What Is a Forward Deployed Engineer, and Why Is AI Going Custom?

A forward deployed engineer is a software engineer who works inside a customer's business, learns how the work is really done there, and builds the AI around it. Job ads for the role grew more than eightfold in a year, and on 2 October Anthropic committed $100 million to train 10,000 of them.

Every business gets the same AI model. What makes it useful in yours is fitting it to how you work: where your information lives, the exceptions your team handles, and who approves what. The companies selling AI have found that this fit does not come in the box, so they are sending engineers to build it, one business at a time.

What is a forward deployed engineer?

Most software companies sell you a product and leave you to set it up. A forward deployed engineer goes the other way. They leave the vendor's office and work in the customer's, often for weeks, next to the people who do the work.

The role started at Palantir, the data company, more than a decade ago. AI companies such as OpenAI and Anthropic now hire for it, and Microsoft has built a whole unit around the idea. Anthropic's new training programme calls the role a Frontier Deployed Engineer.

A typical few weeks looks like this:

1. Watch the work. Sit with the team, see what they do by hand, and find out where the information lives.
2. Write down the rules. Most of what makes the work hard sits in someone's head: which customer needs what, when to pass something up, what counts as a mistake.
3. Build it into the tools the business already uses, such as the inbox, the spreadsheets and the accounting system.
4. Test it on real cases, including the awkward ones, with a person approving the results.
5. Stay until the team uses it, then hand it to someone inside the business who owns it from then on.

Here is how that compares with the two kinds of outside help most businesses already know:

	Consultant	Software vendor	Forward deployed engineer
What you get	Advice and a plan	A product to set up yourself	A working system in your own tools
Where they work	Workshops and meetings	Their own office	Next to your team
When they leave	When the plan is delivered	At the sale, with support after	When your team uses it

Why are AI companies hiring so many of them?

Because their customers bought the software and still could not make it do the work.

In April 2026, Indeed had more than eight times as many job ads for forward deployed engineers as a year earlier, a rise of roughly 729% (Business Insider, May 2026). The biggest AI companies are now spending on it directly:

• Anthropic launched Claude Frontier Academy on 2 October, with $100 million to train 10,000 engineers by the end of 2027 (Anthropic, October 2026). Its stated aim is "to solve one of the most pressing issues in AI implementation: talent." Each engineer spends 12 weeks leading a real AI project inside their own organisation. The first firms taking part include Accenture, Deloitte, McKinsey and Commonwealth Bank of Australia.
• Microsoft set up the Microsoft Frontier Company in July, a $2.5 billion unit of more than 6,000 people, to place engineers inside customers and build AI systems alongside them (GeekWire, July 2026).

These companies sell the most capable AI models there are. If the models fitted a business on their own, they would not be paying engineers to go and fit them.

Why does off-the-shelf AI stop short?

An AI model reads and writes well, but it knows nothing about your business until someone tells it.

Say your sales coordinator turns customer emails into orders. Any AI tool can read the email and pull out the products and quantities. It does not know that one long-standing customer writes "the usual" and means last month's order. It does not know that orders above a customer's credit limit go to the finance manager first, or that the current prices sit in a spreadsheet someone updates every Friday. Your coordinator knows all of this without thinking about it, and none of it is written down.

That missing knowledge is where AI projects tend to stall. The tool works in the demo and trips on the third real email. A forward deployed engineer spends the first days finding those rules, writes them down and builds them in. After that, the AI handles each order the way your coordinator would, and passes the unclear ones to a person.

Singapore's own numbers point the same way. In the Ministry of Manpower's survey of 2,560 firms early this year, 28.5% had started using AI, but only 3.8% had built it into their core business processes (MOM, April 2026). Two of the biggest barriers firms named were a lack of skilled people (42.4%) and fitting AI into their existing systems (37.4%). Closing those two gaps is the forward deployed engineer's whole job.

Is the future of AI custom?

For part of your work, yes.

Drafting, summarising and answering general questions need nothing custom. The AI you already have does them well, and it gets better every few months without any effort from you.

Custom AI solutions earn their place in the work tied to your systems and your rules: orders, supplier bills, client reports, approvals. That is where most of the hours go, and where a general tool trips on the exceptions.

Custom also means something smaller than it used to. Custom software once meant months of building from nothing. Now the model already exists, and the custom part is connecting it to your tools, writing down your rules and examples, and deciding where a person approves. For one piece of work, that usually takes a few weeks.

If you are not sure which kind of work you have, the five-question test for AI agents sorts it quickly.

What if your business will never get one from Anthropic?

Most businesses will not. Anthropic's programme works by nomination through its account teams, the first cohorts are in San Francisco, New York and London, and the first firms taking part include large banks and consultancies.

The work itself scales down. A smaller business does not need a team of engineers for a year. It needs one person to do the five steps above on one piece of work, and then hand it over. That person can come from two places.

Someone inside the business. Anthropic's own finding is that "a small group of deeply skilled people drives an outsized share of what AI delivers." In a smaller business, that group may be one person. If you have someone technical and curious about AI, give them time as well as a licence. Take one piece of work off their plate for a month and let them run the five steps on it. The implementation guide covers how to run the project.

Someone you bring in. An outside engineer who works the same way: with your team, inside your tools, and gone once your team runs it without them.

What does good forward deployed work look like?

Whoever does it, check for these:

• They watch the work before they choose a tool. The first week is spent with your team.
• They write your rules down, in files you keep, so the knowledge no longer lives in one person's head.
• They build inside the tools you already pay for, and add new software only when those cannot do the job.
• They test on your real cases, including the awkward ones, before anything goes live.
• A person approves anything that leaves the business, pays money or commits you to something.
• They stay until your team uses it, and train each person on their own work.
• You own the result: the instructions, the examples and the system, in your own accounts.
• They tell you when AI is the wrong answer. Sometimes a template or one rule in a tool you already have does the job for nothing.

Common questions

Is a forward deployed engineer the same as an AI consultant? The roles overlap. A consultant usually advises and plans, while a forward deployed engineer builds and stays until it works. Many good AI consultants now do both, as the guide to AI consulting explains.

What is a frontier deployed engineer? Anthropic's name for the role in its new programme. The work is the same.

Do we need to hire one full time? Rarely. One piece of work is a project with an end date. After that, the system needs an owner inside the business, and that owner does not have to be an engineer.

How do we know whether custom work is worth it? Work out what the task costs you in hours each year, then compare that with the cost of the build. The method is in how to find the AI use cases worth doing.

Where to start

Pick one piece of work your team repeats every week. Ask the person who does it to write down every rule they follow, including the ones they think are obvious. Then see whether the AI in the tools you already pay for can follow those rules. Where it cannot, you have found the custom work, and you can see how big it is.

If you would like a second opinion on where to start, book a free call. Thirty minutes, no obligation, and if the tools you already pay for can do it, we will say so.