One agent for everything
An agent that does everything is mediocre at everything. Several agents with clear tasks work better. The question is: who does what, and who approves?
Advanced course · 1 day · for teams that already have their first AI agents
First the blueprint. Then the build.
Your team has built its first AI assistants and agents. Now several of them are meant to take over an entire process: research, assess, write, check. On this day your team learns the common design patterns of such systems, sketches the workflow for its own agents on paper and chooses the right tool before anything is built. No coding. At the end, every pair has its blueprint.
Understand first, then design, only then build. By the end of the day, every pair has its agent org chart on paper, with approval points and a tool decision.
The map
Explained in everyday language. Each pattern with an example from a mid-sized business, when it fits and when it does not.
One agent passes its result to the next: research, organise, write. Simple, easy to check, right for processes with a fixed order.
One agent sorts each request and sends it to the right specialist agent. Right when many different cases come in.
An orchestrator breaks down the task, assigns subtasks and assembles the result. Right for big tasks, expensive for small ones.
One writes, one checks against criteria, back again until it fits. Right wherever quality matters more than speed.
Several personas assess the same result and award points against fixed criteria, a moderator sums up. Better than agents pushing texts back and forth.
All agents read and write in one shared place, each takes what it needs. Flexible, hard to keep track of.
At fixed points the system waits for a human: approval, correction, stop. Belongs in every system with external impact.
Sound familiar?
Most teams jump from “one agent” straight to “automate everything”. What is missing in between is a picture of how the agents should work together.
An agent that does everything is mediocre at everything. Several agents with clear tasks work better. The question is: who does what, and who approves?
The automation grows step by step. After three weeks, nobody knows why agent four talks to agent two.
First a platform is bought, then people think about what for. The right tool follows from the design pattern, never the other way round.
You bring your challenges.
We work on them live.
The course produces prompts, assistants and agents for exactly the tasks that cost your company time.
Afterwards
An overview of the seven design patterns as a poster, with examples from your business alongside.
The blueprint for a real workflow from the questionnaire: which agent does what, who hands over to whom, where a human decides, which AI model takes on which role, when to stop.
Fixed assessment criteria that agents use to check other agents' work, instead of just passing texts back and forth.
A decision on which tool will implement the blueprint, with a cost range per run, and a chapter on agents for your AI policy.
What happens on the day
The agenda below is the standard case. Your questionnaire decides which process your team works on. You receive the detailed schedule after the intro call.
Everyone draws the org chart of their department, and the picture stays on the wall, because a system of agents needs the same clarity. Then the seven design patterns, each with an example from a mid-sized business. Live demo: an agent that creates the next agent from a job description. Exercise: four reviewer agents assess the same proposal text against fixed criteria.
Tools compared: assistants on a platform, n8n, environments in which one agent controls others, and developer tools, each judged on operation, cost per run and traceability. Then each team of two designs its blueprint on paper, with the points where a human approves and the right AI model for each role. Finally the question of the human: who may write, how test and production stay separate, who carries responsibility. Each team presents its design in three minutes, the others assess it.
The two best designs go into the n8n course or a sprint. In the follow-up call we clarify which design gets built first and what data is still missing.
We talk about which agents are already running and what makes sense.
Your trainer
Florian Pagel has advised companies on digital projects for more than 30 years, first in marketing for brands such as Google, Nike and Montblanc, today on introducing AI in mid-sized businesses and large corporations. With Hello White Parrot he builds assistants, agents and automations together with teams and trains around 200 people a year, always on their own tasks.
He created the AI programme unlocktomorrow.ai and founded the AI coaching academy Solo-Helden for solo self-employed professionals. He is also the author of “Der KI-Knigge: Authentisch trifft Agentisch” (with Vivian Hecker, BoD 2026): how we stay human when the machine works alongside us. His principle in both roles: results you can apply right away.
Price
On site plus travel costs. 6 to 8 people, in-house or remote. Includes a questionnaire in advance, an overview of the design patterns as a poster, a blueprint template and a certificate of participation under Art. 4 of the EU AI Act. For the leadership team there is also a half-day version: only design patterns, tools and the question of the human, price on request.
FAQ
No. The day works with paper, posters and a no-code platform. Building happens afterwards in the n8n course or an implementation project, using the blueprint from this day.
Then “From Prompt to Agent” is the right place to start. This day assumes your team has already built assistants and knows how to describe a task to an agent.
EU-hosted platforms such as nuwacom and Langdock, n8n for visible wiring, agent environments in which one agent controls sub-agents, and developer frameworks for teams with IT. We show what we run ourselves and say where we see limits.
On the day itself we work with your workflows, not with your data. Which data may later go into which system becomes part of the blueprint: every handover between two agents gets a data rule.
At the points where decisions are made. Write access only with human confirmation, test and production data kept separate, clear stop rules for each agent. This is the part of the day that goes into your AI policy.
Yes, same agenda, same price. The blueprints are then created on a digital whiteboard. Where possible we recommend on site, because drawing on the wall is faster.
Next step
Tell us which agents are already running and which workflow should come together. You will receive an email with a confirmation link. We will then get back to you within one working day with suggested dates.
Name, email and one sentence about your topic. Then an email with a confirmation link. We only see your enquiry after you click it.
Cookies for Google Analytics and Google Ads?We would like to use Google Analytics to understand how our site is used, and Google Ads conversion tracking to measure whether our ads lead to enquiries. For this, Google (Google Ireland Limited) sets cookies, and data may also be transferred to the USA. Without your consent, nothing is loaded and nothing is sent to Google. You can change or withdraw your choice at any time via “Cookie settings” in the footer. Privacy policy