Why Everyone Is Hiring Forward-Deployed Engineers

AI made bespoke software cheap, and that's pulling SaaS back toward the thing it spent 15 years escaping. Also available as a video.

Harvey is one of the largest legal AI companies out there. In September it raised $550 million at a $15.5 billion valuation.

Harvey's cofounder Gabe Pereyra described in a recent interview what he sees as the natural progression for companies like his. You start out selling software as a service. Then, as you move upmarket, which mostly means your customers get bigger, you get pulled beyond pure SaaS into customer-specific infrastructure: custom deployments, custom integrations, sometimes on-prem.

The reason is simple. Large companies want the product wired into their own proprietary data, workflows, and internal systems, and they have a ton of all three.

That's why you now hear the term "forward-deployed engineer" everywhere.

How a vendor ends up inside the customer

Take a hypothetical. Harvey sells to large businesses, so say it lands Walmart.

At first, Harvey pitches Walmart its main platform. Someone inside Walmart becomes a champion for it and evangelizes it across Walmart's corporate structure. As more teams start using the platform, they start wanting more from Harvey.

Now Walmart doesn't just want the main platform. It wants an Excel add-in, or a sales tool, or something custom that lives inside the software Walmart already uses.

This keeps growing until Harvey is building so many bespoke products for Walmart that it has to dedicate entire teams to nothing but Walmart.

Those are forward-deployed engineers: software engineers deployed inside the walls of the client company to build whatever that client needs. In practice, the work is usually integration, connecting what Harvey offers to what Walmart already runs internally. At that point, Harvey's engineer is acting a lot like a Walmart employee.

SaaS was built to avoid exactly this

As a bit of an old head, I find this fun to watch, because it's how software companies operated 15 or 20 years ago. Then SaaS came along with the opposite premise: build one piece of software once, put it on the internet, and sell the same product to thousands or even millions of customers.

What you did not want was a bunch of people manually customizing it for every customer. Do that and you're a consulting company, and consulting companies don't scale the way a Salesforce does.

So small AI companies embedding engineers inside large legacy companies is a reversal of what the industry has been working toward for the last 15 years.

Why AI pushes the other way

The value prop of enterprise AI is that it understands your company specifically, and understands it deeply.

Your company doesn't want a generic AI agent. It could build that internally for much less. What it wants is an agent that understands its data, its weird internal processes, its permissions, its documents, its databases, its Slack channels, and the proprietary software it built 15 years ago that nobody understands anymore.

The catch is that your company often can't do this itself. Bureaucratic red tape makes it move too slowly. So it hires a small, fast-moving AI startup to do it instead.

For that startup, the highest-value version of the product is no longer a standardized subscription app. It's an underlying set of tools, prompts, and skills, plus the customization that makes the platform understand how this one company works.

And that blurs the line between a software company and a consulting company.

What keeps it from being consulting

Imagine Harvey has 50 forward-deployed engineers working inside Walmart. They're Harvey employees, but day to day they're solving Walmart problems. They talk to Walmart employees. They know how Walmart's systems work. Eventually they understand parts of Walmart's technical infrastructure better than a lot of actual Walmart employees do.

In other words, Walmart is letting Harvey accumulate some of its tribal knowledge.

But Harvey isn't a consulting company in the traditional sense, because all of those engineers are building on top of the same underlying Harvey platform. If the work were completely detached from that platform, you'd have Accenture: pure consulting, doing whatever the client asks.

Instead, Harvey takes what it learns at Walmart and turns the repeatable pieces into features of the platform. That's the flywheel. A customer asks for something. Your forward-deployed engineers build it. You realize your other customers have the same problem, so the custom solution becomes a general product feature. The next customer gets it out of the box, or sees it in the sales pitch.

Skip that last step and you're back to being a consulting company.

The engineer's job changes too

Historically, an engineer at a SaaS company sat a long way from the customer. A product manager stood in between, and whatever requirements came out of those meetings eventually made their way down as tickets.

The forward-deployed model collapses that. Companies now want someone who can write the software, think like a product person, act as a solutions architect, and talk to customers directly. Deep domain knowledge matters more too, whether that's finance, healthcare, or legal.

The job stops being "give me a ticket and I'll write some code." It becomes: go inside this complicated organization, figure out what their actual problem is, work out which parts of our platform already solve it, and build the pieces that don't exist yet.

Jevons paradox, for software

All of this is happening for one reason: AI makes bespoke software much cheaper to build. So companies are building a lot more of it.

A lot of people expected the opposite. AI was supposed to make generic SaaS infinitely scalable and eliminate a bunch of software engineering work. What we're seeing instead looks like Jevons paradox, where making the use of a resource more efficient leads to more consumption of that resource, not less.

Here the resource is software engineering work. Specifically, bespoke software engineering work.

Every company gets its own constellation of agents, internal applications, and workflows, built around how it operates. And the people building all of it may not work for the company using it.

You can see startups operating this way in one industry after another. Palantir in government. Harvey and GC AI in legal. My company, AgentSmyth, in capital markets. And a ton more in healthcare.

SaaS was the bet that every customer could run the same software. The forward-deployed engineer is the bet that, now that building is cheap, they don't have to.

Sources