The Deeper You Embed, the Harder You Are to Remove

01 /The Concept

The core problem is that forward deployed engineers are becoming a bottleneck. A company can only take on as many new clients as its FDEs can handle.

Which means improving FDE productivity is a direct growth driver. The counterintuitive conventional wisdom is that FDEs are slow by necessity, because every client is unique.

In reality, the slowness comes from two specific sources:

- FDEs spend too much time extracting information about a client's workflows from the client's own employees.

- FDEs constantly reinvent solutions that other engineers on the team have already built for other clients.

Nixo built a platform to address both.

The first module captures information about a client's IT infrastructure and settings, and serves as the intake channel for client requests – submitted like tickets in Jira. The system automatically prioritizes incoming requests by urgency and importance, so FDEs always know what to tackle next.

The second module is about eliminating repeated work. It aggregates tickets and solutions from all FDEs across all clients, automatically identifying connections between them – so when a new ticket comes in, the engineer is surfaced with relevant solutions others have already built.

The platform integrates with external sources including GitHub, Slack, Notion, and HubSpot to pull in context that might be useful for FDE work.

The technical specifics are still thin, because Nixo is currently in Y Combinator and only just published its launch on the YC site.

02 /Why It Matters

A job title has been circulating more frequently lately: forward deployed engineer, or FDE.

The role isn't entirely new – Palantir pioneered it in the 2010s, embedding engineers directly inside client companies to translate Palantir's technology into working systems seamlessly integrated with the client's business processes and existing software stack. By 2016, Palantir had more forward deployed engineers than core product engineers.

AI startups building enterprise products are now rediscovering this model – because without it, you can build excellent technology, but you cannot build a successful product or a successful company on top of it. At the early stages, an AI product is essentially a chatbot or an API – and that API still has to be integrated into client workflows and IT infrastructure.

Since virtually every startup today is an AI startup, Andreessen Horowitz recently published a piece titled "Trading Margin for Moat: Why Forward Deployed Engineers Have Become the Hottest Jobs at Startups."

Demand for sophisticated AI products in areas like marketing, sales, and legal is growing. Startups shipping these products are managing to reach $5M, $10M, or even $20M in annual revenue within their first two years.

But the situation with complex products is like handing your grandmother a new iPhone. She'll happily accept it. But for her to actually use it, you have to configure it carefully – so she doesn't get confused or frustrated and can do what she needs to do.

For years, the conventional wisdom was that great software should work out of the box, with no configuration or implementation required. This approach enables scale and dramatically improves margins – because implementation and support are expensive. ChatGPT is the most recent canonical example of this model.

At some point, people assumed this rule should apply to B2B products too. But the most successful B2B examples – Salesforce, ServiceNow, Workday, Palantir itself – prove otherwise.

Companies wouldn't use Salesforce if the CRM weren't configured around their specific sales process. Workday wouldn't be useful if it weren't fully integrated with a buyer's HR systems.

That depth of integration carries real costs. At IPO, ServiceNow's margins were 63.2%, and Workday's were 54.1% – well below the 80%+ typical of pure-play cloud businesses. Even Salesforce, considered the gold standard of B2B software, spent $52 million to reach its first $22 million in revenue – before it built out an ecosystem of implementation partners.

But those upfront costs have compounding returns. Revenue growth accelerates over time, and the cost per implementation falls as a library of reusable solutions builds up.

The second reason margins recover is integration depth itself. The deeper a product is embedded in a client's business processes and infrastructure, the harder it is to rip out.

In other words: the more complex the implementation, the higher the switching cost to a competing solution. Deep implementations dramatically improve customer retention – which eventually becomes the single largest driver of revenue growth.

So the pattern holds: implementation costs initially compress margins, but then transform into a competitive moat. And that moat only strengthens as implementation costs per customer fall and retained revenue compounds.

But who should handle these implementations? The key insight is that a modern AI product is essentially an AI agent designed to fully or partially replace a human employee.

Human employees get onboarded by their managers – they're told how the company works, what the processes are, what the norms are. Who does the equivalent for an AI agent being hired in their place?

Company managers can't do it – they understand the business but lack the technical depth. The AI startup's own engineers can't do it either – they have the technical depth but don't understand the nuances of how a specific client company operates.

What's needed is someone who understands both: the client's business processes and the technical architecture. That's the forward deployed engineer – trained on the technology inside the startup, then embedded in the client company to learn its specifics and get the AI agent up and running.

AI startups are starting to recognize this need explicitly. As a data point: of 311 open roles at OpenAI, 22 can reasonably be classified as forward deployed engineering positions.

03 /Opportunities

Implementation has always been a fundamental part of selling complex B2B products. But the key shift right now is that virtually all B2B products are becoming complex – because AI is being embedded inside them, enabling them to take on far more sophisticated tasks than previous generations of software could handle.

As a result, structured, efficient implementation is becoming a mass-market problem rather than a niche one. Hiring the right people is table stakes – maximizing their effectiveness is the real competitive advantage.

The direction worth pursuing: platforms that make implementation specialists more productive. The category is new at this scale, which means early movers have a real chance not just to capture the initial market but to define it – before the wave of companies building complex AI products forces the issue for everyone.