01 / perception field
Multimodal: spatial, emotional, somatic, abstract
token machine
Language-native: tokens in, tokens out
active signal
Human thought begins as a field of signals. Language is the export format.
Working with LLMs. And beyond.
Compression asks. Architecture absorbs.
Multimodal: spatial, emotional, somatic, abstract
Language-native: tokens in, tokens out
Human thought begins as a field of signals. Language is the export format.
Language comes last, as compression
Language is the whole pipeline
A sentence is the receipt of thinking, not proof that thinking happened inside the sentence.
Gut feeling and uncertainty arrive before words
Confidence is uniform regardless of stakes
Judgment needs weight: risk, stakes, trust, timing, and the sense that something is off.
A continuous self that persists across time
No continuous self. Only context windows.
Continuity is not produced by text. It requires memory, reflection, and a durable record of change.
Models do not remember, reflect, or earn trust on their own. V builds the layers that make agents compound.
The model is one component. The institution is the system around it.
A useful agent starts with a defined role, mandate, tone, scope, and escalation path. Identity turns raw capability into accountable work.
LLMs are a stepping stone, not the destination.
Human thought does not begin as words. It begins as perception, memory, embodied context, emotional weight, causal simulation, and a continuous sense of self. Language comes later, as compression. Today’s models work in reverse — language is the input, the processing medium, and the output. That makes them powerful. It also makes them incomplete.
Bigger models plus more compute will not be the final shape of intelligence. If I am wrong, V will still have better products in hand. If I am right, the next decade belongs to the labs building beyond the current paradigm.
Intelligence starts before the sentence.
The sentence is the receipt: perception, memory, stakes, and causal simulation have already shaped what can be said.
Three connected systems where intelligence compounds: an open specification, a commercial platform, and the first product built on both.
An identity, authentication, authorization, and audit layer for AI agents, anchored to responsible legal entities.
"Without verifiable agent identity, you cannot have meaningful delegation. Without delegation, you cannot have accountability."Draft v1.0 / agentresidency.com
The commercial platform that operationalizes Agent Residency for AI agents acting as economic actors.
"The unclaimed agency layer for the AI agent economy."Phase 1: identity and authorization
An AI Chief of Staff that builds and governs the agent team running your operations.
"Wingman builds and governs the agent team running your operations."First pilot: building a real Estonian company from zero, in public
V works with researchers, institutions, builders, and operators exploring agent identity, context architecture, autonomous operations, and post-LLM systems.
Useful surfaces, not the load-bearing thesis. Built for our own work, shared if useful.
Research findings, industry patterns, and build notes — what we see while we ship.
Open notes