You are using an older version of Internet Explorer. For the security of your Leifeng.com account and a better product experience, we strongly recommend using a faster and more secure browser

This is a temporary link, for article preview only, and will expire at that time

0 To the outside world, FDE is blazing hot; but what practitioners on the front lines see is absurdity, farce, and the unbelievable.

A while ago, Wang Tao, a To B veteran, saw a vendor claiming to do FDE whose entire company consumed only a little over 10 billion tokens per day.

"I alone consume 2 billion tokens a day," Wang Tao said, half laughing and half crying. "A company of 2,000 people consumes only 10 billion a day—isn't that pretty weak?"

And what Wang Tao saw was no isolated case.

In fact, over the past few months, a batch of companies of varying quality has emerged on the market, flying the FDE banner and already taking orders, teaching customers how to carry out AI transformation and implementation.

Can you expect them to truly help customers get AI implementation right?

"The answer, obviously, is no."

In China, why can't FDE save AI implementation?

  • FDE practitioners are a mixed bag, with plenty of amateur outfits among them

At the end of July, Liu Yu joined a company in a first-tier city as an FDE. Nearly two months of firsthand experience have gradually worn away his expectations for the role:

"I even feel it's worse than my old job doing ERP development for factories back home." Back then, he could at least lead a project himself, spending months on the factory floor, genuinely figuring out which business pain points the systems he developed could actually solve.

In the AI To B system, the FDE is not the protagonist—it's more like outsourcing.

Liu Yu told Leifeng.com that his company doesn't even have a ready-made AI product; the only presentable "product" is a set of Huawei methodology. What's truly valuable is the consultants' "endorsement"—the experience in their heads and the connections in their hands.

Within the company, the FDE's positioning is more like an ancillary service, bundled with this methodology and sold to clients as a package.

Liu Yu said bluntly: "We're like freebies—and the kind of freebies nobody wants."

After joining, he built two demos. But he gradually realized that what he made was merely there for clients to freeload—shown to clients once, to tell them "we have this capability," so the company could smoothly close deals.

Using FDEs as "freebies" is not uncommon.

When selling AI products to enterprise clients, many cloud salespeople at big companies or resellers also bundle services like scenario setup and employee training. These tasks, which originally belonged to pre-sales or delivery, are now also being labeled as FDE work.

Beyond the awkward positioning, FDE capabilities vary widely, and there are plenty of amateur outfits.

Xu Yang, an AI product salesperson working on the front lines and collaborating with FDEs, began to lose his illusions about the FDE role during an internal collaboration:

For a client project, he approached the internal FDE team to discuss how to support the client's needs. They said they had no time—all their people were tied up with industry-line clients—and tossed back a line:

"Just ask AI, and you'll know how to build what the client needs. That's what we do too—we just throw it to AI."

FDE practitioners vary in quality, yet participants keep pouring in. But most of those who have transitioned into the role possess only a "one-sided" FDE capability—either leaning toward consulting, strong in experience, or leaning toward engineering, strong in technology.

A real FDE, however, needs more than consulting knowledge or technical skill. Today's FDE market is a mixed bag, and many participants have neither a mature product nor enterprise delivery capability.

What struck Liu Yu as even more absurd was that he had somewhat become an outsourced worker. "I've never been to a client site, never even communicated with a client directly."

In his understanding, the FDE should be the person closest to the client, and the value of the role lies in eliminating translation costs. But since he couldn't reach clients and had to rely on his boss or consulting experts to relay messages, an extra layer of translation cost was added instead.

As a result, the FDE easily degenerates into an AI operator revolving around the PRD (product requirements document): feed the document to AI, run the results, then wait for the boss to keep correcting course. The so-called forward deployment never even touches the edge of the business—instead, it's all grunt work, dirty work, and drudgery.

  • Clients have weak digital foundations and are hard to please

The first challenge on the front lines is weak digital infrastructure.

Evan, the commercialization lead for the Data Agent at a BI vendor, feels this deeply: when to deploy FDEs depends on scenario complexity and data foundations.

"If the data foundation is really poor, we don't take the job—we tell the client to do data governance first." Only after the data governance team finishes do the FDEs handle the overall solution design and the specific Agent development.

He sees it clearly: without data governance, no matter how beautiful the FDE solution is, the Data Agent can't be implemented—inaccurate data makes all analysis meaningless.

One of the sticking points that makes data governance so hard is the enterprise semantic layer. In Xu Yang's view, it is an extremely complex data quality-check job, because companies are full of internal jargon.

For example, inside Huawei, "film" means PPT. Such corporate jargon, plus different business units calling the same concept by different names, all require manual semantic alignment one by one—large language models alone can't solve this.

But these are only execution-level obstacles. The truly intractable problem lies in the decision-making layer: the imbalanced power dynamic between client and vendor.

In the Chinese market, the client is always the dominant party, which ultimately relegates FDEs to mere outsourcing.

After systematically studying the Palantir model, investor Xu Li believes that FDEs need initiative, even a bit of aggressiveness—they can no longer adopt the traditional software services posture of "whatever the client says, we do."

But taking the initiative is not easy. In his view, only after the vendor first proves its capability will the client be willing to trust it fully, open up underlying systems and development permissions, and allow the vendor to perform structured semantic alignment.

Only with this near "God's-eye view" level of access can the FDE truly make the role work.

Players rushing in, each with their own agenda

Although many frontline FDEs and industry veterans have already punctured the FDE bubble, this frenzied and baffling craze continues.

The people pouring in each have their own calculations.

