Sharing highlights from a two-hour underwater group chat that took place last night in my FDE community. https://t.co/YaB4sybcpT

---

A Private Chat on “How Enterprises Implement AI” Turned into the Most Insightful Rant Session in the Industry

Last night, I witnessed an online rant session focused on FDE (Forward Deployed Engineer). For those unfamiliar with the “FDE” concept, you can refer to my free open-source book, which essentially covers “how enterprises implement AI.”

This was a spontaneous group call organized by members of my FDE community—something I didn’t expect before my live teaching session scheduled for next Wednesday. It set the tone for the group ahead of time.

The participants came from diverse backgrounds: R&D, pre-sales, consulting, HR, professionals already implementing AI in traditional enterprises, and B2B practitioners serving governments and large corporations.

Over two-plus hours, everyone shared their views on the trending concept of FDE.

• Some see it as a “hexagonal warrior” (versatile expert);
• Others view it as pre-sales + delivery in the AI era;
• Some believe it’s essentially on-site development;
• And a few argue FDE shouldn’t be seen as a job title, but as a new way of delivering enterprise services.

Yet, it’s precisely these differing perspectives that make this discussion valuable.

When a new concept emerges, what’s truly worth observing isn’t just its official definition, but rather:

Why do so many people suddenly need this concept?

And:

What problem is each person trying to solve with it?

After the discussion, my understanding of FDE has become clearer—and perhaps less mainstream. Below is a sanitized summary of some key insights from the chat.

(If you’re interested in this community, visit FDE4.ai or click the “Original Link” at the bottom left to find the entry point.)

1. The Rise of FDE Isn’t Necessarily Driven by Demand Surge

The most straightforward narrative is:

“AI is penetrating every industry, so enterprises need people who understand both AI and business and can implement it. Thus, FDE will become a hot profession in the coming years.”

This logic holds to some extent.

But there’s another half that’s often overlooked:

The supply side of FDE is also undergoing dramatic changes.

Over the past few years, the supply-demand dynamics in software development, especially for general developer roles, have shifted noticeably.

On one hand, the growth of the internet has slowed. On the other hand, AI Coding is rapidly boosting individual developers’ productivity.

Tasks that once required collaboration among frontend, backend, testing, and product teams can now be handled to a large extent by a single capable individual with AI.

The result isn’t that “all programmers will disappear,” but rather:

Completing the same number of software projects requires fewer people.

This means many backend technicians will be forced to seek roles closer to clients and business—positions harder to standardize.

FDE offers a new narrative for this shift.

So, I prefer to interpret the current FDE wave as the convergence of two forces:

Enterprises genuinely need AI transformation; meanwhile, a large pool of technical talent is searching for new value anchors.

The simultaneous occurrence of these two factors has created today’s buzz.

This also signals a cautionary point:

The supply of FDE might grow much faster than the willingness of enterprises to pay for it.

“More people discussing FDE” doesn’t automatically mean “more people are making money from FDE.”

These are two entirely different metrics.

2. The Biggest Issue with FDE: Things Get More Complex, but Clients Don’t Want to Pay More

Traditional software outsourcing rests on a key premise:

Clients generally know what they want.

• I need a CRM.
• I need an order management system.
• I need an approval workflow…

Once requirements are broken down by a product manager, development can begin.

AI projects, however, are often different.

A boss might just say:

“Our company needs to go AI.”

Then all the questions fall on the service provider:

• Which department should be transformed first?
• What processes are suitable for AI?
• Is the data ready?
• What accuracy rate is required for deployment?
• Who’s responsible if something goes wrong?
• Which model should be used?
• How are token costs calculated?
• Will employees know how to use it?
• Who maintains it after deployment?

Even more complicated:

Clients themselves might not know where their problems actually lie.

Thus, FDE starts taking on multiple roles:

Consulting, requirements analysis, product management, pre-sales, R&D, delivery, training—and sometimes even managing organizational relationships.

