October 21, 2026

You're Using AI at 10% of Its Capacity. The Missing 90% Is Context.

Most teams use AI like a smarter search bar. The gap isn't prompts, it's context. Here's why your best data stays out of AI, and how to safely put it back in.

  • how to use AI at work
  • context engineering
  • enterprise AI
A dark card with a progress bar filled to 10% and the text "THE MISSING 90% — It's not your prompts".

Here’s a pattern I see in almost every company I walk into. Someone opens an AI tool and asks it to “write a follow-up email to a client about their late documents.” The AI writes something polite, generic, and about 60% right. The person rewrites half of it and moves on, mildly impressed and mostly unchanged.

Multiply that by a whole team and you get the state of AI in most businesses today: a very expensive autocomplete.

My estimate, from what I see in the field, is that most professionals use AI at about 10% of what it can do for them. And the reason isn’t what they think.

It’s not your prompts

There’s a whole industry selling prompt courses. Prompts point the way like a lighthouse in the middle of the night, but that isn’t enough to bring you into port.

Take the client email again. To write it well, the AI would need to know:

  • Which documents are actually missing from your file, and since when.
  • That your client has been late three quarters in a row.
  • That the partner prefers a warm tone with this client because they’re a long-standing relationship.
  • That the filing deadline is in nine days, and what it implies if that deadline is missed.

No prompt can supply that. That’s your context, and it lives in your files, your systems, your history, and in your people’s heads.

AI without your context is a very smart stranger. Brilliant, eager, but useless on anything that matters to your specific business.

The 4 levels of AI use

A simple way to see where your team stands:

  1. Ask. You use AI as a smarter search engine: “What’s the VAT rule for X?”
  2. Draft. AI writes the first version of an email, summary, or report from what you type.
  3. Work with your context. AI reads your real documents, client files and history, and gives answers specific to your business.
  4. Act in your workflows. AI agents handle multi-step work inside your systems (collecting documents, reconciling entries, preparing files) with a human approving the result.

Most teams are stuck between levels 1 and 2. That’s the 10%. Levels 3 and 4 are where the other 90% lives, and both run on context.

Why your best context stays out of AI

If context is the unlock, why don’t people just give it to the AI? Because in most companies they can’t, and they’re completely right.

The context that would make AI brilliant is exactly the context that’s most sensitive: client financials, contracts, HR records, pricing, strategy. Pasting it into a cloud AI tool means sending it to someone else’s servers, under someone else’s policies and, often, someone else’s laws. For regulated professions like accountants, lawyers, and healthcare, it is explicitly prohibited.

So your people do the sensible thing. They sanitize. They strip out names, blur the numbers, summarize instead of sharing. And, obviously, an AI working from a blurry copy of reality returns only empty, generic answers.

This is the central trade-off: with cloud AI, you have to choose between privacy and performance. Most companies choose privacy, correctly, and then wonder why AI doesn’t keep its promises…

Simulate your situation in your organization

A few questions, and you leave with practices you can apply today.

Run the simulation

Context engineering, not prompt engineering

The industry has started calling this “context engineering”: designing what an AI knows before it answers, not just what you ask it. That’s the right frame. But context engineering has a prerequisite most articles don’t mention.

You can only work with the context you’re authorized to use.

If your most valuable data can’t safely reach the model, even the best context strategy has nothing to work with. The problem isn’t technical skill. It’s the question of where you choose to put your data.

Bring the AI to the data, not the data to the AI

That’s the idea behind Sceptra.

Instead of sending your data to cloud solutions, we deploy the AI inside your infrastructure. Models run on your premises, so they can work with your real files, systems and history, and none of it ever leaves your network. That removes the trade-off. You no longer have to choose between privacy and performance.

Inside Sceptra, context is organized into spaces: per client, per team, per project. Each space has its own documents, its own history and its own access rules. The AI working on Client A’s file sees Client A’s context, and nothing else. Your team controls what goes in, and can see what comes out.

For an accounting firm, that’s the difference between “write a reminder email” and an AI that knows exactly which bank statements are missing for which client, drafts the right message in the partner’s tone, and flags the deadline risk. The accountant reviews it and hits send, and gets the time back for work that needs a professional.

That’s levels 3 and 4, the other 90%.

How to move your team past 10%

  1. Take stock of your current AI usage. If it’s mostly emails and summaries, you’re at level 2.
  2. List the context your team wishes they could share. Those are the documents they strip out or never paste. That list is your roadmap.
  3. Work on the architecture before the prompts. If sensitive data can’t safely reach the AI, no training course will close the gap.
  4. Organize context by boundary. Client, team, project. The right context for the right task, and nothing more.
  5. Keep a human on the final step. Level 4 isn’t autopilot. It’s an AI that does the work and a professional who owns the result.

The bottom line

The next leap in AI productivity won’t come from a bigger model or a better prompt. It will come from companies that work in coordination with AI without handing the company’s real value over to cloud models along the way.

The strength of your AI is only worth the context you can entrust it with.

Your context. Your models. Your values.