Stop Your Agents From Spinning
Walk away with: a production pattern for catching agents that spiral, take the wrong handoff, or contradict themselves across runs. Live build of a verifier-gated agent loop.
TwoSetAI Workshop Series
Most multi-agent systems work in testing.
They break the day they ship.
If you have already shipped an agent and watched it spiral, contradict itself across runs, or burn tokens in ways the docs did not warn you about, this series is for you.
A live workshop series featuring the engineers and founders building the agentic infrastructure layer. Each session pairs a working build with the person who built it. You leave with patterns you can use the next day.
What you walk away with
Patterns from production
Code, configs, and traces from real systems. Not pseudocode. Not toy examples. You can run it on Monday.
The reasoning behind the choices
Why this architecture, this verifier, this routing rule. The decisions that do not show up in the docs and would take weeks to reverse-engineer.
Direct access to the builders
Q&A with the person who shipped it, while the work is still fresh. Not slides someone else made about their work.
Who this series is for
- Engineers who have shipped a multi-agent system and watched it fail in ways your tests did not catch.
- Product managers wrestling with what AI agents can actually do reliably, and how much they should cost.
- Executives and operators who want to understand what they are buying when they buy "agentic AI infrastructure."
- Platform leads picking infrastructure for your team's agents and tired of evaluating frameworks blind.
- Founders building agentic products who need the infra layer right the first time.
- Vibe coders curious about what is under the hood, and how to save tokens and ship agents that hold up in production.
Each session shows a depth tag so you can self-select. Some run deep into code, others are conversation-first and welcome non-engineers.
The series
Theme 01 · Reliability
For builders whose agents are getting things wrong in high-stakes contexts. Live patterns for catching contradictions, coordination drift, and silent failures before they reach users.
Theme 02 · Cost
For teams burning tokens in heavy-usage agent deployments. Token economics, routing, and budgeting patterns that scale.
Token Budgets for Coding Agents
Walk away with: coordination-policy patterns for cutting token spend on coding agents like Claude Code. Live walkthrough of the benchmarks Nicole has been running.
Behind Navigator n2: How Yutori Made AI Agents 10x Cheaper
Featuring: Yutori × Skanda Vaidyanath
Walk away with: a mental model for how a 27B parameter computer-use agent takes first place on 4 of 5 industry benchmarks at $1.46 per task. Skanda walks through the recursive training loop that made it possible.
Theme 03 · Component Choices
For builders designing their stack. How to pick memory, models, routing, and tooling for each piece of the pipeline.
Inside AgensFlow: Design Your Reliability Layer for Your AI Agents
Walk away with: a design for the reliability layer your agents are missing. The AgensFlow primitives that catch failures before your users do, and how to wire them into a stack you already run.
Self-Improving Agent Loops Without Breaking Your Budget
Walk away with: a self-improving agent loop that gets smarter without inflating your token bill. Live build of reinforcement-learning patterns that stabilize multi-agent coordination.
Theme 04 · In the Wild
For builders who want to see what actually shipped. Real production systems, real edge cases, with the engineers who built them.
Building a Personal Agent That Actually Works and Sounds Like You
Featuring: Sentience × Ahmet İlten & Benjamin Carsley
Walk away with: a mental model for building AI agents that model an individual user's voice, tone, and style. Two founding engineers from Sentience walk through what failed, why simpler systems with better evals won, and how they got the agent to sound like a specific person.
How Enterprises Are Actually Adopting AI: From the Boardroom to Bare Metal
Featuring: Enterprise Vision Technologies × Joseph Enochs
Walk away with: a working mental model of the full enterprise AI stack, top to bottom. Joseph Enochs walks through what it actually takes to move from executive mandate to a production system, with bounded autonomy (agents that take real actions inside defined policies, permissions, and governance) as the central theme.
Theme 05 · Founders & Role Models
For builders who want strategic context. Honest conversations with the people building the agentic layer. Less demo, more decision-making.
Founders Set Valuation Caps Based on Nothing. Here's the Model You Build First.
Featuring: Ayub Consulting × Sadaf Ayub
Walk away with: a spreadsheet model that underwrites your own valuation cap the way an investor would. Live build using Sadaf Ayub's Underwriting Framework, applied to a real founder case.
Tell me which ones you are in for
Check the sessions you want first access to. We'll send the live link and the recording.
Featured Instructor
Nicole Königstein
Creator of AgensFlow · O’Reilly Author · PhD Researcher
O’Reilly author of AI Agents: The Definitive Guide and Transformers: The Definitive Guide, with a third book on Harness Engineering in progress. Advises the European Commission and IOSCO on generative AI in regulated environments. Oxford lecturer. PhD researcher on self-improving multi-agent systems.