Abstract

In the U.S. IT industry, "FDE (Forward Deployed Engineer)" has emerged as the most sought-after job role of 2026. This role—embedding within client companies' offices to integrate AI models into their actual business systems—was invented by data analytics company Palantir in the early 2010s, and is now being aggressively hired by OpenAI, Anthropic, Google, and Salesforce in a competitive race. This article organizes the job responsibilities, annual compensation, and differing approaches of each company in a way that is accessible even to beginners, and then concretely examines the actual systems they have built and the business improvements they have delivered at aircraft factories and financial institutions. Furthermore, drawing on primary sources, it explores from multiple angles how Silicon Valley venture capital (VC) firms are interpreting this trend, what happened in May 2026, and what will be measured in 2027–2028.

Note that the yen conversions in this article are calculated based on the mid-May 2026 exchange rate of approximately 158 yen to the dollar (the same applies hereafter).

What is an FDE? The "Engineer Who Lives at the Client Site," Invented by Palantir in the Early 2010s

An FDE refers to an engineer who embeds deeply into the client company's workplace—rather than staying in their own firm's development office—and builds complex software products (in recent years, AI models in particular) to the point where they "actually function in real-world operations." Joe Schmidt, a partner at the major U.S. VC firm a16z (Andreessen Horowitz), explains the need for this role with the analogy: "Companies buying AI are like a grandmother who has just gotten an iPhone. She desperately wants to use it, but unless someone sets it up for her, she can't make it work." The FDE is precisely that "person who sets it up"—observing the customer's operations, defining requirements, writing production-quality code, debugging broken parts, and staying on-site until that deployment actually moves business performance metrics.

What beginners should grasp first here is that an FDE differs from both a "consultant" and an "ordinary software engineer." A consultant advises, leaves behind a proposal, and departs. By contrast, an FDE rolls up their own sleeves to build the system and accompanies the customer over the long term. On the other hand, while an ordinary software engineer builds "one feature for many customers," the FDE conversely builds "many features for one customer." Palantir has expressed this with the slogan "one customer, many capabilities." While being an engineer, they understand the customer's operations more deeply than anyone else and can speak both the technical and business languages—this is why they are called "a third option between engineer and consultant."

It was Palantir that invented this role. In the early 2010s, the company faced a constraint that its main customers at the time—U.S. intelligence agencies—"could not openly share what they needed." If they couldn't ask the customer for their requirements, then the only option was to place engineers inside the customer's environment and have them learn through observation and experimentation—and thus the idea of "forward-deploying" engineers into the field was born. According to a16z partner Tom Hollands, around 2011 Palantir renamed the role—previously called "solutions engineer" or "integration engineer"—to "Forward Deployed Engineer," and internally gave them the designation "Delta." Palantir built a structure combining an "Echo" team responsible for industry knowledge with a "Delta" team specialized in execution, and until around 2016, FDEs reportedly outnumbered ordinary software engineers.

Why is this role from over a decade ago now in the spotlight? Behind it lies the "last one-mile problem." The MIT NANDA Initiative's 2025 report "The GenAI Divide: State of AI in Business 2025" pointed out that while companies have poured $30–40 billion (approximately ¥4.7–6.3 trillion) into generative AI, 95% of organizations have not obtained measurable business returns. The failure is not of the models themselves; rather, AI fails the moment it meets "messy workflows, legacy infrastructure, and fragmented data"—and FDEs are needed as the specialists who bridge this final gap.

Job Openings Up 800%, Salaries Range from $200K to $1M — The Overheating FDE Talent Market

The FDE job market has literally exploded since 2025. According to a report by the UK's Financial Times (FT), hiring interest in FDE positions has increased by 800% since January 2025. On its corporate site, a16z notes that FDE-related job openings have "grown by 800–1,000% this year alone." The US's Fast Company reported that the CEOs of both Google and Box (a cloud company) jointly called the FDE "the most in-demand job in the tech industry for 2026."

Annual compensation levels are also high. According to multiple articles compiling data registered on career media outlets and levels.fyi, the average total compensation (base salary + bonus + equity) for an FDE in the US is approximately $238,000 (about ¥37.6 million), with a typical range of $205,000 to $486,000 (about ¥32.4 million to ¥76.8 million), and at the staff level, examples exceeding $630,000 (about ¥99.5 million) have been reported. Limiting the scope to Palantir's Forward Deployed Software Engineer (FDSE), registered data on levels.fyi shows an annual range of $171,000 to $415,000 (about ¥27 million to ¥65.6 million), with a median of around $215,000 (about ¥34 million). Some reports indicate that FDE packages at OpenAI and Anthropic reach a scale of $350,000 to $550,000 (about ¥55 million to ¥87 million). These levels exceed those of typical software engineers, and the compensation tables for Cloud (cloud business) divisions are said to shape the market's upper limit. The reason for the high compensation is simple: this job is directly tied to enterprise revenue.

The career path is also distinctive. While Palantir hires FDEs with roughly one year of experience after graduating from college, the fintech Ramp requires more than five years for senior positions. FDEs may eventually move into core product development at headquarters, or they may take the path of founding a startup, leveraging the business knowledge gained in the field. In fact, startups (such as Serval) have even emerged that position the FDE role as a training ground for "future founders" and hire prospective entrepreneurs as FDEs. The list of companies seriously building out their FDE function is endless—OpenAI, Anthropic, Palantir, Salesforce, Google, Databricks, Cohere, Ramp, Rippling, Intercom—and emerging players in voice AI such as ElevenLabs, Deepgram, and Vapi are entering the field one after another as well.

