02 Applied AI
Agents that ask
before they act.
Working AI agents for real teams.
A person accountable for the result.
Why “Agency”
The ability to act.
Through SINT Labs I build agents that do real work for client teams: outreach and marketing, and assistants for wealth managers, funds and other professional firms. Each one gets a clear job, the right tools, and a rule for when to stop and ask.
The builder doesn’t decide what ships.
This is the loop I run every day.
Step through it, from the question to the release.
Turn the idea into a spec that can fail.
- Receives
- A question, the people it affects, and what would count as success.
- Returns
- A locked spec: what to build, and tests that say when it’s done.
- Human judgment
- I write this part myself. It’s where the product gets decided.
Build from the spec, nothing else.
- Receives
- The locked spec and a controlled workspace.
- Returns
- Working software and notes on what was done.
- Human judgment
- I set the architecture and review the code.
Grade the work against the spec.
- Receives
- The build, the tests, and no knowledge of how it was made.
- Returns
- A pass or fail for each test, with the evidence.
- Human judgment
- A fail goes back to step two. Nobody argues with the test.
Ship it or hold it.
- Receives
- A build that passed, and the evidence behind it.
- Returns
- A release, or a reason to wait.
- Human judgment
- Publishing is always a person’s call.
The practice
Design the responsibility.
Define the job, what success looks like, and where the agent should stop.
Set the limits first.
What it may spend, what it may touch, and what needs a human. Decided before it runs, not after.
Keep the receipts.
Every action leaves a record someone can check later.
A little clarity
What do the agents actually do?
Defined jobs with approved tools: finding and contacting the right people, preparing material, answering questions from a firm’s own documents. They pass their work to a person before anything consequential happens.
What stays human?
People set the goal, approve access and money, and accept the result. The rules for that come from SINT Protocol.
The principle
Autonomy should earn trust.
One useful task at a time.
Start with the work.