From a workload perspective, it’s clearly heavier than traditional outsourcing.

But here’s the problem:

Will clients pay more because of this?

The answer is likely the opposite.

Many business owners might react:

“With AI now, shouldn’t it be faster and cheaper?”

This creates a dangerous gap:

The service provider bears increasing uncertainty, while the client’s price expectations keep decreasing.

If this isn’t resolved, so-called FDE can easily become:

One person doing the work of three or four, yet getting paid only slightly more than a regular engineer.

That’s not industry upgrade—it’s just a fancy packaging for high-level drudgery.

3. The “AI Anxiety” of Business Owners ≠ Real Demand

This is what I believe is the most easily misjudged aspect of today’s AI B2B market.

When you talk to traditional enterprise owners, you’ll find their enthusiasm for AI might be higher than the tech circle imagines.

ChatGPT, Agents, digital employees, knowledge bases, GEO…

Whatever’s trending, they’ve probably heard of it.

Especially in industries under pressure, many owners feel anxious:

• Are competitors already using it?
• Will my company fall behind?
• Can AI help me hire fewer people?
• Can it boost profits?

This anxiety is real.

But it’s different from actual demand.

At least three levels need distinction:

• First: The owner is willing to listen about AI.
• Second: The owner is willing to spend some money to try it out.
• Third: The enterprise has a quantifiable, repeatable problem that can generate ongoing value.

The first two levels are common now.

The third level is far less prevalent than imagined.

Some enterprises haven’t even completed basic informatization, data consolidation, or process standardization, yet they’re already thinking about AI Agents.

In such cases, once an FDE steps in, they’ll realize:

They’re nominally working on AI, but actually doing digital transformation, process consulting, or even management consulting.

Thus, for future enterprise AI services, a crucial skill isn’t “identifying needs,” but:

Determining whether a need is a genuine business problem or a projection of the owner’s anxiety.

These two are priced very differently.

4. The Companies Best Suited for FDE Might Not Be the Most Traditional Ones

An interesting question is:

What kind of company is best suited for AI transformation?

Intuitively, people think of the most traditional companies with almost no digitalization.

Because “there’s a lot of room for improvement.”

But in practice, these clients might be the hardest.

If a company:

• Has no data consolidation;
• Lacks unified systems;
• Processes are highly manual;
• Employees have limited computer skills;
• Much knowledge is stuck in WeChat groups, Excel, paper documents, and individual minds…

Then to implement AI, the first step isn’t Agents.

You’d need to help them catch up on the informatization they missed over the past decade.

The cost is very high.

On the other end, truly AI-native companies with strong R&D capabilities might not need external FDEs.

Therefore, I think the most noteworthy segment is the middle tier:

“Semi-digitalized companies.”

Typical characteristics:

• They already use computers extensively;
• Have ERP, CRM, OA, enterprise WeChat, Excel, etc.;
• Accumulated documents and data;
• But systems are siloed;
• Lots of repetitive data entry and manual judgment;
• Lack a strong internal R&D team.

These companies’ problems are most likely to be quickly solved by AI.

They have data, but it’s not usable; They have systems, but systems don’t talk to each other; They have processes, but many steps rely on manual handoffs.

AI’s value in this environment isn’t about “creating a brand-new system,” but:

Reconnecting broken information flows.

This might be the most comfortable battlefield for FDE.

5. The Real Challenge Isn’t Technical—It’s People

This was a clear consensus in the discussion.

When doing enterprise services, you inevitably hit organizational issues.

Because “AI efficiency improvement” sounds neutral, but within the enterprise, it often means:

• A certain role loses importance;
• A department loses part of its power;
• Someone’s long-built information barriers are broken;
• What used to take five people might now take one;
• Certain experiences are extracted and turned into corporate assets.

So why do employees resist AI?

Often, it’s not due to lack of understanding.

