No two workshops are the same, but most start from one of these three. Here is what they often look like.
For leadership and department heads
No developers and no prior AI knowledge needed.
Where AI Pays Off
Where AI helps your company most, what kind of solution fits, and what to do first.
Often with: the managing director and leadership team, plus the heads of the departments where the work happens, such as sales, service, operations, finance or HR. Any industry, with or without your own software team. Usually 4 to 10 people. One preparation day, one workshop day, one wrap-up day.
What the workshop day often looks like:
- Morning: we walk through three or four of your workflows end to end, with the people who run them, and mark where time, errors and handoffs pile up. The handoffs between departments get special attention, because that is often where most time is lost and nobody owns the problem.
- Then: for each spot, we decide what kind of solution fits: an off-the-shelf tool, a tool you already have, set up differently, a changed process (sometimes without any AI), something built for you, or nothing for now, and why.
- Afternoon: we rank by benefit and effort, work out the top three, and decide the first step and who owns it.
You leave with: a ranked list of where AI could help, the three most important areas worked out (solution, what it needs, risks, how you will know it worked), the places where AI is not worth it for now, and a decided first step. All of it written up on the wrap-up day.
If the first step is a build: I can build a prototype with you within weeks and take it into production. If you have no software team of your own, we decide early who runs it after me.
From practice: logistics dispatchers need training dates for their drivers, scattered across thousands of driving-school websites. Researching them by hand doesn't scale. For Modulschulung.jetzt, a service of BKF Online-Schulungs GmbH, I built an AI system that reads those websites and extracts the dates. Every date is checked against its source, and cost and quality are measured continuously. Manual research became an automated process, with quality and cost under control.
For the same company, AI turned training videos into multilingual assessments. Producing that material went from weeks to minutes. Read the case study
For teams that build software
Your own team, or the partner who builds for you.
Agentic Engineering in Your Codebase
Your developers work real tickets with AI agents and keep a setup they go on using.
Often with: a development team and its tech lead who use AI coding tools, or are about to, and want them to work in an established codebase, including in regulated environments. Usually 6 to 12 people; larger teams in several groups. One preparation day, two workshop days, one wrap-up day.
Preparation day, here: time in your codebase, choosing the tickets with your tech lead, and taking a baseline from your delivery data (for example cycle time, rework and review load) so you can measure the effect later.
You leave with:
- Context and conventions for the agents, committed to your repository
- Real tickets that every participant worked with an agent, through your normal review
- Guardrails the team agreed on: what agents may change and see, where human review is mandatory, which checks run automatically
- Connections to your issue tracker or documentation where they help, within your security rules
- A short team playbook: which tasks go to an agent, which not, and how to brief it
Wrap-up day, here: I refine the setup with two people from your team who carry it forward, and write up the playbook and a simple way to compare your numbers with the baseline.
From my own work: I build this way every day. My own agents and a set of skills work like a virtual product team, covering product, architecture, UX and research. Every change follows the same path: a written spec, an AI-assisted build, a stored review diff, and sign-off by two reviewers before it is merged.
AI Pilot to Production
What your AI pilot needs to become a system you can run, maintain and hand over.
Often with: companies whose AI prototype (an assistant, an automated pipeline, an agent) convinced the right people and now has to go into production, whether your own team or a partner built it. The business owner, the people who built it, whoever will run it, and whoever owns compliance. Usually 4 to 10 people. One preparation day, one workshop day, one wrap-up day.
Preparation day, here: a review of the pilot's code and architecture, and conversations with the people who use it.
We work through: where the AI component sits and how it stays separate from the rest of the application; how you know the results are good enough, and notice when they get worse; cost, response times, failure modes and fallbacks; data and compliance boundaries; who owns it after launch.
You leave with: a target architecture sketch, the gaps between pilot and production ranked by risk, an approach for checking output quality over time, and a clear go, change or stop recommendation with a plan for the first 90 days, written up on the wrap-up day.
From practice: the questions above are the ones behind the Modulschulung.jetzt system: answers checked against their source, costs logged and capped, quality measured against a recorded test set. In the BKF video pipeline, the AI layer stays separate from the application; it runs on its own, and the client's team owns the platform.
