Text | Beyond the Page Layout, Author | Drawing
Over the past three years, the most expensive people in the AI industry have been model scientists.
Today, the types of people that OpenAI, Anthropic, and Google most want to hire have changed.
Not a researcher, not an algorithm engineer, not even a large model expert.
Instead, they are a group of people who need to travel, be stationed on-site, attend meetings, and revise processes.
They have a new name: Forward Deployment Engineer (FDE).
This may seem like an insignificant position, but it could represent the biggest shift in the AI industry over the past three years: the myth of the model has officially ended, and the battle for practical application has begun in full swing.
Silicon Valley's tech giants have finally realized that the models themselves are no longer the problem. The real challenge is for companies to know how to use them. As a result, a previously unwanted position has seen its value skyrocket overnight.
LinkedIn's 2026 Workforce Report shows that global FDE (Software as a Software Developer) job postings increased 42 times between 2023 and 2025, while AI engineer job postings increased 13 times during the same period, with the former growing at about three times the rate of the latter.
This unconventional frenzy for talent has torn away the veil that the entire AI industry has been keeping hidden for the past three years.
I. The model was implemented, but the organization failed to keep up.
Since the inception of ChatGPT, the main theme of the AI industry has been clear: from who can create the strongest model to who can create the best agent.
By 2026, the question had changed. Enterprise customers began asking another question: We bought AI, so why haven't we seen much change?
This is the biggest illusion in the entire industry: that a model equals productivity.
The reality is that many companies have spent a lot of money to purchase AI/Agents, employees have registered accounts, and the IT department has created a demo of an internal knowledge base, and they were excited for a month.
Then... six months passed, and nobody used it. The working methods were exactly the same as before.
It's not that employees are uncooperative, nor that management lacks determination, nor that the model is inadequate. The real Achilles' heel of a company in the production environment is never how it communicates, but rather where its historical data is located, whether the format is correct, and its quality. Where do approval responsibilities follow, and who has the authority? How is customer data imported, how is the ERP system integrated, and how are existing compliance and security systems compatible?
These are not technical issues, they are organizational issues.
It's like putting a rocket engine on a horse-drawn carriage. The engine is real, the thrust is real, but the horse is still a horse, the track is still a dirt road, and the driver has never learned how to press the accelerator, let alone where the emergency brake is.
The model company has always sold it as a tool, giving users the most powerful digital brain and letting them find a way to fit it into their bodies.
The result was that most companies, after two years of pretending, still had their brains on the table, but their bodies remained completely still.
II. Palantir's Legacy
The company that truly turned FDE into a profession was not OpenAI, but Palantir Technologies.
This mysterious big data unicorn, founded by Silicon Valley godfather Peter Thiel and which helped the U.S. military kill Osama bin Laden, has been ridiculed in Silicon Valley for fifteen years.
The reason lies in its overly asset-heavy business model. Instead of selling standardized software, it sends engineers to clients' sites and stays there for as long as six months. VCs have labeled it: a consulting firm disguised as a software company.
In Silicon Valley's hierarchy of disdain, SaaS is high-end, while projects built on sheer numbers are low-end. Palantir stands at the very bottom of this hierarchy.
In 2011, Palantir discovered a recurring problem when selling data software to government and defense agencies: customers who bought the software simply didn't know how to use it.
But this problem changed everything. The traditional model of sales gathering requirements and engineers developing remotely completely failed when faced with highly confidential and extremely complex clients. The clients themselves didn't know what they wanted; they only knew that what they had wasn't working properly.
Palantir's approach wasn't to provide better instruction manuals; instead, they sent their engineers directly to their clients' sites. They went to the CIA, energy companies, and banks. The engineers sat alongside the clients, observing how they worked, studying data flows, understanding organizational structures, and then modifying software, processes, and even work methods.
This model was never replicated on a large scale in the era of standardized software. In the past, the process was defined by the product, and if customers were not satisfied, it was because the training was insufficient.
The era of large-scale models completely shattered this logic. AI has no standard usage; its limitations depend entirely on how private data is accessed, workflows are designed, and it is implemented within the organization. Each company's siloed systems are completely different, and generic products simply cannot solve the deep-water problems of customization.
As a result, Palantir's methodology, which had been refined over more than a decade, suddenly became the textbook for the entire industry.
Today, OpenAI is beginning to replicate this model, which essentially means acknowledging that AI has transformed from a software development problem into an organizational evolution problem.
III. Within a month, three industry giants made the same judgment.
If Palantir merely set an example for the industry, then in May 2026, the three leading giants in the global AI field simultaneously used real money to complete a collective conspiracy for application implementation.
On May 4, Anthropic, together with Blackstone, Goldman Sachs, Hellman & Friedman and several global asset management institutions, launched a joint venture with a total committed capital of $1.5 billion. Its core business is to help companies deploy the Claude big model.
Following this, on May 11, OpenAI officially announced the establishment of an independent deployment subsidiary, Deployment Company (DeployCo), with a total initial investment of over $4 billion and a total of 19 institutions in the partnership, including private equity investors such as TPG and Bain Capital, as well as consulting integrators such as McKinsey and Accenture.
OpenAI has simultaneously acquired Tomoro, an AI field consulting firm. Following the acquisition, Tomoro will provide DeployCo with approximately 150 frontline deployment engineers. Tomoro's existing clients include Tesco, Virgin Atlantic, Red Bull, and Supercell.
