Everyone rolling out AI Agents at scale is saying the same thing right now: we can’t hire enough forward deployed engineers.
And they’re right. But most companies are drawing the wrong conclusion from that fact.
The shortage of FDEs isn’t just a hiring problem. It’s a structural problem about what it actually takes to get AI deployed inside a real enterprise — and why the people and teams you’d normally lean on to do that work can’t do it anymore.
What Forward Deployed Engineers Actually Do
FDEs sit at the intersection of product, engineering, and customer success. They go on-site or deep into a customer environment, understand the actual workflows, and configure the product to work inside those workflows. They’re not building from scratch. They’re not doing basic support. They’re doing the hard middle work of making software actually land in the real world.
This role has existed for a long time. Palantir essentially built their entire go-to-market around it. You couldn’t buy Palantir’s product and self-serve your way to value. You needed their people inside your organization, configuring and training the system for your specific context.
That model worked. It was expensive and it didn’t scale the way SaaS was supposed to scale. But it worked, because complex software in complex environments requires human judgment to deploy well.
Now almost every serious AI product has the same requirement. And almost no one has enough of the people who can do it.
Why CS Can’t Fill This Gap. Please, Don’t Try.
The instinct at most companies is to solve this with customer success. CS is already post-sale, already focused on adoption and value realization. Just upskill them, right?
Wrong.
Traditional CS was built for a different era. The job was: help customers use software they’ve already decided to buy, make sure they hit their renewal metrics, escalate bugs. It was reactive, relationship-driven, and optimized for retention.
FDE work is different in almost every way. It’s proactive, technical, and optimized for deployment. You’re not waiting for a customer to have a problem. You’re going in before the problem exists and configuring the system so the problem never happens.
The skill set required is closer to a solutions engineer or a junior product manager with strong customer empathy than it is to a traditional CS rep. Most CS teams don’t have it. Retraining takes longer than most companies want to admit. And the velocity of change in AI tooling means that by the time you’ve upskilled your CS team for today’s deployment requirements, the requirements have shifted.
The Deeper Problem: Agents Can’t Deploy Themselves. Yet.
Here’s the thing that makes this especially acute right now.
The whole premise of AI agents is that they automate work. But deploying an AI agent is itself significant work — and it’s work the agent can’t do for you. Not yet.
Someone has to understand the customer’s workflows deeply enough to know where the agent fits. Someone has to train the agent on the right data, the right context, the right edge cases. Someone has to test it, catch where it breaks, and iterate. Someone has to get internal buy-in from the people whose jobs will change when the agent goes live.
That’s FDE work. And it’s manual, high-judgment, human work.
I saw this up close recently. A $6 billion AI company — one of the hottest in the market — had an agent quoting our team incorrect pricing. Telling us we’d need to quadruple our spend. When we asked how long the product had been in beta: a year. No one had properly trained the agent. The company’s own people hadn’t done the deployment work on their own product.
If a $6 billion AI company can’t train its own agent correctly after a year in beta, imagine what’s happening inside the enterprises buying these tools.
Why the Shortage Is Only Going to Get Worse Before It Gets Better
Palantir announced recently that they’ve gotten deployment times down over 90% using forward deployed engineers. That’s remarkable. It also means the best-in-class operator in this model — the company that essentially invented scaled FDE deployment — is still deploying manually, just faster.
90% reduction in deployment time is not the same as automating deployment. It means better tooling, better playbooks, more experienced FDEs. The human is still in the loop.
Every serious AI vendor is now competing for the same small pool of people who can do this work. The companies that came up through Palantir, the solutions engineers from the major cloud platforms, the implementation consultants from the enterprise software world — everyone wants them, and there aren’t enough of them.
Meanwhile the demand is exploding. Every enterprise that decides to deploy AI agents — and they’re all deciding, fast — needs FDE-caliber people to make it work.
What To Do About It
Three things worth considering if you’re on the vendor side:
- Do group training. The companies winning at deployment are the ones building serious enablement programs — not just documentation, but hands-on training that gives customer-side operators the skills to configure and train agents themselves. If you can’t train them 1:1 for smaller deals, at least try to do it in groups.
