It’s not the AI model that changes your company, it’s knowing how to use it

When I run a training session, there’s a moment that repeats almost every time. I show what a new model can do and the room lights up: questions, excitement, “wait, it actually does that?”. It’s fun, and it’s right to be impressed. But it’s also the moment when I stop them and say the line I repeat in every room. The model is not what will change your company. What changes it is knowing how to use it: the right workflows and the skills to put it to work on your real processes. And that capability isn’t a gift for the few, it’s learned. It’s exactly what I teach the people in front of me: some will use it themselves every day, others will carry it into their team. The model is the easy part, and getting cheaper. Knowing how to make it work is the hard part, and today the most valuable one.

And this week the biggest players in the world proved me right in the most spectacular way possible. Within a few days, almost every giant in artificial intelligence announced the same move. Not a more powerful model. Armies of people to send inside companies.

Anthropic, together with the financial giant Blackstone and Hellman & Friedman, on July 15 launched a company called Ode: a billion and a half dollars, a hundred engineers, with a single purpose, to sit inside client companies and build there the AI systems that solve a concrete problem. Its CEO, Chris Taylor, said it without mincing words: “It’s pretty easy to imagine this as a trillion-dollar company someday if we execute well.”

A few days earlier Microsoft had announced Microsoft Frontier: two and a half billion dollars and six thousand experts dedicated to exactly this, making AI work inside companies. Amazon put around a billion into a similar initiative. Meta is building a unit that places its engineers directly in clients’ offices. Four giants, the same bet, in the same week.

And the most interesting part is who is putting up the money. Blackstone is not a technology company, it’s one of the largest asset managers in the world. When finance, the people who have to make the numbers work, bets a billion and a half not on the model but on whoever installs it, it’s telling you where it thinks the profits of the next ten years are. It doesn’t mean they’re necessarily right, but when that kind of capital moves, it has done the math.

There’s a name for this job: forward deployed engineer

An AI forward deployed engineer is a technical specialist who works inside the client company rather than in a lab: they take a concrete business problem, build the AI solution by connecting it to the company’s real systems and processes, and see it through to a measurable result. The term comes from military jargon, where it means someone stationed on the front line, close to the action. It’s not the person who builds the model. It’s the person who makes it work where the work actually happens. Today that person is worth more than the model they use.

The name isn’t new. Palantir invented it in the early 2010s. Palantir sent its engineers inside its clients, intelligence agencies and then large companies, to build solutions on site, because the data was classified and the problems couldn’t be solved from a distant office. What’s new today is that this model, once rare and reserved for a few, is being applied to AI and is going mainstream. OpenAI, Anthropic, Google and others are copying it wholesale. That’s why it makes sense to talk about an AI forward deployed engineer: the same figure as always, now at the center of the AI wave.

For three years the race was about who had the smartest model. That race is becoming secondary. The model is now a raw material: powerful, and ever cheaper. To give you a sense, having a mid-tier model process the equivalent of seven hundred and fifty thousand words costs a few dollars today. The real problem, and therefore the real value, is something else. How do you make this stuff work inside a company that wasn’t designed for AI, with its processes, its scattered data, its habits? That’s where the trillion-dollar bet comes from.

And there’s a story inside the story. If even Anthropic and Microsoft, the ones who build the models, have understood that selling you the tool alone isn’t enough and they have to send people inside companies, then it’s the definitive proof of one thing: the gap between having the tool and getting a result is so wide that people are building billion-dollar companies on top of it. And the numbers back it up. According to MIT’s research on the state of AI in business, 95% of generative AI pilots produce no measurable impact on the bottom line. And the cause, the study says, isn’t the quality of the models. It’s how they get integrated into the organisation.

But is it really an engineer? The skills it takes

At this point an honest question: is it really an engineer? At the top tier, yes, no discount. The AI forward deployed engineer at OpenAI, Anthropic or Palantir writes production code, builds the data pipelines, connects APIs and models, works on cloud and infrastructure. It’s a genuine technical role, and the market prices it that way: in the United States we’re talking about salaries from three hundred to over five hundred thousand dollars, and job postings for this figure grew around 800% in 2025. It’s not an inflated title.

But the rare part isn’t the code. It’s a “T-shaped” profile: deep technical ability plus the skill, which most technical people don’t have, of sitting with a client, understanding their real problem and taking it to a measurable result. Good programmers are plentiful. People who can also do that are very few. That’s where the value is, and that’s why those numbers get paid.

The skills sit on two levels. The first is the one that matters most, and that a professional often already has: framing the right problem, knowing a sector deeply, talking to people and managing change, defining the number that has to change, and leaving the team able to carry on by itself. The second is the technical part, connecting AI to real systems. At the top tier that’s real software engineering. At the scale of a small business it’s lighter: APIs, automation tools, integration with the management software and the CRM, a bit of scripting, building a smart search over the company’s documents. Lighter, but not zero.