Less than two weeks later, Google Cloud CEO Thomas Kurian publicly announced on LinkedIn a large-scale recruitment drive for FDEs (Field Application Developers). Google Cloud is offering more than 1,500 AI-related positions, with FDEs being the core recruitment category.
Three of the world's top AI companies did the same thing at the same time: instead of releasing a more powerful model, they established entities specifically to help businesses implement AI.
This is a signal that deserves more attention than any model release.
OpenAI COO Brad Lightcap even said the following:
While AI systems designed for individuals are already quite powerful, we haven't yet seen AI truly penetrate enterprise business processes. Enterprises are complex organizations with fragmented systems, numerous compliance constraints, and cumbersome legacy processes; the biggest challenge at present is integrating AI into the core business processes upon which enterprises rely for operation.
Simply put, the model is good enough. The problem lies within the company and organization.
It is precisely because they have seen through this that OpenAI and others spare no expense in acquiring disciples of Accenture and McKinsey, upgrading them in batches into frontline FDEs.
This multi-billion dollar talent war has directly drained the underlying assets of the traditional consulting and IT implementation industry, and has also ushered in a revolution in large-scale model delivery.
Fourth, the ultimate goal of selling tools is to sell results.
Many people believe that AI will destroy the consulting industry. McKinsey is finished, Accenture is finished, and large IT implementation companies are finished.
The opposite happened; AI has revitalized the consulting industry.
But behind this lies a deeper change: the business model of the entire software industry is undergoing its biggest shift in the past two decades.
This is precisely the survival principle that Palantir has developed over a decade ago: Don't sell software. Deploy outcomes.
This is a fundamental transformation. In the past, Microsoft sold Office, Salesforce sold CRM, and Adobe sold suites; they were all delivering tools, and whether you used them well or not was your problem. Today, what OpenAI and Anthropic are doing is sending their own people into the customer's company and delivering results.
FDE stands for Result Deliverant. They research the organization, the processes, the data, and ultimately output a system that actually runs in a production environment, not just a pretty demo.
In the past, consultants delivered PowerPoint presentations, while FDEs delivered agents. Consultants used to offer advice, while FDEs provided code. The essence is the same: helping companies solve the problem of how to work more efficiently; only the deliverables have changed.
This is why Anthropic's FDE recruitment has a strange requirement: maintain a low sense of self and a collaborative attitude.
This is the most difficult aspect of engineering culture: it requires both sufficient technical depth to solve any problem on-site and the ability to set aside any pretense of knowledge in front of clients and patiently understand why they do not trust the output of AI.
An annual salary of $300,000 to $500,000 is not because FDE has stronger technical skills, but because a qualified FDE can replace four people: a product manager, a technical architect, a project manager, and an AI engineer.
On the front lines of delivery, an FDE is an army.
V. The biggest obstacle to the implementation of AI has never been technology.
The vast majority of AI projects that fail nowadays are not due to technical failures, but rather organizational failures.
Even the world's top financial empires and retail giants are not immune to this.
Goldman Sachs encountered a classic mid-level compliance defense when advancing its AI migration. Its technology department had developed an AI auditing system that could automatically generate analyst reports and conduct initial reviews of IPO compliance documents.
But when the system was ready to be deployed to the production environment, the middle and senior managers in the risk control and compliance departments jointly pressed the pause button. They submitted a thick inquiry report to management, arguing that if the "illusion" of the grand model appeared in the listing documents, who would be held responsible for the potential billions of dollars in fines?
No matter how beautiful the technical prototype was, the project was stuck for half a year because it could not overcome the deep-rooted culture of absolution within the organization. It was only after the FDE team intervened and redefined the boundaries of rights and responsibilities in human-machine collaboration that it barely made it through.
If Goldman Sachs was hampered by a conflict of power and responsibility, then the famous early setbacks of American retail giants Target and Palantir were due to the walls of organizational interests and culture.
At the time, Palantir sent a large FDE team to Target to try to reconstruct its supply chain and inventory forecasts, which generated hundreds of billions of dollars in annual revenue, using data models.
However, Target's most powerful team of senior buyers was extremely resistant, believing that their decades of fashion acumen should not be bowed to an algorithm. Middle management dragged their feet on data interfaces, while frontline employees deliberately ignored the system's replenishment orders. This multi-million dollar technological overhaul ultimately ended in a disastrous failure due to the power struggle between people and machines within the organization, with Target unilaterally breaking the contract.
The code is flawless, but the project just won't move. This is the most realistic scenario: technology only accounts for 20%, the remaining 80% is all about the organization's internal power dynamics, allocation of responsibilities, and historical baggage.
For example, a bank's loan approval process is backed by decades of established power and responsibility allocations and regulatory requirements. A hospital's scheduling system is linked to the profit distribution of all departments. A factory's quality control process is connected to supplier contracts and quality insurance.
These will not change automatically because of a GPT account.
These obstacles cannot be solved by an engineer who only understands technology. What is needed is someone who can think from both technical and organizational perspectives.
Therefore, what FDE is really doing is not just deploying AI, but more importantly, helping organizations complete the AI migration. If the IT department was responsible for digitizing paper-based processes in the past two decades, then in the next decade, FDE will be responsible for AI-enabled digitized processes.
This is the next stage of the same thing.
A note outside the main page:
As models become cheaper, computing power becomes cheaper, and agents become cheaper.
The truly expensive thing is starting to become another kind of capability: understanding organizations, transforming processes, and driving change.
This is why FDE became so popular.
It's not that this position is particularly important; the essence is that the entire AI industry has finally acknowledged one thing:
The most difficult part of a technological revolution has never been technology.
It is people.
