Zaela, the AI harness for your company

An AI agent is a model inside a harness. The model writes and reasons. The harness gives it a loop to work in, tools to act with, memory, rules and checks. In short: agent = model + harness. If that is new to you, start with what an agent harness is.

Most of what is written about harnesses is about coding. Claude Code, Codex and OpenCode are harnesses for one developer working in one code repository. But a marketing team, a sales team or a service desk also wants an AI that knows its rules, works in its systems and stays within agreed limits. That needs a harness too, built for a whole company. Zaela is that harness.

Inner harness, outer harness

Birgitta Böckeler of Thoughtworks splits a harness into two layers. The inner harness comes with the product you use: the loop, the built-in tools and the defaults of Claude, ChatGPT or Claude Code. The outer harness is what you add on top: your instructions, your knowledge, your methods, your connections and your checks.

Zaela is an outer harness. It does not replace the AI your team already uses. It works inside it: in Claude (desktop, Cowork and Code), in ChatGPT or in any assistant that supports MCP, the open standard for connecting AI to data and tools. Your people connect one link, and everything comes along.

Models are rented. The harness is the part you own.

That is why the outer harness matters most. When a better model arrives, you switch. Your knowledge, methods, roles and rules stay where they are.

Every part of a harness, and where it lives in Zaela

LangChain describes a harness as all the code, configuration and logic that is not the model itself. Microsoft lists what that layer does: it drives the model and tool calls, manages state and context, and applies approval policies. In Zaela, those jobs map onto four building blocks and a set of rules around them.

• Context and memory: Memory. Shared knowledge per topic or brand: brand rules, tone of voice, templates, logos and earlier work, with version history so nothing disappears quietly.
• Instructions: Skills. Methods written down once and used by everyone: how to write like a human, how to translate meaning instead of words, how to make a reel.
• Roles: Agents. Trained roles such as a content writer, a report builder or an SEO specialist, each with its own instructions, tools and limits. They know each other and pass work along.
• Hands: Tools. Connections to your platforms and data through one gateway, from Google Analytics 4 and Google Ads to image and video generation.
• Permissions: access per person. Your AI sees exactly what you may see, nothing more. Keys are encrypted and never visible, not even to the AI.
• Logs: every action recorded. Who did what, through the website or the AI.
• Brakes: budgets and limits per person, and approval by a person where it counts, such as before a website goes live.

Everything the AI uses, your team can also see, adjust and add to in the Zaela web app. The harness is not hidden in code.

What changes when the harness is for a company

A coding harness and a company harness do the same job for different people. The differences are practical.

• Shared, not personal. Memory and methods are used by the whole team, not kept in one developer’s files.
• Procedures, not code tasks. The work is a monthly report, a post in your tone or a competitor scan. The checks are a quality bar, a recalculation or an approval, not only a test suite.
• Business systems, not a terminal. The tools are your analytics, advertising accounts, CRM and publishing platforms.
• Strict separation. Every organisation gets its own workspace, strictly separated from every other one. Its memory and agents stay its own.

One request through the harness

Take a request many teams know: “Make our monthly marketing report.”

1. Your AI, Claude or ChatGPT for example, receives the request and conducts.
2. Memory supplies the report template and the brand rules.
3. The report builder takes on the work.
4. Tools read Google Analytics 4, Google Ads, Meta Ads, LinkedIn Ads and Search Console, and a calculator recalculates every ratio from the raw numbers.
5. You get an Excel report with every channel and a short summary of what worked, within exactly what you are allowed to see.

Nobody had to write the perfect prompt. The harness already knew what your report should look like.

A harness that grows with every slip

A harness is never finished. Each time the AI gets something wrong, ask which rule, method or check would have prevented it, and add it. That work has a name: harness engineering.

In Zaela it follows one simple rule: ask for the same thing twice, and turn it into an agent. Every lesson becomes an agent, a method or memory, and from then on the whole team has it. That is how Zaela takes your shape: the more your team uses it, the more it works like you.

What Zaela is not

• It is not a model. It works with the models you already use, and you can switch.
• It is not another chatbot. It is the brain and the hands behind the chat you already use.
• It does not replace a coding harness such as Claude Code. It works inside it, as the layer with your company’s knowledge, methods and tools.

Want to see the building blocks at work? See how a request flows through Zaela. For a wider comparison of coding, managed and company harnesses, read the complete guide to AI harnesses on i-scoop.eu, which is also published by Edge.