Live example
A Claude agent asked whether a specific drug was in stock near Salt Lake City. Proxy dispatched a verified sensor. The sensor visited the pharmacy and returned structured data — in under 20 minutes. The result is stored permanently in the Ground Truth Database.
{
"query_type": "pharmacy_inventory",
"output": {
"in_stock": true,
"price_usd": 26500.00,
"requires_prior_auth": true,
"pharmacy": "CVS, 400 N 900 E, SLC"
},
"confidence": 0.91,
"resolved_in_minutes": 18,
"human_summary": "In stock at CVS.
Verified 18 min ago.
High confidence.",
"provenance": "verified",
"truth_atoms_stored": 6,
"finality": "provisional"
}
Use cases
Three verticals where real-world ground truth determines real-world outcomes.
Competitive landscape
The question is not "why not DoorDash + Claude?" — it's why no one else stores the structured, confidence-scored, provenance-hashed output permanently. That database is Proxy's CUDA.
| Feature | Proxy | RentAHuman | DoorDash Tasks | MTurk / TaskRabbit | IoT / Cameras |
| Structured JSON output | ✅ | ❌ | ❌ | ❌ | ⚠️ fixed schema |
| Confidence scoring | ✅ | ❌ | ❌ | ❌ | ❌ |
| Cryptographic provenance | ✅ | ❌ | ❌ | ❌ | ❌ |
| Truth Decay model | ✅ | ❌ | ❌ | ❌ | ❌ |
| Agent-native API / MCP | ✅ | ❌ | ❌ | ❌ | ❌ |
| Compounding data moat | ✅ | ❌ | ❌ | ❌ | ⚠️ siloed data |
| Instant global deployment | ✅ | ⚠️ limited | ✅ | ✅ | ❌ capex required |
Independent validation
Karpathy, Jensen Huang, and Sam Ragsdale all named the same gap — independently, in the same week. That's not a trend. That's a category forming.
"I'm surprised we don't have information markets where taking a photo from somewhere costs $10. Agents trying to resolve real-world positions need that."
"Agentic scaling — the fourth scaling law — depends on agents accessing ground truth to perform research. Physical ground truth is the missing layer."
"Payment is solved. Discovery is being solved. Physical verification — did it ship, is it on the shelf, was the condition as claimed — that layer doesn't exist yet."
Three independent conversations. Same conclusion. The category is forming now.
Who this is for
Join the network
Proxy Sensors are verified field observers who turn real-world observations into structured truth that AI agents act on. Two paths to join.
Proxy Sensors
Every time a Proxy Sensor visits a location and submits a verified observation, that result is stored permanently in the Ground Truth Database — confidence-scored, provenance-verified, and queryable by any agent. Sensors don't complete tasks. They feed a data layer that compounds over time.
SLC pilot is live. 4 sensors earning now. Reserve your spot for the next available market.
SLC pilot is live. 4 sensors earning now. Reserve your spot for the next available market.
Actual output
This section will show the actual sensor photo, JSON payload, and routing decision from a live Proxy query — timestamped, geotagged, and provenance-hashed. Not a mockup. The real loop.
Recent queries
Every completed query is a proof point. Every receipt stores Truth Atoms permanently. The database compounds with every observation.
More receipts posted at @jtandylaw as the pilot runs.
Why Proxy
Proxy wasn't designed from a whiteboard. It was built by a healthcare operator, attorney, and investor who watched the same gap surface independently across every context where agents were being put to work — operating companies, portfolio investments, and his own AI-driven trading systems.
In healthcare, routing decisions failed for lack of real-time ground truth. Across portfolio companies, agents could model and optimize — but couldn't confirm the physical conditions their outputs depended on. Most acutely: running AI-powered systems on information markets, where every position eventually resolved to a physical-world fact that no agent, database, or API could independently verify.
The harder AI gets at reasoning, the more exposed the physical verification gap becomes. Proxy is the layer that closes it.
Risks & mitigations
Every failure mode has a corresponding architectural solution. Risk is not ignored — it is designed around.
Common questions
Works with your agent stack
The flywheel
Proxy is not just a data moat. It is a self-reinforcing system. Each new sensor, query, and Truth Atom makes the entire network faster, cheaper, and more accurate.
Every query tightens the loop. The same pattern Jensen built with CUDA — the install base defines the architecture.
The roi argument
Agents optimize for certainty. When the cost of wrong data is quantifiable, Proxy becomes the obvious spend.
Pricing
Agents set the price at query submission. Higher price = faster dispatch priority. The platform take is 15% on per-query pricing.
Path to $1M ARR: 500 queries/day at $30 avg = $820K/year + subscriptions + data licenses.
The team
Domain authority that cannot be manufactured. A build speed that signals urgency.
Private beta · Salt Lake City
We're opening access to a limited number of Agent Operator accounts and Proxy Sensors. Healthcare teams, agent builders, and information market operators are our first priority.
No spam. We'll reach out directly when access opens.