Your AI Agent Is Probabilistic. Its Decisions Don't Have to Be.
Your AI agent can draft the email. Should it be allowed to change where a customer's payments go?
Lower the temperature, repeat the prompt, add more tests. The answers get more consistent, but none of it proves an action is authorized or safe.
And an agent acting on a confident guess can move money, change records or touch the wrong account before anyone notices. This workshop is about what closes that gap.
Daniel Davis is co-founder of TrustGraph, an open-source platform that checks every action an AI agent proposes against business rules and evidence before it runs. His team's guide on what determinism means for AI agents is the starting point for this session.
Daniel starts with what determinism has always meant in software and in safety-critical systems like aircraft: the same input gives the same result, and when something deviates, the system catches it and stops safely. Then he shows why that breaks with language models, and what replaces it: not identical outputs, but a predictable decision about whether the agent may act. He walks through the six things to check before any action (authority, scope, risk, time, uncertainty and the value of acting) and the three possible answers, allow, pause or deny, using one example: an agent asked by email to change a customer's payment account.
You'll walk away with:
A definition of determinism for AI agents you can use with your team: predictable decisions, not identical words
Six questions to answer before an agent is allowed to act
When an agent should go ahead, stop and ask, or refuse, and how to make that call the same way every time
A way to record why each action was allowed, so you can show it later
Come if you are:
An engineer building agents that take real actions: payments, tickets, data changes
On a platform, security or risk team deciding what agents may do
A PM or founder putting agents in front of customers
Anyone who has been asked "how do you know the agent won't do something it shouldn't?"
Want the backstory? Watch TrustGraph co-founder Daniel Davis on the TwoSetAI podcast.
Format: 60 minutes, live on YouTube. ~40 min technical deep-dive, ~15 min live Q&A.
About the series
Part of the TwoSetAI Workshop series.
More sessions:
July 17: Why multi-agent systems fail
July 31: Inside AgensFlow
August 14: Token Budgets for Coding Agents
August 28: Self-Improving Agent Loops
September 18: Building A Personal Agent That Actually Works and Sounds Like You
September 25: Behind Navigator n2: The Recursive Training Loop That Made a 27B Model Beat GPT-5.6 and Claude
October 2: How Enterprises Are Actually Adopting AI: From the Boardroom to Bare Metal
October 16: One Person, Many Voices: How to Build Agents That Sound Like a Real Person
October 30: First-Time Founders - Are You Raising? Set a Valuation Cap You Can Actually Defend.
November 13: How Many AI Agents Are Running in Your Company?
December 4: Anyone Can Build. Not Everyone Gets Heard.
December 11: Your AI Agent Is Probabilistic. Its Decisions Don't Have to Be.
See all upcoming sessions: https://twosetai.com/lab/workshops/
About the instructor
Daniel Davis is a technology leader with deep expertise in AI, knowledge graphs, and semantic technologies. He co-founded TrustGraph to democratize access to enterprise-grade context graphs, enabling developers and organizations to build intelligent AI agents with precision-grounded context.
About the host
Angelina Yang runs TwoSetAI and hosts TwoSetAI and Decoding AI Builders, long-form interview platforms where she's interviewed 100+ AI founders shaping this era. fast.ai fellow. 2024 Anthropic Build Tournament winner. Curates the workshop series and connects the speakers with builders who care.
About TrustGraph
TrustGraph builds ontology-grounded context graphs and governs AI agent actions with a deterministic runtime: every proposed action is allowed, paused or denied, with the evidence and policy version behind it recorded. Open source under Apache 2.0 and built on W3C standards (RDF, OWL, SHACL, PROV-O), with 2,800+ stars on GitHub.
Hosted by TwoSetAI.
Past events replay
July 17: Stop Your Agents From Spinning, with Nicole Königstein
July 31: Inside AgensFlow: Design Your Reliability Layer for Your AI Agents, with Nicole Königstein
August 14: Token Budgets for Coding Agents, with Nicole Königstein
August 28: Self-Improving Agent Loops Without Breaking Your Budget, with Nicole Königstein
September 18: Building A Personal Agent That Actually Works and Sounds Like You, with Ahmet İlten of The Sentience Company
September 25: Behind Navigator n2: How did Yutori make AI agents 10x cheaper? With Skanda Vaidyanath of Yutori