Tiller: Structure, Gates, and State for Human + Agent Teams
Join us at The Agent Loop - Chicago while we host JC Grubbs as he presents his talk "Tiller: Structure, Gates, and State for Human + Agent Teams"
Abstract
The agentic coding world keeps offering us the same two defaults. Either you personally review every line your agents produce, which stops scaling somewhere around the second parallel work stream, or you go lights-out and let the swarm run, which works great right up until somebody asks who decided this was done. Most of us live in the uncomfortable middle with a pile of ad hoc prompts, half-followed conventions, and a nagging feeling that the process is running us.
Tiller is Made In Tandem's answer to this chaos. It's an opinionated workflow tool for humans and AI agents building software together, born on our first multi-human, multi-agent client platform build. The division of labor is the whole design: Tiller holds the structure, the gates, and the state; your agent does the reasoning; you craft the solution design, provide steering along the way, and approve at the gates. Work lives as one tree (epics, features, tasks) in your repo as markdown and YAML, no database, no tab to click away to. Acceptance criteria become evals with two polarities, introduce new behavior and preserve what already works. And the done gate is the pull request merge, performed by a human every time. An agent never marks anything done. It's less traffic cop, more really good executive assistant. It knows where everything stands, files the work, nudges you toward the practices you'd pick on your best day, and leaves every real decision with you.
In this talk I'll show Tiller on real work, and I'll argue for the ideas underneath it: why your delivery is already a production system whether you designed it or not, why "all tests green" is a suggestion rather than a definition of done, why brownfield is the general case, and why the fastest teams are adding checkpoints rather than removing them. I'll also be honest about what broke along the way and where v1 stops on purpose.
If you're running coding agents past the toy stage (Claude Code, Codex, Pi, take your pick; Tiller doesn't care which), you'll leave with a working mental model for designing your own human-and-agent delivery system, and a repo you can clone to skip the first 6 months of figuring it out.
Location
This event is hosted in the Teamworking by TechNexus space. Upon arrival, guests can usually come directly to the 12th floor. If we have a large turnout and you arrive after 6pm, you may need to check-in with the security desk before heading up. The reception team will be there to greet you and help guide you to your room.
Parking & Transportation
The Civic Opera Building does not have a dedicated parking lot. There is limited metered parking nearby. We recommend using iParkit or SpotHero to find discounted parking options. The CME Building (10 S. Wacker) is the nearest parking garage, with several additional garages located within a two-block radius.
The venue location is highly walkable and conveniently situated near multiple CTA train lines and bus stops, making it easy to access via public transportation.
Agenda
6:00 - 6:30 Networking and Bites
6:30 - 6:45 Rising Star Speaker
6:50 - 7:30 Presentation
7:30 - 7:40 Q&A
7:40 - 8:00+ Networking and Wrap Up
Sponsors:
Food and Beverage: Focused an exclusive LangChain partner and boutique consultancy dedicated to building production-ready AI systems that work in the real world.
Venue: Teamworking by TechNexus
Continue the conversation between gatherings. Join our discord (https://discord.gg/bmdwxs2Q5F)