Talks

I speak about building AI agents that do real work in production: how to make them secure, how they actually fail, and what changes for product and engineering leadership when execution gets cheap. Every talk comes from systems I’ve built and run, not from slides about other people’s work.

I’m based in Berlin, speak English and Portuguese, and give talks in person across Europe or remotely anywhere.

Talks for 2026

A Capability, Not a Credential: Giving AI Agents Real Access Without Handing Them the Keys 🇺🇸

Agents become useful when they can act on real systems, and dangerous for the same reason. This talk walks through the architecture we use to let agents write and run code against customers’ GitHub, AWS and other systems from a sandbox that never holds a credential: the designs we rejected, the three trust zones, the checks every call goes through, and the limits that remain. For engineering leaders, security teams and anyone putting agents near production.

Based on: A Capability, Not a Credential

What 639,000 Execution Steps Taught Me About How AI Agents Fail 🇺🇸

Most research on agent failures studies benchmarks. This talk looks at five months of production telemetry instead, classified with the MAST failure taxonomy, and at the headline finding that turned out to be an infrastructure bug. It covers what agents really fail at in production, why the dangerous failures look like success, and how to avoid fooling yourself with production data.

Based on: What 639,000 Execution Steps Taught Me

The Model Isn’t the Moat: Building the System Agents Need to Be Trusted 🇺🇸

Model quality is converging. What separates agents that ship from agents that stay in demos is everything around the model: state, permissions, durability, auditability and governance. A practical look at the agent harness, with lessons from building one for enterprise compliance, and why regulation like the EU AI Act makes this an architecture question rather than a legal one.

Based on: The agent harness and the bootstrap phase of AI agents

Execution Is Cheap, Judgment Isn’t: Leading Product and Engineering in the Age of AI 🇺🇸 🇧🇷

When AI makes building cheap, bad decisions scale faster than good ones. This talk is about what that means for teams and leaders: why the line between product and engineering decisions is disappearing, how to adopt AI across a team without losing quality, and the kind of judgment that becomes the real bottleneck.

Based on: Execution Is Cheap. Judgment Isn’t

Formats

Invite me to speak

Email hello@hugo.im with the subject “Speaking: event name” and a few details: the event and its audience, the date and location (or remote), the format and length you have in mind, and how travel is handled. I reply to every invitation, usually within a week.

Short bio: Hugo Nogueira is CPTO at Complyance, the AI-native GRC platform backed by GV, where he leads product, engineering and design and built the company’s agent infrastructure. Over 20 years he has co-founded and sold a startup, built 8+ ventures as a venture CTO at BCG Digital Ventures, and writes about AI agents in production at hugo.im.

A high-resolution photo is available here.

Past talks