How I work

Two ways to integrate AI. I pick the one that fits your business.

Most of what's sold as AI for business is a chat window and a hope that your people will use it. I do it differently. Here is how, and why.

The two ways

There are two ways to put AI into a business, and I offer both. Which one you get depends on your circumstances.

A chat-based agent

The first way is an AI agent with tools, such as Grok Bot. You and your people talk to it, and it does the work: it reads, writes, looks things up and acts in the tools you give it.

It's fast to start. But it costs more to run, because the model does every step, not just the decisions. And it misses or forgets things from time to time. For work where a person is directing it and checking the result, that's fine. For work that has to happen the same way every time, it isn't.

An invisible workflow

The second way is a program. It runs your process step by step, the way you would write it down for a new hire, and it calls AI only at the decision points: reading a document, matching it to the right record, judging whether something looks wrong.

It takes longer to build. But once it runs, it's cheaper, safer and more reliable than an agent.

Why the invisible way has no dashboard

I create no interfaces and no dashboards. Technology should be invisible.

The usual problem with AI in a company is getting people to use it. Owners ask me how to get their staff to adopt it. With invisible AI, nobody has to adopt anything. The work is done, and the output appears where people already need it: in the system they already use, in the place they already look.

That means your people can't resist it, have very little new to learn and can't interfere with it. Nobody has to remember to open a tool. Nobody has to manage it.

Some companies want a dashboard so their staff can "manage the system." But it doesn't need to be managed. If the system needs a person to babysit it, the work hasn't disappeared.

How I choose between the two

It comes down to the work itself.

Many businesses end up with both. The first look at your situation is where we work this out, and it costs you nothing.

A small error rate is a fair trade

AI may make mistakes. I won't pretend otherwise.

What matters is that the error rate is small and known, and that one person checks the output. That is still cheaper than paying several people to do the work by hand.

At the trucking company I work with, work that took three people now takes one person about a third of their time to review. The owner accepted a small error margin to get there. Read the case.

Frontier models, not local AI and not resold bots

I use the frontier models from the major AI companies.

Owners often ask about running AI on their own hardware. The hardware doesn't pay off, so I can't recommend it unless it is absolutely necessary.

I also steer owners away from vendors who sell their own bot on top of a weaker or resold model. If a vendor's bot isn't built on one of the major companies' models, you can just close the door.

How an engagement runs

I work in stages: I look at the situation at no charge, then a proof of concept, then a prototype running inside your systems, then the live deployment, built until it is right. You can pay in installments, there's no long-term contract, and if I don't deliver the value I promised, I refund you. The offer on the home page has the details.

The point of all of it is simple. You stop carrying staff you only tolerate, you pay your best people more, and you grow.

For US owners with about 8 or more employees

If you want to grow without hiring, start here.

I work with select owners. Tell me what the work is, how big the company is and where you want it to go. I'll look at it at no charge.

Tell me what work you want gone