An AI agent that walks you through building other AI agents. By the end of an hour, you will have something that works, helps with something specific in your life, and that you understand because you made it. The harness builds context before the model runs, then runs PassWeb verification (patent pending) before the agent speaks to you.
Register interest
The main path is Substack: the same channel as AI PassWeb outreach. Signing up registers you for builds, tutorials, PassWeb framing, and Simpliloquy milestones. Subscribe there first.
You can also drop your email below to receive a direct ping when your install path needs a heads-up outside Substack. Entries forward to Mark only and do not replace Substack signup for registration purposes.
Opens markulett.substack.com
We send one email per submission through this site relay. Spam filters may delay delivery.
What it is
You sit down, talk to it, and over the course of an hour you build a real working agent that helps with something specific in your life. Your morning routine. Your client work. Your training plan. Your reading list. Whatever matters to you.
You do not learn agent theory and then build something. You build something, and the theory lands in the moment because you are doing it.
No class to schedule. No videos to watch. No prerequisites. The conversation is the curriculum.
The assembler stages your canon and skills deterministically before the model wakes. That tackles Configuration Impersonation in structure. PassWeb, which is patent pending, is layered on as deterministic verification instead of trusting ad hoc retrieval or a vibes check.
What you walk away with
A working AI agent that lives on your machine and helps with something you actually do. Not a sandbox. Not a toy. A real tool that knows real things about your real life.
You will know how to build agents now. The first one teaches you the pattern. The next is faster. The one after that you build alone. The skill compounds and goes with you.
You will understand what your agent knows and why. You will know it cannot just say anything. You will have built something powerful that you can still control.
How AI agents help real people
Each of these is a real agent someone like you could build in an hour with Simpliloquy. The agent reasons from what you wrote down about your life. It does not invent. It does not improvise. It works inside the boundaries you set.
Knows the kids' schedules, the after-school programs, the doctor's contact info, what each child eats. Drafts notes to the school, helps think through the week, surfaces what is coming up.
Knows your services, your pricing, your boundaries, your style. Drafts proposals and emails grounded in what you actually do. Will not invent commitments or quote a rate you do not offer.
Knows your sport, your training history, your typical recovery patterns. Reads your sleep and HRV data. Surfaces patterns. Suggests tweaks. Operates from your specific body, not generic advice.
Knows what you are building. The model train layout. The garden. The novel. The home renovation. Carries the context across sessions so you do not have to re-explain yourself every time.
Knows your medications, your appointments, your children's contact info, your financial advisors, your home maintenance schedule. Helps you stay on top of everything without becoming dependent on it.
These are not features Simpliloquy ships. These are agents you build. Simpliloquy gives you the platform and the teaching. You bring the knowledge of your own life. The agent that comes out is yours.
Who it is for
I know I need to learn this. I just do not see how to start.
The time to learn AI was 2023. The thing that matters now is agents. You do not need a Computer Science degree. You do not need to understand transformers. You need a tool that meets you where you are and walks with you. That is what this is.
Patent-pending verification
The harness gathers your canonical stack in code before the model speaks, which targets Configuration Impersonation structurally rather than trusting ad hoc retrieval. PassWeb is patent pending and wired in here as deterministic fill-in-the-blank verification alongside that staging. Translation: assembly reduces the odds the mismatch happens first; PassWeb proves the staged load matches what you wrote before downstream work begins.
You stay out of the weeds until you want detail. Verification runs every session without hand-holding.
How to start
Open markulett.substack.com, subscribe, then use the invite flow when replies go out.
Download from GitHub. Install Claude Code if you do not have it. Run it. Everything stays on your machine.
Open Simpliloquy. It greets you. By the end of the conversation you will have built an agent that helps with something you care about.
Going deeper
You do not need to be. The Simpliloquy agent calibrates to where you are. If you are new, it explains every concept the first time it shows up, in plain language, with one example. If you are experienced, it moves fast and skips the explanations.
You will need to install Claude Code and run a command in the terminal once. After that, the entire experience is conversational.
Simpliloquy is free during private beta. Subscribing to Substack doubles as registering interest outside this site. Announcements ride on that channel first.
The ongoing cost during use remains the Claude Code tokens tied to your Anthropic subscription.
For PassWeb standalone use separate from builds here, subscribe on Substack to register noncommercial handbooks or reach mark@beargrassai.com before shipping commercial products built on PassWeb.
When you write down what your agent should know (its canonical), you also write a fill-in-the-blank question whose answer is a specific phrase embedded in that canonical. Before the agent speaks to you, the harness asks that question, checks the answer with deterministic matching, and only opens the gate if it aligns with the anchored text.
The verifier stays outside the neural stack. Two models debating each other never enters this path.
Collectively those checks are labeled PassWeb. PassWeb carries a patent pending designation as of April 28, 2026, when the US provisional landed on file.
You are working on your own machine, with files you can read, in a system that asks for explicit authorization before writing anything important. You can always look at what an agent knows. You can always change it. You can always delete it and start over.
The architecture is designed so that the failure modes of inexperienced agent builders (the agent that hallucinates a commitment, the configuration that drifted, the agent that did not load what it claims to have loaded) are caught structurally. You can still build something that does the wrong thing. You will know it is wrong, because you wrote what it knows.
Three agents come with the platform. Each has one job and stays in its lane.
Bundled narratives and manifests stay under ordinary copyright protections. Terms for redistribution will ship before public GA.
PassWeb is patent pending. Noncommercial experimentation that follows newsletter registration at markulett.substack.com is encouraged. Commercial redistribution of PassWeb concepts inside monetized SaaS stacks needs a scoped conversation.
For commercial PassWeb integrations or reseller programs, mail mark@beargrassai.com.
For personal questions about this harness, mail mark.ulett@gmail.com.
Yes. An agent is a folder of files (a profile, a canonical, a test pair) on your machine. You can share that folder. The other person, with their own Simpliloquy install, can spawn the same agent.
If you write a great body agent or work agent, sharing it is encouraged. The pattern of "here is an agent I built that helps with X, try it" is part of how this gets better over time.
Mark Ulett. mark.ulett@gmail.com. Independent. Building this because the tools that exist did not meet his cousin where he was, did not meet his brother where he was, and probably do not meet you where you are either.
Free during private beta. Runs on your machine. Bring your Claude Code subscription. We bring the platform, the agents, and the teaching.