100+ Members and 16,000+ Messages: What an Open-Source Cohort Taught Me About AI

Earlier this year I ran a cohort with over 100 mentees that was entirely focused on open-source contributions and emulating a real-world developer workflow. There were 14 teams, participants sent over 16,000 messages in the Discord, 651 contributions were reviewed... and then I ran an exit survey.

And that survey was telling: 71% of the participants used AI to augment their work, 58% wished I provided better guidelines around how to do so... this was the lowest rated portion of the experience when I polled participants at the end of the cohort.

I want to share my boots-on-the-ground experience, what the data show, how AI helped our participants stay engaged, and how it harmed my ability to identify skill gaps that led to churn.

Mentorship becomes increasingly important as AI shifts the playing field of our industry, and as MCP-enabled tooling reshapes what "AI-augmented workflow" means. The lessons I learned from this cohort belong to every maintainer.

The key takeaway? AI is drastically impacting the shape of the open source ecosystem. It's the core of the gap between contributors who stay and contributors who churn. And we need to adapt to the ever-changing agentic AI domain.

Architecting Agentic AI

I have ADHD. Executive function is the hardest part of my day. A year ago I built an AI agent - gave her a name (Hikari), a face, a personality, and access to my filesystem, my repos, my Discord, my calendar. She works alongside me from 8am to 9pm every weekday, drafts code, runs my sprints, manages my community, and reminds me to eat.

This talk is the field guide nobody gives you: how a single person, working at home, can wire together an LLM + tools + context + triggers into a real agent that does real work. Not a chatbot you prompt - a coworker you trust with carte blanche on anything you can undo.

I'll walk through the five pieces (brain / identity / hands / memory / ears), demystify what most people overcomplicate (MCP, hooks, gateways), and share the safety framework that lets me hand her broad access without it being reckless.

Built solo on Claude Code with MCP servers, a Discord gateway listener, and an Electron desktop app showing her current state through sprite swaps. Talk ends with a 4-step minimum starter kit you could build tonight.

Cultivating a Remote Workplace

Most workplace allyship conversations focus on policy. This one doesn't.

"Cultivating a Remote Workspace" is a practical, no-overhaul-required guide to the daily habits that make a measurable difference to LGBTQ+ colleagues in a distributed team. Attendees leave with five concrete actions they can start today - and a short to-do list they'll actually use.

The session opens with a moment of genuine celebration, because Pride exists precisely because queer people built joy in the face of things designed to erase them - and that's worth honouring before moving on to the work. It then addresses the current landscape honestly, without dwelling: the conversation around LGBTQ+ rights has gotten quieter in recent years, and that's not the same as things getting better.

From there, the session moves into what any colleague - in any role, at any level - can actually do. The five areas covered are: names and pronouns, navigating questions, confidentiality, speaking up in rooms where queer colleagues aren't present, and the cumulative impact of inclusive language. None require a manager's sign-off, a policy change, or perfection. All of them are choices.

The session closes with a short, tiered action list and a reframe: treating queer colleagues like their existence is just normal isn't a grand gesture - it's giving back energy that would otherwise be spent simply existing in those spaces. That energy goes somewhere more interesting when it doesn't have to go there.

Presented internally at a company all-hands~

Building a Voice Agent with Deepgram

Voice AI is having a moment - and honestly, it's more accessible than you might think. In this session, I'll walk you through building a fully functional voice AI agent from scratch, using real-time speech-to-text, a large language model for reasoning, and text-to-speech to talk back.

We'll cover the full loop: capturing audio, transcribing it with low latency, feeding it to an LLM with the right context, and streaming audio back - fast enough to feel like a real conversation. Along the way I'll dig into the design decisions that actually matter: interruption handling, conversation state, and keeping latency low enough that your agent doesn't feel like it's thinking too hard.

By the end of this session, you'll have a clear mental model of how voice agents work under the hood, a working architecture you can build on, and enough practical knowledge to ship something at the hackathon today.

No prior voice AI experience needed - if you've worked with an API before, you're good to go.