On the contrary, they might be fully aware:

If this system succeeds, it could hurt their bargaining power.

Thus, enterprise AI projects have a typical structure:

• The boss is excited;
• Middle management is cautious;
• Frontline employees may not cooperate

Without someone with real authority pushing it, no matter how capable the FDE, it won’t matter.

This is why good enterprise service professionals eventually have to learn something most programmers dislike:

Understanding organizations.

• Who is the real decision-maker?
• Who controls the budget?
• Who can coordinate business departments?
• Who benefits from this project?
• Who might be harmed?
• Who reports the project’s success?
• Who takes the blame for failure?

These questions are often more important than whether you use Claude, GPT, or Gemini.

6. When Enterprises Buy from Vendors, Sometimes They’re Not Just Buying Technology

A very “client-perspective” point in the discussion felt particularly real.

Why don’t enterprises do it themselves and instead seek external vendors?

Of course, they might lack R&D capabilities.

But another often-overlooked reason is:

Risk transfer.

If an innovation project is entirely driven by internal employees and fails, the responsibility is clear.

But if a vendor is involved:

If the project fails, you can say the vendor wasn’t capable enough; If the solution doesn’t work, you can switch to another; At least the risk doesn’t entirely fall on the internal project lead.

Thus, mature B2B services deliver more than just software.

They also include:

• Professional endorsement;
• Clear boundaries of responsibility;
• Reportable milestones;
• Verifiable deliverables;
• A sense of security within the organization.

This explains why some technically strong individual developers struggle when entering B2B.

They think clients are buying the “optimal technical solution.”

But clients might be buying:

A solution that can still be justified if something goes wrong.

These two goals are entirely different.

7. The Ultimate Value of FDE Might Not Be “Efficiency” But Making Organizations Computable

Many enterprises talk about AI with “cost reduction and efficiency improvement” as the first words.

But if you dig deeper, a more fundamental need emerges.

Traditional enterprises普遍 face this problem:

A lot of how the organization runs exists in people’s heads.

• A salesperson leaves, and client relationships go with them;
• A veteran employee departs, and no one knows how a process works;
• New hires lack a complete training system;
• The boss doesn’t even know what employees do daily.

After AI enters the enterprise, it could achieve something hard to do before:

Gradually digitizing work processes.

• What problems did employees handle today?
• What knowledge did they access?
• What judgments did they make?
• Which processes frequently bottleneck?
• How do high-performing employees handle certain problems?

Once these are continuously captured, they form a new organizational asset.

So from the boss’s perspective, a crucial potential value of AI is:

Transforming the company from “relying on certain people” to “relying on the accumulated system of the organization.”

This is far more important than saving a few minutes of work.

It means the organization can become:

• Observable;
• Measurable;
• Replicable;
• Transferable;
• And ultimately more replaceable.

In a sense, AI is extending BI (which could only see business results) forward to the work process itself.

This might be one of the long-term values of enterprise AI.

8. Relying Solely on FDE Project Fees Might Not Be a Great Business Model

This was another strong impression from the discussion.

FDE has several inherent flaws:

• Highly non-standard;
• Heavy pre-sales efforts upfront;
• Unpredictable project timelines;
• High client organizational costs;
• Easy scope creep;
• Revenue stops after delivery.

If income depends entirely on one-off on-site engagements, it’s essentially still a labor-intensive business.

Just instead of selling “development person-days,” you’re selling “AI transformation experts.”

The more interesting model is:

Using FDE as a client entry point, then connecting to ongoing revenue streams.

For example:

• Model API calls;
• Enterprise AI infrastructure;
• Managed operations;
• Industry SaaS;
• Standardized workflows;
• Data services;
• Vertical agents;
• Subscription models.

This changes the entire economic model.

FDE is no longer the end product but becomes:

The front-end for discovering needs and entering clients.

The real money comes from reusable assets downstream.