- Hire for deployment instinct, not just technical skills. The best FDEs I’ve seen aren’t always the most technically deep. But they really, really need to know how AI Agents work. And love tinkering with them.
- Don’t let CS pretend to be FDEs. This is the mistake that costs the most time. If the deployment requirement is genuinely FDE-level, staff it as such. Asking a CS rep to do FDE work and calling it “high-touch onboarding” is how you end up with a $6 billion company whose agent gives wrong pricing after a year in beta.
On the buyer side, it’s simpler: treat deployment as a first-class project, not a cleanup task after the contract is signed. Give it a dedicated internal owner with enough authority to actually change workflows. Train the agent properly before it goes live. And if your vendor can’t give you FDE-level support, factor that into the evaluation — because a tool that’s 20% worse but comes with serious deployment help will beat a best-in-class tool that self-serves into chaos.
Yes, It’s Even Harder for SMB
Everything above assumes you can afford the help. Most SMB customers can’t.
Here’s the math problem. A serious FDE engagement — the kind that actually gets an agent deployed and working — costs real money. Vendor time, configuration work, training cycles, iteration. That cost structure makes sense when you’re paying $100K+ a year for the product. It makes no sense on a $5,000 or $10,000 account. You simply cannot staff a meaningful FDE against a customer paying $400 a month. The unit economics don’t work, and no amount of goodwill changes that.
So SMB customers get pointed toward self-serve. Documentation. Help centers. Onboarding flows. “Get started in minutes.”
The problem is that self-serve only gets you so far — and with AI agents, “so far” isn’t far enough.
Zendesk’s own CEO said it plainly: enterprise customers using proper training and support hit 60-80% automation rates. Self-serve customers, going it mostly alone, land around 20%.
That gap isn’t a small efficiency difference. It’s the difference between an agent that transforms how your team works and one that gets turned off after 90 days because it never quite worked.
The two-human rule we’ve landed on for serious deployments — an FDE from the vendor and a dedicated GTM engineer on your team who owns the AI operations — is genuinely out of reach for most SMB companies. They don’t have a GTM engineer. They often can’t get meaningful time from the vendor’s implementation team. They’re running the deployment themselves, with whatever documentation exists, learning as they go.
And here’s the core problem that makes all of this worse: AI agents do not train themselves well. Not yet.
You cannot point an agent at your knowledge base, flip a switch, and expect it to work correctly at a level your customers will actually trust. It needs to be fed the right data. It needs to be tested against real scenarios. It needs someone to catch where it breaks — like the $6 billion company whose agent was giving our team incorrect pricing after a year in market — and iterate. That work requires human judgment. Consistently. Over weeks, not hours.
For enterprise, that work gets done. There’s budget, there’s vendor support, there’s an internal owner.
For SMB, it often doesn’t. The agent goes live undertrained, makes a few embarrassing mistakes, loses the confidence of whoever deployed it, and gets quietly turned off. Another failed pilot. Another team that tells their peers “we tried AI, it didn’t really work for us.”
The vendors who figure out how to solve SMB deployment — genuine self-training that doesn’t require months of human babysitting, or a services model that works at $5K ACV — are going to unlock an enormous amount of the market that is currently failing quietly. We’re not there yet. And until we are, SMB teams that want AI to actually work need to go in with eyes open: treat deployment as a multi-month project, find the one person internally who will own it obsessively, and don’t expect the agent to figure itself out.
The Irony That We Need Humans to Train AI Agents For Now is Real. But So What.
The biggest irony in AI deployment right now is this: we are building tools designed to automate human work, and the thing blocking those tools from reaching their potential is that we don’t have enough humans to deploy them.
That gap closes eventually. The tooling gets better. Playbooks mature. Agents get better at configuring themselves. Some version of “FDE as a service” will emerge and scale.
But today, the companies that win are the ones that take deployment seriously — that staff it properly, train it properly, and don’t pretend that buying the tool is the same thing as deploying it.
The agent doesn’t deploy itself. Not yet. And that “not yet” is where most companies are losing right now without knowing it.
