Zyte Developer community meetup #2 x Humanbound.ai
Your ai-agent doesn’t just answer questions anymore. It browses, scrapes, calls tools, and ships code, often unattended. This meetup covers two places that can go wrong before anyone notices, and what to do about each one.
First, Humanbound shows a price-monitoring agent doing exactly what it was built to do: scrape a competitor’s page every night. A single line of fine print on that page is enough to walk your own cost basis and floor price straight back to the competitor. It’s not an exploit or a broken rule, just an agent reading untrusted content it was told to read. They close the loop live: a failing security test becomes an exported guardrail, dropped into a stock LangChain agent in two lines.
Then Zyte opens the hood on the infrastructure running its own coding agents. A new open-source library runs agents as declarative, swappable background jobs, local or in the cloud, across harnesses like Claude Code and Codex, without locking into one LLM provider. You’ll see it powering Zyte’s spider-writing agents, plus how the team evaluates the code those agents produce.
Agenda (BST)
3:00 PM: Welcome & introductions
3:05 PM: AISecOps for the Agentic Age: Model, Test, Monitor, Demetris Gerogiannis, Co-founder & Co-CEO, Humanbound (talk + Q&A)
3:30 PM: Zyte CDP, John Rooney, Developer Engagement Manager, Zyte
3:35 PM: Open-Sourcing Zyte’s Remote Coding Agent Infrastructure, Konstantin Lopukhin, Head of R&D, Zyte (talk + Q&A)
3:55 PM: Closing & what’s next
==========
Talks in full
Talk 1: AISecOps for the Agentic Age: Model, Test, Monitor
Demetris Gerogiannis, Co-founder & Co-CEO, Humanbound
AI agents don’t just answer questions anymore. They browse, scrape, call tools, and act on whatever they read. That moves the security boundary: untrusted input is no longer just what a user types, it’s every page your agent fetches. The OWASP Top 10 for Agentic Applications 2026 ranks Agent Goal Hijack first (ASI01), and the shortest path in is content the model reads and the user never sees.
Humanbound will show it, not slide about it. A price-monitoring agent scrapes a competitor’s product page, as it’s built to, every night, and a single line of fine print on that page walks your cost basis and floor price straight back to the competitor. There’s no exploit and no broken rule here: the agent just used a tool it was allowed to use, on data it was told to read.
Then we close the loop live. That’s AISecOps: model, test, monitor.
Model: map where untrusted content enters your agent, and what it can reach once it’s in.
Test: turn each threat into an adversarial test that runs on every change.
Monitor: because a new tool, model, or page can reopen what you closed, without touching your code.
You’ll watch a failing security test become an exported guardrail, dropped into a stock LangChain agent in two lines (no new dependency), and the same attack fail on re-run.
Free and open source: clone the guard, run one scan, ship the rules it finds. Every attendee leaves with the repo, free Humanbound usage keys, and the one-line command to test their own agent the same day.
------
Talk 2: Open-Sourcing Zyte’s Remote Coding Agent Infrastructure
Konstantin Lopukhin, Head of R&D, Zyte
Zyte is releasing a new open-source library for running coding agents as remote background jobs. Define an agent declaratively once, then run it locally or in the cloud, swapping agent harnesses (Claude Code, Codex) and LLM providers without vendor lock-in.
Konstantin walks through the design decisions behind it, shows it powering Zyte’s own spider-writing agents in production, and shares how the team evaluates the code those agents produce before it ships.
Open source from day one: the repo ships alongside the talk, ready to clone and point at your own agent workloads.
What you’ll take away
Why the content your agent reads, not what your user types, is the primary injection point, and how to test for it
How a failing security test becomes a guardrail your own framework enforces, in code you keep
A new open-source way to run coding agents as remote background jobs, harness- and provider-agnostic
Free Humanbound usage keys and the one-line command to test your own agent the same day
The Zyte agent infrastructure repo, ready to clone
Speakers
Demetris Gerogiannis, Co-founder & Co-CEO, Humanbound
Konstantin Lopukhin, Head of R&D, Zyte
Host: DevRel team, Zyte
Who it’s for
Developers and engineers building or shipping AI agents that touch the open web, whether that’s scraping, tool use, or autonomous code generation. Useful for anyone who has watched an agent fail, get blocked, or do something it technically wasn’t told not to do.
Register free, spots are limited on Zoom.
Join the Zyte developer community: https://www.zyte.com/join-community/
Humanbound: https://humanbound.ai
Meetup #1 recap + slides: https://zytelabs.github.io/community-meetups/meetup-1/