A very ideal path might be:

The first time entering an industry, doing highly customized projects; By the second or third client, starting to identify commonalities; By the fifth or tenth client, 80% of the work can be reused; Eventually, turning services into products.

If you start from scratch for every client, you’re not accumulating assets.

You’re just continuously selling your time.

9. What’s Worth Accumulating Isn’t “FDE Skills” but Industry Assets

If I were to redefine this opportunity, I wouldn’t say:

“FDE will have great prospects in the future.”

I’d rather say:

People who deeply understand a vertical industry and can independently deliver solutions with AI will become increasingly valuable.

These two statements are very different.

Because the first might lead people to learn a new job title;

The second makes people think:

What industry do I know better than others?

For example, if you’ve spent ten years in recruiting.

Your moat isn’t whether you can call large models, but knowing which steps in the recruiting process seem simple but are actually highly counterintuitive.

If you’ve spent ten years in construction.

You know which information handoff points in engineering projects are most error-prone.

If you’ve spent years in cross-border e-commerce.

You know the real pitfalls in advertising, customer service, supply chain, orders, and after-sales—only those who’ve actually been in the trenches know.

AI provides the leverage.

Industry knowledge determines where to apply that leverage.

Therefore, the most worthwhile endeavor isn’t necessarily becoming a “general FDE.”

It might instead be:

Choosing an industry you truly understand, re-doing its most expensive, repetitive, and error-prone problems with AI.

And then continuously reusing that work.

10. A Cautionary Signal: Industry Standards Might Mature Before the Industry Itself

Right now, FDE is spawning training programs, courses, communities, certifications, and methodologies.

This is normal.

Every new profession goes through this phase.

But the issue is:

It’s not yet fully validated what enterprises are actually willing to pay for.

If training on “how to become an FDE” matures quickly while few are profitable from actual FDE projects, a very familiar structure emerges:

Those digging for gold haven’t made money yet, but those selling shovels already are.

This doesn’t mean training has no value.

But at least several specific questions should be asked:

• How many person-days does a typical project require?
• What’s the customer acquisition cost?
• How long is the pre-sales cycle?
• What’s the conversion rate from POC to full project?
• Will clients pay again the second year?
• After accounting for communication, on-site work, rework, and after-sales, what’s the actual gross margin?

If no one can answer these numbers, the industry is more accurately described as:

An emerging opportunity, not yet a proven business model.

Final Thoughts

As the discussion wrapped up, I increasingly feel:

The term FDE itself might not be that important.

Today’s debates about whether it’s a job, a capability, a delivery method, or a startup opportunity largely stem from the entire industry still being in the phase of contesting interpretations.

Everyone is trying to align the concept with their own capabilities and interests.

• Tech companies call it technical delivery;
• Consultancies call it consulting;
• Developers call it hexagonal engineering;
• Training institutions will define it as a learnable profession;
• Platforms will aim to standardize it;
• Business owners might just think: “Can you actually help me use AI effectively?”

But strip away all concepts, and the core business questions are quite ancient:

• Can you identify problems others are truly willing to pay for?
• Can you solve them at a lower cost than others?
• Can you turn one-time services into assets that can be sold repeatedly?

If these three things hold, what you call it doesn’t matter.

It can be FDE, consultant, AI engineer, or nothing at all.

Conversely, if none of these are in place—

Then no matter how trendy the new job title, it won’t change the fact that it’s ultimately a difficult business to run.

So, compared to “Will FDE become popular,” I’m more concerned with another question:

After AI rapidly commoditizes technical capabilities, what will truly be scarce in the future?

My current answer is:

Client access, industry understanding, organizational trust, and the ability to continuously turn non-standard problems into standardized assets.

These things might be harder for AI to commoditize than “knowing how to write code.”

They might also be where the real pricing power lies in the next round.

(If you’re interested in this community, visit FDE4.ai to view paid content and benefits, or first check out my open-source book for free.)