Key Takeaways
Remember that email?
Remember That Email?
Let's go back to where we started. A client sends a change request by email.
AI drafts the reply in 10 seconds. You review and send. Faster, but the bottleneck is still the mailbox.
The email triggers a prompt chain. AI extracts the request, checks it against project data, drafts a response with context. You review a complete package instead of starting from scratch.
The email never reaches a human inbox. An agent receives it, classifies the request, pulls the relevant data, drafts the response, routes it for approval if needed, and sends it. You see a dashboard of what was handled overnight.
Same email. Three factories. Three completely different outcomes.
Key Takeaways
Context is everything.
Research first. Always. Ground it in reality.
Never ask AI to DO finance.
It will be wrong. Ask it to BUILD the tool that does the math.
Use AI to build "old school" software.
Dashboards, scripts, tools. No AI embedded. You don't need AI IN the product to benefit FROM AI building it.
Ask AI to generate your prompts.
Don't start from scratch.
Reuse instructions.
Same thing twice? Make a skill or template.
Test, test, test.
No developer ships on the first try. Neither does AI.
Describe steps thoroughly.
It's a new employee on day one.
Process > Tool.
The tool is only as good as the workflow around it.
Start with Phase 1, but plan for Phase 3.
The 6% who succeed planned the transformation from the start.
“The next Google or Amazon of AI probably hasn't stood up yet. Maybe it's in this room.”
Building AI Systems Is a Continuous Loop
This is not a one-off IT project. Building with AI is more like running a business than managing a system. You iterate, learn, adapt - continuously.
The tools change. The models improve. Your processes need to evolve with them. The companies that treat AI as a “set and forget” deployment will fall behind those that treat it as an ongoing practice.
About Me
Florian Smeyers - Founder of SFLOW. I help companies integrate AI into their workflows - not by replacing people, but by redesigning processes.
How We Use It at Agidens
Real-world application: we use AI to fetch User Stories from Azure DevOps, draft documentation, generate content for our media pipeline, and build internal tools - all with human review gates in place.
The key: AI does the heavy lifting, humans do the quality control. Every output goes through review before it ships.