Level 01 · Advice
01
Chatbots
A passive assistant in a browser tab. Knows nothing about your business until you paste it in.
Autonomy ~18%
From a chatbot in a browser tab to a coordinated system running an entire operation — and how each step moves more autonomy to the model while the structure around it grows up.
The model is replaceable. The harness is your engineering asset.
Each step moves more decision-making to the model and gives it more of your business in context. Underneath, "harness engineering" is just files in folders feeding the right information at the right moment.
A passive assistant in a browser tab. Knows nothing about your business until you paste it in.
A pipeline that fires on a trigger. AI fills gaps inside steps you defined — same order, every time.
You give a goal — the model picks the path. Reason → Act → Observe → Iterate inside a harness.
A coordinated team of skills, MCPs and shared memory running an entire operation — humans in the loop where it counts.
The core idea: the leap from a smart chatbot to a production-grade operation is structure, not code — skills, memory, and MCPs are just markdown files and folders that feed the right context to the right agent at the right time.
Advice OperationsA passive assistant that lives in a browser tab. Smart on the surface, powerless beneath it: no business context, no execution, no awareness of last week's work.
Projects, gems, and custom GPTs help — but it's still static context you paste in manually.
Brand voice, audience, post history, what worked last week — none of it is in the model unless you paste it in every time.
Waits for you to prompt. Won't pull anything, run anything, or check anything on its own.
First drafts, brainstorms, explanations — anywhere a thinking partner is enough and execution isn't required.
A pipeline that fires on a trigger. The AI fills gaps inside steps you defined — same order, every time. Magical at first, brittle at scale.
If this week's topic is better as a Twitter thread than a LinkedIn post, the workflow can't tell — it runs the same steps in the same order.
The voice file you wrote three months ago doesn't know your carousels are outperforming text right now.
The workflow can't iterate on its own.
Repeatable, well-defined work where the steps genuinely never change.
You give a goal — the model picks the path. This is where AI starts thinking, not just doing. The shift from "follow my recipe" to "figure it out".
You didn't write those steps. The model decided them based on the goal you gave it.
A harness is the infrastructure that turns thinking into doing — read files, run tools, check its own work. Without one, you have a chatbot in a tab.
Not one agent doing one job — a coordinated system running an entire operation. Multiple skills, shared memory, real tools, with you in the loop where it counts.
Closer to organising a Notion workspace than writing code. Which is why the audience is business owners and knowledge workers — not just developers.
It's not more code — it's more structure: skills, memory and MCPs are markdown files and folders feeding the right context to the right agent at the right time.
— the thesis behind every Jentrix engagement
Skills, memory, rules — diffed, reviewed, CI-checked. Improvable like any other engineering asset.
A new joiner reads brand-voice.md the same way they'd read a Notion doc.
Drop in a better model next quarter; your harness — the asset — keeps shipping.
~95% autonomous, but you sit at the input and the publish step. Where it actually matters.
Bring the workflow you're not sure about. We'll spend 30 minutes mapping whether a harness pays off — and you'll leave with a written recommendation either way.
The model is replaceable. The harness is your engineering asset.
30 minutes with the people who own the workflow. We map roles, rules, memory, validation — and where humans stay in the loop.