Home / AI consultancy / AI agents

Service

AI agents that work like a team

One agent with a long prompt is a demo. A useful set of agents has roles, tools, a way of checking each other's work and a person who signs off. We build the second kind, and Luros, our own kit, is built on the same idea.

What goes wrong with agents

• No clear job. An agent asked to "help with marketing" does a little of everything and nothing well.
• Marking its own homework. The agent that wrote the work also decides it is finished.
• Missing tools. It can describe the task but cannot reach the system that does it.
• No memory of how you work. Every run starts from zero, so the same mistakes come back.
• Nobody can see it. Work happens somewhere, and no one knows what, or why.

What we build

• Roles and a reporting line. Each agent owns one job, and someone else reviews its output before it ships.
• Playbooks. Your way of doing a task, written down once, so agents follow it every time.
• Tools and access. Connections to the systems the work lives in, with limits on what each agent may touch.
• Sign-off. The points where a person approves, and everything else runs without them.
• A view of the work. Who did what, and the evidence they did it.

Why us

This is one part of our AI consultancy work. We build agent products. Luros is the marketing team you cannot afford to hire: it puts search, answer engines and conversion on autopilot for a founder with no marketing team, and it is built out of the same idea, a team of agents with roles, skills and playbooks. ToolsWanted is a public board of the capabilities agents have got stuck on, so we see where agents get stuck beyond our own. OfficeSim shows a company of agents as a live pixel-art office. More in our work. Our founder, Syd Lawrence, is a two-time exited founder and data and AI operator.

Questions

What is an AI agent?

Software that uses an AI model to decide what to do next and then does it with tools: reading, writing, calling systems. It works towards a goal rather than answering one question.

Will agents replace our team?

They take on defined, repeatable work. People still set the direction and sign off what ships.

How do you stop an agent doing something it should not?

Each agent gets only the access its job needs, work is reviewed by a different agent or a person, and the risky steps need a human to approve.

How do we start?

Email help@minithlabs.com with the work you want agents to take on. We reply with where we would start.