And here’s the note that holds the whole argument together. AI has lowered the technical bar: today a sector expert who learns to build can do things that three years ago required an entire development team. But “build” is still the key word. If you can only advise and you can’t actually make the AI run inside the client’s systems, you’re not a forward deployed engineer, you’re a consultant. The difference, once again, is the result that works, not the advice.

We’ve seen this before, with the internet

Let me give you a historical parallel, because it helps to understand where we are. In the late Nineties, when the internet arrived, for a while the spotlight was all on browsers and search engines, on the infrastructure. But the real money, for most normal companies, didn’t come from building a browser. It came from those who knew how to take the company online: who built the website, the management system, the e-commerce, who rethought the processes around the network. The infrastructure was the precondition, not the value.

With AI the same thing is happening, but faster. The model is today’s browser. Implementation is the website that sells. And this time the window to position yourself is opening now, not in five years.

Why this matters to you, whether you run an SME or work solo

Here comes the part that counts, whether you run a small company or you work alone.

Ode and Microsoft Frontier are aiming at multinationals. They will never come to work for your thirty-person business, and they’ll never touch the accounting firm down the street. But beneath the multinationals there’s an ocean of small and mid-sized companies that no army of engineers will ever reach. And in Italy this ocean is huge: AI adoption in companies has more than doubled in a year, from 8.2% to 16.4%, but we’re still below the European average. And the brake isn’t technological. It’s the lack of skills.

Sixteen percent means that more than eight companies out of ten aren’t yet using AI in a structured way. Whoever moves now has a real competitive window. But you don’t buy the advantage with the tool: you build it with someone who knows how to connect it to your processes. Before spending on technology, the question is only one: who, in your company, actually knows how to use it? If the answer is nobody, that’s the first investment, not the software.

And if you’re an independent professional, this figure is gold. Demand for AI skills is exploding exactly where supply is scarce. Companies’ lateness is your opportunity. Which is also why the serious money in AI today comes from businesses and not from personal subscriptions: around 85% of Anthropic’s revenue comes from companies and developers, not from consumers. The money follows the businesses, and the businesses are looking for someone to help them.

Two concrete examples of AI implementation

Picture an accounting firm with forty people. It can buy a Claude or ChatGPT subscription tomorrow morning. But nothing changes if no one can connect it to the management software, the client documents, the firm’s processes. The value isn’t in the subscription. It’s in the person who sits there for a week and builds the workflow that saves twenty hours a week on a repetitive task. The subscription costs twenty euros. That result is worth thousands.

Or take a small manufacturer handling quote requests by email. Today a technician reads the email, looks up the prices by hand, writes the quote. Two hours per quote. The AI model, on its own, solves nothing: it doesn’t know your price list, it doesn’t know your discount rules, it doesn’t touch your management system. The value is born when someone builds the bridge between that inbox, the price list and the management system, and turns two hours into ten minutes of review. That’s what an AI forward deployed engineer is, in practice.

Strategy first, then the model

There’s one point, though, where I see people go wrong most often, and it’s the one closest to my heart. Connecting the model to the processes isn’t just technical work. It’s strategy, before the tool.

Let me give you the example I use in marketing training. Before I let anyone touch an AI model, I insist it’s clear who the ideal client is and who we’re talking to. It’s not a box to tick, it’s the condition. Because a model, without that strategy, doesn’t stay neutral: it does damage. It writes perfect copy aimed at the wrong person. It takes an imprecise message and multiplies it a thousand times, automatically, at machine speed. It floods the market with generic content that weakens your brand instead of strengthening it, right at the moment when visibility is moving inside AI answers. The same power that, with a clear direction, makes you fly, without direction amplifies the error and does it faster.

That’s why the work that counts, the part that creates value, isn’t technical. It’s understanding the problem and the direction before switching on the tool. It’s exactly what a subscription can’t do for you, and what an AI forward deployed engineer puts at the center: strategy and workflow first, the model after. Whoever reverses the order is buying speed to get to the wrong place faster.

Who builds it for you, and who teaches you to build it

And here’s where I fit, because it’s a distinction that matters. There are two ways to close the gap. The first is to bring in a forward deployed engineer who builds the system for you, in production: deeply technical, done-for-you, and it makes sense when you’re big enough to afford it. The second, the one I work in, is to build the capability inside your team: I come in, find where AI actually helps, and teach your people the workflows and the skills, how to design, schedule and run AI agents on your real processes. I leave you a working system and the ability to keep it alive and extend it. Not theory: a built result, plus the skills to maintain it. The test stays the same for both paths, the number has to change.

