The problem is not the AI. It is how you talk to it.
When someone tries an AI tool and concludes "this is useless for my business," the problem was almost never the tool: it was the instruction. They asked for something vague and got something vague. Garbage in, garbage out —at the speed of light.
Modern AI understands natural language, and according to McKinsey, the tasks that depend on exactly that language make up around 25% of total work time. The bottleneck is no longer the machine: it is how well a human knows how to ask. And asking well is a skill —it is learned and it is taught.
Think of the prompt as a job description
You do not tell a new hire "handle sales" and walk away. You explain what they have to do, with what information, what limits they have, and what you expect them to deliver. A good prompt is exactly that: a clear briefing for a lightning-fast collaborator who does not guess what you did not tell it.
The framework to do it fits in four letters: TCCA —Task, Context, Constraints and Ask. Every time a result comes out weak, it is almost always because one of the four was missing.
T — Task: what you want exactly
Start with a clear verb and a concrete deliverable. Not "help me with an email," but "write a collections email for a customer who is 15 days late." The task defines the work; if it is fuzzy, everything else blurs.
C — Context: what a stranger would need to know
The AI does not know your business, your customer or your history. Context is all the information you would give an outsider so they could do the task well: who you are, who it is for, what happened before, what tone you use with that customer. A good shortcut: paste an example of something you liked how it turned out. One example is worth ten adjectives.
C — Constraints: the limits and the format
Here you set the rules of the game: how long, what tone, what to avoid, in what format. "Max 120 words, cordial but firm tone, do not sound threatening, do not mention legal action." Constraints are what separate a generic text from one that sounds like your company and not a robot.
A — Ask: the specific request and output shape
Close by saying exactly what you want back and in what form. "Give me three versions so I can pick," or "return it as an email ready to send, with a subject line." The Ask avoids the back-and-forth: instead of asking, looking, correcting and asking again, you get something usable on the first try.
A real example: from poor prompt to TCCA prompt
Poor prompt: "write me a collections email." Result: a flat, robotic text you will not send.
TCCA prompt: "Write a collections email (Task) for a frequent customer who always pays on time but this time is 15 days late on an USD 800 invoice (Context). Max 120 words, warm and understanding tone because I value the relationship, no threats or mention of late fees (Constraints). Give me two versions —one softer and one a bit more direct— ready to send, with a subject line (Ask)." That prompt hands you something you can actually use today.
How to make it a company skill, not a personal one
The value jump is not that you learn to write good prompts: it is that your company saves and reuses them. Every time someone lands a TCCA prompt that works —the collections email, the complaint reply, the meeting summary— save it as a template and share it. That way knowledge stops living in one head and starts living in the business.
That is exactly moving up from level 1 (loose chat) to level 2 (instructing with templates). And it is probably the highest-return AI skill your team can learn this week, without spending an extra dollar on tools.
The Q.AI Take
Most companies want to buy more AI when they have not gotten even a third of the value from the AI they already pay for. Before selling you a new tool, at Q.AI we check whether the problem was simply how you were instructing it.
Often the best "AI project" of the quarter costs zero: teaching your team to ask well. — Martín, founder of Q.AI Consulting
- When "AI is useless," the problem was almost always the instruction, not the tool: vague in, vague out. Asking well is a skill that is learned and taught.
- Think of the prompt as a job description for a lightning-fast collaborator that does not guess what you did not tell it. The framework: TCCA —Task, Context, Constraints, Ask.
- Task = clear verb + deliverable. Context = what a stranger would need (include an example). Constraints = length, tone, format, what to avoid. Ask = what you want back and in what form.
- A good example in the context beats ten adjectives, and the Ask avoids the back-and-forth: you get something usable on the first try.
- The real return is not that you learn, but that your company saves and reuses the prompts that work as templates. That is moving from level 1 to level 2 —without spending on tools.