AI is the ultimate tool. The prompt is the skill. | Advait Sontakke
AI & ThinkingJune 20266 min read
AI is the ultimate tool. The prompt is the skill.
Most people type into AI the way they text a friend — whatever’s on their mind, however it comes out. What arrives back sounds helpful. What it actually is, more often than not, is a confident answer aimed at everyone and no one. Here is what changes when you stop doing that.
You have a problem. You open an AI tool. You type the problem. You get an answer. You use the answer.
Fast. Simple. Feels like progress.
Here is the part that takes a while to see clearly. AI does not know who you are. It does not know your scale, your team, your constraints, or what you have already tried. When you ask it a question without any of that, it answers for everyone. Which means it answers, in any useful sense, for no one.
Think of it this way. AI is like an extremely powerful tool with no aim. Without direction, it fires across the entire domain of human knowledge. You get a little of everything. You get very little of what you actually needed.
Watch what happens when you aim it. Three layers. Same question. Same AI. Press each one.
Layer 1I need this
Layer 2Better context, I need this
Layer 3Max context, I need this
You typed — Layer 1
Recorded 25 Jun 2026, 9:00 AM IST · Claude Sonnet 4.6
Claude Sonnet 4.6 respondedSearched on 25 June 2026, 9:00 AM IST · Claude Sonnet 4.6 · Results may vary with context
Press the button above to reveal Claude’s response for this layer.
What Layer 3 was actually doing
The responses from the same AI shifted dramatically — not because Layer 3 was smarter, but because it was aimed. Here is what each part of Rahul’s message was doing that the earlier layers weren’t.
The nine layers — and what each one added
WHO I AM — Solo founder context changes everything. AI advice for a bootstrapped one-person operation is completely different from advice for a funded company with a management team.
MY BUSINESS — Revenue and channel mix signal operational complexity. ₹75 lakhs across Amazon, a website, and 12 retail stores is a very different problem than ₹75 lakhs through one channel.
MY TEAM — 6 people across 4 departments is a micro-team problem — not a scaling startup problem. This ruled out half the generic advice before a word was written.
WHY IT’S BREAKING — The growth context matters. This is a transition problem — not a startup problem. The old system worked. It just stopped working at a new scale.
WHERE IT BREAKS — Specific failure points gave AI a real diagnostic. Not a vague pain — a mapped breakdown Rahul could verify against.
WHAT I TRIED — Prevented AI from recommending the three things Rahul already tried and failed with. This alone saved three rounds of useless back-and-forth.
WHAT I WANT — The 48-hour independence test is a specific, measurable outcome. AI could now recommend solutions against a real success criteria.
MY LIMITS — Budget and tool constraints ruled out half the remaining recommendations before AI made them. The output became immediately practical.
WHERE DO I START — The actual question came last, after all context was in place. By that point, AI had no choice but to answer specifically. The question was the arrow. Everything above it was the aim.
“When you give AI more context, it is not being smarter. You are being smarter. The tool is finally aimed.”
But here is where people go wrong in the other direction
Once someone realises that more context produces better results, they start sharing everything. And this is where a different problem begins.
People start treating AI like a confidant. They type things into a chat window they would not say in a meeting, would not write in an email, would not share with someone they just met.
AI is not your therapist. It is not your lawyer. It processes data. What you type is what you give. That line matters.
What people actually type — that does not belong in a chat tool
“My business partner has been withdrawing money without telling me. Here are the transaction details…”
Financial disputes with real numbers belong with a lawyer — not an AI session.
“I’ve been feeling really low lately and I don’t know if this business is worth continuing. Let me tell you what’s been happening at home…”
Personal mental health belongs with a person trained to respond to it — not a data tool.
“My employee said something inappropriate. Here is the full WhatsApp conversation…”
Sharing a third party’s private communication — without their knowledge — creates real exposure.
“I have a pitch to [Investor Name] next week. Here is our full financial model and cap table…”
Confidential business information in an open chat session without understanding where it goes.
The right context for AI is professional, purposeful, and specific. Not personal, emotional, or confessional. The moment you share something out of feeling rather than purpose — you have crossed from directing a tool into something it was not built for.
What mature use actually looks like
The people getting the most from AI are not the ones who trust it most. They are the ones who understand most precisely what it is — and what it will never be.
When you reach that point, something changes in how you arrive at the blank window. You stop greeting it. You stop explaining yourself emotionally. You stop hoping it will somehow understand what you haven’t said.
Instead, you arrive with the thinking already done. You know the problem. You know the context that matters. You know what a bad answer looks like so you can describe around it before it arrives. What you type is not a question searching for an answer. It is a direction delivered to a tool that is now precisely aimed.
The blank window in front of you is not empty because you don’t know where to start. When you truly understand AI, it is blank because you already did the work before you opened it. The context is set. The direction is clear. What you are about to type is an arrow — not a search. What comes back, when you have done this right, is something you can actually use.
Five things to change starting today
Always tell it who you are before you ask anything. Your name, your work, your customer, your context. Even two sentences changes the output completely.
Name the gap, not just the goal. Don’t say “I need a better system.” Say “My business stalls when I’m not there and I’ve already tried SOPs and a manager.” The problem is more useful than the wish.
Tell it what a bad answer looks like. “I don’t want generic advice or advice to hire more people.” Ruling out the obvious saves half the work.
Keep personal matters out of it. Emotions, disputes, confidential data — these belong with the right human, not a data tool. Know the boundary.
Do your thinking before you open the chat. AI amplifies what you bring. A sharp thought in produces a sharp response out. The quality of the context you carry is the quality of the output you receive.
Questions this usually raises
Is this hard to learn? Does it take a long time?
No. The layers shown above are something you can apply in your next AI conversation. It is not a technique you study — it is a habit you build. Most people who try it once notice the difference immediately and never go back to asking bare questions.
What if I’m not sure how to describe my own context clearly?
That is actually useful to know. If you cannot describe your customer, your failure point, or your constraint clearly enough to type it into a chat window — that clarity gap is the real problem, and no tool will solve it. The exercise of writing your context down, even imperfectly, often produces the insight before the AI has responded at all.
Does AI actually understand my context, or is it pattern-matching?
It is pattern-matching — extraordinarily sophisticated pattern-matching across an enormous range of human knowledge. It does not understand the way a person does. But the quality of the pattern it matches depends entirely on the specificity of what you give it. Specific input finds specific patterns. Vague input finds every pattern. That is why context works.
I’ve been using AI for over a year. Am I already doing this?
Some of it, probably. But most long-term AI users are comfortable at a middle stage — using it seriously, but still putting context in after the fact instead of before. The tell is this: if your AI conversations regularly need multiple follow-ups before the output is useful, the context work is happening inside the chat rather than before it. Moving that work upstream is the shift.
A
Advait Sontakke
Brand director, commercial photographer, and ex-CA based in Mumbai. Founder of Advait Sontakke Visual Solutions and 7Miles Communication. Thinks in commercial outcomes first — whether the instrument is a camera or a chat window. Meet Advait →