Teaching people to use it well, by the way, also means teaching them not to get buried by it: working with AI all day carries a real cognitive cost, and I wrote about it in AI brain fry.

And when the job needs real production-grade technical work, I don’t improvise it: I hand it to a technical partner I work with, Broken Ice Technologies. I bring the strategy, the training and the knowledge of your commercial processes; they bring the technical hands when the project calls for building deep. For most SMEs, the sustainable path is keeping the capability in-house instead of renting it project after project.

What to do now, without chasing an AI forward deployed engineer

If you run an SME, stop waiting for the giants to come down to you, because they won’t. The right move is internal and surgical:

1. Write down the five activities that eat the most hours in your company every week.
2. Pick one, the most repetitive and least creative, because that’s where AI pays off most and risks least.
3. Assign that problem to a person with a clear goal and a deadline. For example: in six weeks this activity has to cost half the time.

Don’t buy ten licenses hoping something happens on its own. Buy a result on a specific problem, and then replicate it. If you don’t have that person in-house, that’s exactly when targeted training pays off, because you’re investing in a result and not in a generic course that then nobody applies.

And if the problem needs a technical level you don’t have in-house and don’t want to build, the good news is you don’t have to go hunting for a half-million-dollar unicorn. There are specialised firms, like Broken Ice Technologies, that do exactly this work at the scale of a small business: they come in, build the technical part and connect it to your systems. Your job stays the same, pick the right problem and insist on the number.

If instead you’re an independent professional, I owe you an honest note here, because this is where the most damage gets done. I’m not telling you to pass yourself off as a forward deployed engineer. That’s a role with serious technical training behind it, and improvising it means promising a client things you can’t deliver. Your advantage is something else, and it’s worth a lot: you know the real problems of a sector, and the real problem is what people pay for. You can be the one who understands where AI actually helps, who builds the lighter automations on processes you already know, and who leans on people who know how for the heavy technical work. Don’t sell “AI”, an abstract concept that scares people. Sell a solved problem, and be clear about where your competence ends. That’s exactly the model I work with myself.

AI implementation or hot air: a rule to tell them apart

A warning, because I don’t want to sell you a fairy tale. Around the word implementation a huge market of hot air is already forming: people who slap the AI label on old consulting and charge triple. How do you tell real value from noise? With a single rule, valid whether you’re buying or selling: real value is measured by a number, before and after. Hours saved, errors reduced, quotes done in less time, more clients handled. If someone pitches you AI and can’t tell you which number changes and by how much, it’s not implementation, it’s theater. Always ask for the number.

The right question, by now, is no longer which artificial intelligence to use. It’s who knows how to make it work, and whether that capability can grow inside your company instead of always staying outside it. Models will keep improving and getting cheaper, and that’s fine, that’s the easy part. The hard part, the one that creates value and today is worth billions, is the piece in the middle: between the tool and the result. Whether you run a company or work alone, that’s where it pays to put your attention over the coming months.

Frequently asked questions

What is an AI forward deployed engineer?

A technical specialist who works inside the client company rather than in a lab: they take a concrete business problem, build the AI solution by connecting it to the company’s real systems and processes, and see it through to a measurable result. Not the person who builds the model, the person who makes it work where the work actually happens.

What skills does the job take?

They sit on two levels. The first, which a professional often already has: framing the right problem, knowing a sector deeply, managing change with people, defining the number that has to change. The second is technical: APIs, automation tools, integration with management software and CRM, a bit of scripting, smart search over company documents. At frontier companies it’s real software engineering. At small-business scale the bar is lower, but it isn’t zero.

How much does a forward deployed engineer earn?

In the United States, at frontier companies like OpenAI, Anthropic and Palantir, from around three hundred thousand to over five hundred thousand dollars a year. Job postings for the role grew around 800% in 2025.

What’s the difference between a forward deployed engineer and an AI consultant?

The forward deployed engineer builds: they make AI run inside the client’s systems and take it to a number. The consultant stops at the advice. If you can point at the direction but can’t make AI run on the company’s real processes, you’re not a forward deployed engineer.

Does my small business need to hire a forward deployed engineer?

Almost never, and you certainly don’t need to hunt for a half-million-dollar profile. You have two more realistic paths: build the capability inside your team through training, and rely on a specialised firm for the production-grade technical work. In most cases it’s the combination of the two that works.

Do you want AI, in your company, to produce a measurable result and not just one more unused license?

That’s exactly what I do: I bring AI skills inside SMEs with practical programs, built on your real processes and not on generic slides. The goal isn’t to let you try a tool, it’s to leave the capability to get results in your hands. Take a look at my training programs for sales teams and for marketing teams. And if you’re a professional who wants to learn to bring AI into your sector properly, without promising what you can’t deliver, follow me: I’m preparing content and courses designed exactly for you.