What a Forward Deployed AI Officer Actually Does

A few months ago, a friend asked me to help him set up an AI workflow for his business because he had seen all these social media posts about how amazing AI tools are in increasing productivity. I sent him a detailed walkthrough with the exact tools, the exact steps, links to tutorials, even a YouTube video showing the whole process. He's smart. He runs a real company. He thanked me, said he'd get to it over the weekend.

Two weeks later I asked how it went. "Yeah, I tried Claude but couldn't figure out the rest."

We met for coffee and he handed me his phone and laptop. Fifteen minutes later, the workflow was running. Same tools. Same steps. The difference was that I was there.

That fifteen minutes says more about the AI industry's actual problem than any conference or research report ever will.

Everyone's buying AI. Nobody's using it.

Here's a pattern I see constantly: a founder or executive buys ChatGPT Pro, or signs up for Claude Cowork, or installs some AI SaaS tool. They try it for a few minutes. They get inconsistent results. They run out of tokens on the free tier, or hit the limits of the $20 plan. They conclude that AI is impressive in demos but doesn't hold up in practice.

Then they go back to doing everything manually.

Meanwhile, I'm using the same tools, often the exact same platforms, to run operations that would take a team of people to handle. The gap between what AI can do and what most companies are getting out of it is staggering. Not because the technology isn't ready. Because nobody is there to wire it into the way the business actually works.

I tried the obvious approach first. I'd recommend the exact stack someone should use. Here's the tool, here's the API, here's how to set it up. I tell them to ask the AI itself on any obstacles or roadblocks as it can help clear those up. It never works. People are busy. They have a business to run. Learning a new tool stack isn't their job, and it shouldn't be.

The only thing that consistently worked was showing up and doing it with them.

Two stories

A healthcare administrative company had five full-time staff members spending most of their days extracting and organizing data from high volumes of PDFs and Excel files. They'd tried using ChatGPT and Claude directly: copying and pasting documents, trying to build repeatable processes. It never stuck. The results were inconsistent, they kept hitting token limits, and eventually they wrote it off as not worth the effort.

I tried to guide them remotely. Sent instructions, walked them through the setup. They told me it wasn't worth their time. Five people doing manual data entry, and setting up the automation to fix it "wasn't worth their time." That's not a criticism because they were right. It wasn't worth their time to learn a new technical skill on top of running their actual business.

So I embedded directly in the company. In less than a week, I set up an automated extraction workflow using their own accounts and API keys: their preferred platforms, their existing tools, nothing foreign. The manual data work that consumed most of their team's capacity just stopped being a task for humans.

The real impact wasn't the time saved on extraction. It was what happened next. They'd been unable to grow their customer base because every new client meant more manual processing load on an already maxed-out team. With that bottleneck removed, they're now planning to double their customer base with the same headcount. Same team, twice the output.

A different engagement: a venture capital fund was paying for an expensive enterprise CRM platform that their non-technical team struggled to maintain. Data was stale, contacts were duplicated, portfolio updates were manual. I rebuilt the entire system in Airtable, which is a tool their team already understood, with an AI layer on top. It automatically keeps the CRM clean, pulls in public information from reliable sources, sends notifications for breaking news about their portfolio companies, and generates CRM updates from meeting notes so nobody has to manually log activities. All at a fraction of what they were paying for the enterprise tool that wasn't working for them.

Neither of these required cutting-edge technology. No custom models, no six-figure infrastructure projects. Just someone who understands both the AI tools and the business context, sitting down and connecting the two.

The problem with AI consulting

Most AI consulting today is education-focused. Someone shows you a magical demo, then sells you a course or hides the useful stuff behind a signup wall. The entire model assumes that what you need is knowledge. That if you just understood AI better, you'd be able to transform your business.

That assumption is wrong. Customers don't care about how the sausage is made. They don't want to become AI experts. They want the outcome: less manual work, faster turnaround, more capacity, lower costs. The "how" is someone else's problem.

The other pattern I see is agencies that force-fit every client into their own proprietary stack. They have their platform, their framework, their way of doing things, and the engagement is really about getting you onto their system. The client's actual workflow, tools, and team capabilities are secondary.

My approach is the opposite. Embed inside the company. Learn how they actually work and not how they say they work in a discovery call, but what people actually do at their desks every day. Then directly address the highest-value areas with whatever tools make sense for that specific team.

Why "forward deployed"

The concept isn't new. Dev shops have used forward deployed engineers for years. Palantir made it famous by sending engineers to sit inside client organizations because complex technology doesn't implement itself. The military uses the same term for positioning assets where they're needed, not where it's convenient.

"Forward Deployed AI Officer" isn't a cute title. It's a description of the method. Show up. Understand the operation. Build the systems. Deploy them. Transfer ownership to the team. Leave.

I'll be honest: this is not a scalable model. I can't embed in fifty companies simultaneously. Frankly I max out at 3 today, even with all the AI tools at my disposal, because this is a human to human role at this time. But it is the most effective way to launch a company onto a highly productive path with AI. The alternative of selling courses, sending decks, running workshops feels scalable and sounds nice in a YouTube video or witty X post, but the outcomes are dramatically worse. I'd rather do fewer engagements that actually change how a company operates than run a volume practice that produces PDFs.

The fifteen-minute lesson from my friend's office still holds. The gap between "here's how to do it" and "it's done" is enormous. Someone has to close that gap, and the only reliable way I've found is to be in the room.