The first group of promoters are the hype machines that float outside frontline B-end scenarios yet talk up FDE with wild enthusiasm—training institutions.

They don't touch delivery and take no responsibility for results; they only heat up the concept, amplify the anxiety, and harvest both individuals and enterprises.

In the individual market, the targets are clear: mainly the unemployed and those with AI anxiety, including those who fiddled around in OPC but "couldn't make money and couldn't keep it going."

"In fact, the same crowd that ran OPC training is now running FDE training," Wang Tao pointed out in one stroke.

Their pitch is seductive: ride this wave, and as long as you can use AI and get trained, you can smoothly become an FDE and land enterprise AI transformation orders.

The B-end market has also attracted plenty of outright fraudsters. As Xu Yang observed: "As long as they can drop a few AI buzzwords, they can go out as teachers and lecture enterprises on strategic transformation."

The second group of promoters are consulting firms or outsourcers. Under the name of FDE, they repackage their existing capabilities and raise their premiums.

Wang Tao couldn't help but complain that consulting firms still "sell whatever's trendy": a few years ago it was data middle-platform consulting, then AI consulting, and now FDE consulting.

AI-savvy developers or outsourcers can now take on the work of an entire team single-handedly, so they too enter enterprises under the FDE banner, taking orders everywhere.

The third group of promoters are companies with their own large model products. For example, the BAT cloud giants—their demand is simple: use FDE to sell their own models into clients' hands.

Over the past year, several cloud vendors have set AI revenue targets that are almost all doubling. But selling AI products is no easy task, and it poses a major challenge to their existing organizations.

For example, in traditional sales scenarios, frontline salespeople were used to dealing with clients' IT departments. With little difference between cloud products, sales could basically close deals just by cultivating business relationships—sharing a few meals and downing plenty of drinks.

But selling AI products requires reaching business departments, even the chairman or CEO. To reach business departments, the team must have people who truly understand the business. This is where the importance of the FDE role stands out: they become the bridge between the client's business departments and sales, and a key factor in whether a deal gets closed.

So over the past six months, several big tech companies have been loudly promoting FDE, and some vendors have even published FDE white papers, mainly to attract more FDE talent.

Besides these three main groups of promoters, a large number of FDE practitioners have also been drawn in by the hype, thinking they should grab a spot first and figure things out later.

Of course, it takes two to tango. Beyond the promoters, business owners eager to get AI into production are also pushing FDE into the spotlight.

Many companies need AI transformation, but internally they either lack such roles or most employees lack the drive.

"Inside a company, who would use AI to smash their own rice bowl—build a tool that makes themselves obsolete?" Even the IT departments in charge of technology deal with traditional IT problems, not driving reform and transformation.

So, the outside monk chants the scriptures best. Hiring a "professional" FDE to come to the rescue seems to have become a kind of essential need.

Chen Wei, CTO of a SaaS vendor, believes that since companies have paid the money, they expect the vendor to deliver results, not just sell a model or a product. Precisely for this reason, a project only reassures the company when it comes with people who can help with implementation.

One side is willing to strike, the other willing to take it. In such a market, FDE is growing wildly.

FDE is just "old wine in a new bottle"

In fact, the FDE proposed by Palantir is an excellent working paradigm.

Palantir itself has a reusable platform centered on the ontology layer, and FDEs can use this platform to build ontologies for clients. The generalizable parts of frontline experience also flow back to the product team and get consolidated into platform capabilities.

It's just that FDE has gradually changed flavor in China. "It's getting more and more surreal—all kinds of strange labels can be slapped onto FDE," Wang Tao said with a wry smile.

FDE in the spotlight is being exquisitely packaged on one hand, while carrying the stigma of "rebranded outsourcing" on the other.

But peel off FDE's "shell," and inside are just familiar old roles—only now with AI as a new variable.

In the eyes of many industry insiders, this role used to be called delivery consultant; it's just that under the framework of AI and large models, the skill sets and deliverables are different.

Wang Tao gave Lei Feng Wang (WeChat account: 雷峰网) an example: SAP used to have consulting consultants and implementation consultants, and you can see the shadow of FDE in them too.

The difference is that SAP implementation consultants had to know ABAP (a proprietary high-level language designed by SAP for developing and customizing its enterprise software products) and also handle a pile of execution-side tasks. Whereas with Palantir's FDE, much of the configuration can be done by AI.

At domestic big tech companies, FDE also has corresponding roles.

Wang Tao believes that those who pivoted to FDE in order to push AI products are essentially no different from Alibaba Cloud's BTE (Business Technology Engineer)—a sales role that understands both technology and business, which can be understood as a tech-savvy sales expert, or an architect who sells solutions.

During the data middle platform boom a few years ago, many consultants did data organization work at client sites. Looking back now, they could also count as FDEs.

In fact, Wang Tao's own company has such examples.

Before the FDE concept was flying everywhere, Wang Tao's company called these roles PE. Later we saw that this was almost identical to FDE: both drive product deployment and rapid implementation on the client side.

"The only difference is the AI element, which lets one person finish the work that used to require one PE plus several implementation engineers."

In his view: "FDE is nothing more than that."

The author of this article covers big tech AI To B. Feel free to add the author's WeChat skylar_12_14 to exchange ideas.

Note: Wang Tao, Liu Yu, Xu Yang, Evan, Xu Li, and Chen Wei in this article are all pseudonyms.

This is an original article by Lei Feng Wang. Unauthorized reproduction is prohibited. See reprint guidelines for details.

Related articles

Editor

For your account security, please verify your email

Please verify your email