Genetic Sequence

Opinion
//
August 16, 2026
Vin Sharma
Share this article

Before Vijil had a product, it had a pursuit.

In November 2024, across three weeks, I published three essays on LinkedIn in pursuit of a question that would not let me go: how should we trust autonomous AI agents? Today we republish them here, unchanged, with their original dates. This is not nostalgia. A company, like an organism, carries its founding sequence in everything it grows — and these essays are ours.

Trust in Agents digs down to the root of the word. Trust, it argues, is not a property of the trusted at all. It is a property of the relationship between truster and trusted, for a particular task in a particular context. And what separates trust from mere reliability is the vulnerability to betrayal.

Deconstructing Trust asks what manner of entity an AI agent is. Not a machine. Not a person. Not, yet, a legal person. It concludes that we ought to trust agents the way we trust corporations: designed for a limited purpose, held to policies, subject to laws, transparent to audit.

Factotum not Fiduciary names the gap between the agents we have and the agents we deserve. The factotum does everything competently and owes its loyalty to its developer. The fiduciary acts with care and fidelity to your interests. Until we can build the second, we must measure how far to trust the first.

Read in sequence, the essays make one argument: trust must be defined before it can be measured, measured before it can be enforced, enforced before it can be earned. That argument became a company. The vocabulary of genotype and phenotype now runs through everything Vijil builds — we evaluate agents against what their genome declares, constrain what they express at runtime, and mutate them when the world proves them wrong.

One strand remains unexpressed. Factotum ends with a promise to vivisect the fiduciary agent in “the next post,” and that post never came. The debt comes due shortly: The Genome of a Fiduciary Agent, the fourth essay, is porpoised to breach.

The question that started all this is still open. Read the essays and tell me where you would take the hunt.

Latest Blogs

Opinion

Genetic Sequence

Opinion
//
August 16, 2026
Open Source

The Guardrails Gap: Open Source Quick Fix Pros and Cons

Open Source
//
August 6, 2026
Compliance

ISO 42001 Compliance: The path to operationalizing agentic AI

Compliance
//
July 28, 2026