Start by telling me
what needs to change.
Intent-to-System Engineering keeps what you wanted to accomplish connected to what you build and how you test it.
Take a vehicle waiting for photos.
Suppose your used-car manager has to chase three people to find out why a vehicle still isn’t ready. Before choosing an AI tool, we need to understand where the information lives, who updates it and what the manager needs to do next.
We then build a small version and try it against actual examples. Did it identify the right blocker? Did it miss anything? Can the manager act on the answer?
Those questions give us something to design and test. They also help us spot work that needs fixing before AI will be useful.
What should the
system actually do?
That question stays with us from the first conversation to the final test. I call the approach Intent-to-System Engineering. Here’s how we work through it.
01Frame the intent
Say who needs help, what should change and how we’ll know it has improved.
OUTPUT / OUTCOME BRIEF & SUCCESS MEASURES02Observe the work
Watch how the job gets done today. Where do people wait, chase information or work around a problem?
OUTPUT / WORKFLOW MAP & RESEARCH QUESTIONS03Design the system
Work out what software can handle, where AI might help and which decisions need a person.
OUTPUT / SYSTEM DESIGN & DECISION BOUNDARIES04Run the experiments
Build a small version to answer the biggest unanswered question. Use AI to help research and code, then check its work.
OUTPUT / PROTOTYPES & EXPERIMENT RECORDS05Evaluate & refine
Try everyday examples and awkward ones. Look at what the system missed and change the design where it falls short.
OUTPUT / EVALUATION RESULTS & KNOWN LIMITATIONS06Prepare to operate
Decide who would maintain it, how it would connect to your tools and what still needs testing before anyone relies on it.
OUTPUT / READINESS DECISION & IMPLEMENTATION PLANMost useful systems need more than a model.
Composite AI means combining these parts around a job. A rule might flag a missing field, AI might read a note, and a person decides what to do about it.
Where AI helps with the R&D.
We can use AI to examine documents, compare approaches, write prototype code and draft tests. Your team still checks the sources, reviews the code and decides whether the results stand up.
In the vehicle example, a simple rule might find a missing sign-off. A language model could interpret a technician’s note. The manager handles an unclear case. That mix of people, software and AI is what I mean by composite AI.
We keep the examples, versions and test results so someone else can follow what we tried. See how the vehicle-readiness project could work.
Learn it by using it.
In the 90-day programme, I teach the method as your team applies it. The point is to get better at choosing what to test, judging the result and explaining the next decision. Those are skills you can take into your next project.
Tell me what you’re working on.
A few lines about the problem and your team are enough to start. We can talk through the rest together.