Prompting looks like it needs no training until you have to do the same task fifty times. A prompt you improvised on a Tuesday produced something excellent, you cannot reproduce it on Wednesday, and a colleague running the same words gets a different answer again. This day is about turning that luck into something repeatable: prompts written deliberately, tested against real examples, and saved as templates the rest of your team can run.
You begin with anatomy. Every reliable prompt states the task, the context the model needs, the constraints it must respect and the exact shape of the output. You practise separating your instructions from the material being worked on, so a document containing the word 'summarise' does not quietly become an instruction, and you learn why positive constraints outperform a list of things not to do. From there you add examples: one or two demonstrations of a correct answer usually do more than another paragraph of description, and you see exactly how much difference that makes on the same input.
The middle of the day covers output and grounding, which is where most business use stands or falls. You specify a format the receiving system can rely on, whether that is a table with fixed columns, a JSON object with named fields or a five-line reply with no preamble. You then give the model your own source material and require it to answer only from that, to quote the passage it used and to say plainly when the source does not contain the answer. That single habit converts an assistant from a plausible writer into something you can check.
The afternoon is diagnosis and reuse. You take prompts that fail and work out why, distinguishing ambiguous instructions from missing context, conflicting requirements, a task too large to do in one pass and the drift that sets in over a long conversation. You break big jobs into chained steps, each verifiable on its own. Then you build templates with placeholders for the parts that change, assemble a small test set of five awkward inputs, and run each template against them so you know it works before anyone else depends on it. Prompts that are only checked on the easy example are the ones that embarrass you later.
Techniques are taught to transfer between assistants rather than tied to one product, because the tools move faster than any syllabus. Text is the focus; image prompting is covered in Adobe Firefly: Generative AI for Creatives and Photoshop AI and Generative Fill. You leave with a personal prompt library of at least five tested templates. The class runs live online in your own timezone or in the classroom in Colombo, priced in LKR for learners in Sri Lanka and USD internationally, with the recording available for twelve months.
