Antoni Watts
My projects & ideas.
I'm a builder, endlessly curious — testing and building ideas across domains, patenting the genuinely new.
Trust is at the heart of every business — I make it mechanical.
Work
What I'm building.
Execution Protocol
Every AI action. Authorized. Executed. Or stopped. With proof.
One structured message in, one terminal outcome out — executed, instructed, or blocked. The boundary sits outside the model, and every outcome ships a signed receipt you can verify offline.
- the agent only proposes
- boundary outside the model
- receipts verify offline
is this exact action allowed? — deterministic
the agent proposes · authority decides · the proof outlives the action
Gutencode
The hard backend domains — done for you, and proven.
Auth, billing, RBAC, tenancy, an agent runtime — an agent-ready backend spine, printed identical in Python, Go, and Node. Clone it, run the verifier, watch it go green, build on top.
- 35+ domains
- Python · Go · Node — identical
- you can't fake GREEN
$ git clone … && python verify.py
tests python ✓ · go ✓ · node ✓
parity python == go == node
durable state survives restart · errors identical ×3
==== VERIFY: GREEN ====
Ideas
The overflow of the backlog, thought out loud — notes and essays across AI, systems, and science. The best of it graduates into whitepapers.
13 patents filed
Authorization — “who says the agent can do that?”
Reliability — “does the agent actually work?”
Accountability — “what did the agent actually do?”
Writing
Ideas, written down.
Give Your Agent a Card, Not Your Wallet
The authorization chain confuses everyone — until you notice you already trust one every time you tap to pay.
The Authorization Chain
Authorization is not a check. It is a chain.
The Action Layer
Models generate. Agents decide. Protocols route. None of them control what is allowed to happen.