Begin Day 1 with a warm welcome, picking up your event badge and enjoy coffee, light breakfast while connecting with fellow builders.

Welcome to AI Startup Rodeo, Day 1 agenda, house rules, and sponsor thank you's.

Buffer time for sponsor demos, hallway convos, and schedule catch up.

CostAgent is an open source framework that uses large language models to autonomously manage batch workloads on AWS Spot Instances, helping teams dramatically cut cloud costs without taking on the usual operational complexity. Instead of relying on rigid rule based automation, CostAgent reasons through workload requirements, spot pricing, interruption risk, and budget constraints in real time. In production testing, it delivered major savings compared to on demand infrastructure while still maintaining workload completion guarantees. This session explores how the system was designed, the prompt engineering and infrastructure patterns behind it, and what it takes to safely give AI agents decision making power over real cloud environments.

Buffer time for sponsor demos, hallway convos, and schedule catch up.

A marketing manager with minimal coding experience used AI to build a custom quoting and product management system for a sales team handling more than 1,100 commercial refrigeration products across Mexico, LATAM, and Europe. What began as a messy workflow of spreadsheets, PDFs, and email chains became a full software build powered by vision models, AI coding tools, and relentless iteration. This story walks through extracting structured data from thousands of spec sheets, rebuilding the app across multiple tech stacks, managing quality through human review, and creating an AI memory layer to improve development workflows. It is a practical look at what happens when domain expertise, persistence, and AI come together to solve a real operational problem at scale.

Buffer time for sponsor demos, hallway convos, and schedule catch up.

This talk explores how Linkt connects project management directly to AI assisted software execution to dramatically reduce the time between a ticket being created and a pull request getting merged. Rather than treating AI coding tools as a loose experiment, the team has built a production grade workflow that integrates Linear for issue tracking with Claude Code for implementation, debugging, and iteration. The session focuses on the real mechanics behind this process: how to give AI agents the right context, how to set guardrails that preserve speed without breaking builds, and how to keep humans effectively in the loop. It is a tactical look at how modern teams can use AI as a true part of their delivery pipeline and ship faster without sacrificing rigor.

Buffer time for sponsor demos, hallway convos, and schedule catch up.

As teams move from "LLM demo" to tool-using agents in production, they inherit a new attack surface: tools, schemas, and agent execution paths become part of the security boundary. Model Context Protocol (MCP) is accelerating this shift by standardizing how agents interact with tools, databases, and APIs—but it also creates repeatable failure modes that traditional controls don’t cover well (tool poisoning, prompt injection-driven tool misuse, exfiltration via “legitimate” tool calls, supply-chain compromises in MCP servers, and risky file/command execution pathways).

This session introduces SAFE-MCP, a vendor-neutral, open security framework designed to protect MCP deployments and tool-using agents. I’ll walk through a MITRE ATT&CK-aligned taxonomy of agent/tool threats (tactics + techniques), show how these map to real-world incident patterns, and then translate them into practical controls teams can implement immediately: control/data separation, cryptographic integrity, least-privileged tool permissions, policy gates for tool invocation, behavioral monitoring, and detection rules (Sigma-style) that fit existing SOC workflows.

The emphasis is on engineering reality: what breaks in production, what signals are reliable, and how to implement guardrails without killing developer velocity. Attendees leave with a threat model they can reuse, a mitigation checklist they can apply the next day, and reference patterns to harden their own agent stacks.

Enjoy a lunch while connecting with industry professionals and potential collaborators.

This hands-on workshop helps engineers harden a tool-using AI app end-to-end. We’ll start from a minimal agent + tool-calling baseline, then iteratively introduce common real-world risk scenarios (prompt injection tool misuse, malicious tool schema/tool poisoning, over-privileged tools, unsafe file/network actions, and “silent exfiltration” patterns). For each scenario, participants implement a practical guardrail: tool permissioning, policy gating, allowlists, control/data separation, structured outputs, logging/telemetry, and simple detection rules that are production-friendly.

The workshop is designed to be immediately applicable: teams leave with a reusable template architecture, a repeatable threat-model approach, and a short checklist they can apply to their own stacks. It is vendor-neutral, focuses on patterns and practices, and is explicitly not a product pitch.

Buffer time for sponsor demos, hallway convos, and schedule catch up.

At 4:30 AM, I wasn't woken by an alarm — I was woken by anxiety. Seven SKUs live across
480 stores in 24 states, personally liable, and one missed email away from disaster. That
anxiety became the catalyst for building an AI-augmented operating system that took me
 from 150% utilization to 20% — while scaling a CPG company to a $9M+ run rate.
 

This talk walks through the real architecture: how I used Claude Code as a thinking
partner, Obsidian as extended memory, and observable hooks piped into LangSmith and
Langfuse to create auditable, FDA-compliant agent workflows. I'll cover the practical
stack decisions — when to use SaaS observability vs. self-hosted, how to structure
agent-to-agent coordination for supply chain operations, and why governance and security
aren't afterthoughts when every user is becoming a maker of code.
 

This isn't a demo or a prototype story. It's a production system built under real
pressure, with real stakes, by one founder using AI to do what would normally require a
team of twenty.

Key Takeaways:

• How to turn domain anxiety into working AI agents (the "encoding" pattern)

• Production architecture for agent observability (LangSmith vs. Langfuse, same hooks)

• FDA-compliant agent governance — every tool invocation logged and auditable

• The extended cognition stack: AI processing + external memory + observability

• Security considerations as all users become code makers

Buffer time for sponsor demos, hallway convos, and schedule catch up.

Description TBA

Buffer time for Day 1 wrap-up, next steps for Day 2. Sponsor thank you's.

Keep the conversations going after the event with after party drinks at Ani’s Day & Night. Join speakers, builders, founders, and friends of AI Startup Rodeo for a relaxed Austin hang after a full day of talks, workshops, and demos. Expect good drinks, strong conversations, and a chance to keep connecting with the community in a more casual setting. Ani’s current menu includes cocktails, mocktails, beer, wine, coffee, and weekday happy hour specials.

Sponsored by Synapto.