"Similar but Distinct" FDEs Across Companies—Palantir, OpenAI, Anthropic, Salesforce, Google

Even when companies hang out the same "FDE" sign, the actual practice at each firm differs in subtle but decisive ways.

The originator, Palantir, gives its FDSEs an "already-built playground" in the form of its own platforms (Foundry, Gotham, AIP) and has them configure it to tackle customers' toughest problems. According to the company's blog post "A Day in the Life," FDSEs work in small teams that own high-difficulty projects end-to-end, spending roughly 25% of their working hours at customer sites. Abhishek Singh, a partner at the consulting research firm Everest Group, describes Palantir as "a category of one," analyzing that it is a rare company that excels across all three areas: "product engineering," "embedded execution in the field," and "the reliability to operate mission-critical systems from day one."

Heading OpenAI's FDE organization is Colin Jarvis (formerly head of Solution Architecture). According to his remarks in public talks, the team launched with two people and rapidly grew to several dozen in just over a year (he has cited a scaling target of 52 people). According to Gergely Orosz, who runs the engineering newsletter "The Pragmatic Engineer," OpenAI's FDEs are distributed across eight cities on three continents and operate in three phases—initial scoping (a few days on-site), validation (building evaluation metrics), and delivery—coordinating biweekly with the research team to contribute to product development such as the Agents SDK. Jarvis captures the stance of sublimating field challenges into generalizable product insights with the distinctive phrase "eating pain and excreting product."

At Anthropic, FDEs belong to the "Applied AI" team and are embedded directly with strategic customers to build production applications using Claude models. According to job postings, their deliverables include technical artifacts used in production workflows, such as MCP (Model Context Protocol) servers, sub-agents, and agent skills. More than three years of experience in technical, customer-facing roles is required, and the role places weight on articulating reusable deployment patterns and feeding them back to product and engineering teams.

Salesforce stands out in scale. As of March 2026, its FDE team had reached at least 1,000 people and become the core function supporting customer adoption of its agent platform "Agentforce." The company has also launched a "Forward Deployed Engineering Partner Network" that brings in external partners (note: the "1,000 new-graduate and intern hires" announced by CEO Marc Benioff in April is a separate initiative aimed at AI product development and should be understood as distinct from FDE hiring). According to a May 12 report by the U.S. outlet The Information, Google is hiring FDEs in the hundreds and stationing them inside customer organizations to convert its own frontier models into working production systems. With Cloud now Google's most important growth axis, the calculation is that having talent who can sit in Fortune 500 offices and implement is the surest way to shorten the sales cycle.

Thus the emphases diverge: Palantir takes a vertically integrated approach of "deploying our own stack and our own people directly into production"; OpenAI and Anthropic take a research-driven approach of "feeding field learnings back into the product"; and Salesforce takes an ecosystem approach of "mobilizing partners as well." Some companies, like the AI agent management firm Lindy, expect FDEs to play the role of "the customer's technical consultant," and the line between engineer and consultant wavers from company to company.

What We Are Building at Client Sites (1) — An Aircraft Factory and "From Gravel Roads to Paved Roads"

This is where the heart of this article lies. What kinds of systems, at what scale, are FDEs actually building at customer sites, and what business improvements are they achieving?

The most extensively disclosed case is Palantir's aircraft manufacturing engagement. Palantir dispatched FDEs to Airbus's Hamburg and Toulouse plants in Europe, where—inside secure facilities physically cut off (air-gapped) from external networks—they tackled the final assembly line itself. At the time, Airbus was struggling to ramp up A350 production, and "visibility" was failing to take hold across the complex processes of manufacturing and the supply chain. The FDEs built, on top of Palantir's Foundry, a system that provides integrated visibility into manufacturing processes, parts inventory, and scheduling. According to figures Palantir publishes on its own site, this effort served as the principal driver behind a 33% increase in the A350 production rate.

What deserves attention is what came afterward. The bespoke solution, originally crafted to solve Airbus's urgent problem, expanded internally into more than 20 use cases spanning supply chain management, scheduling, and finance, and was ultimately generalized into "Skywise," a common platform for the aviation industry. Skywise now connects more than 100 airlines, and Palantir states that the platform generates more than $850 million (approximately ¥134 billion) in annual revenue opportunities and delivers more than $1.7 billion (approximately ¥269 billion) in annual cost savings. This is the true essence of the FDE model. In the FDE industry, this structure is called "gravel roads and paved roads." FDEs quickly lay down rough, bespoke solutions (gravel roads) for individual customers, while the headquarters' core engineering team observes them and reshapes them into standardized product capabilities (paved roads) that work for many customers.

Another archetypal Palantir example—one that comes up even in FDE job interviews—is "shortening 911 (emergency call) response times." A major city wants to integrate 911 call data, traffic data, and ambulance GPS data to reduce dispatch times—binding such fragmented data into a single working system is precisely the FDE's job. On Palantir's blog, an FDSE who worked on a COVID-19 response project recalls bringing a meaningful solution into production "within days" and witnessing the customer's decision-making genuinely improve in real time. Singh of Everest Group calls Palantir's approach a "boot camp" style of finishing a working application in days using actual customer data.

These methods are also reflected in the numbers. In Palantir's Q1 2026 earnings, announced on May 4, 2026, revenue reached $1.63 billion (approximately ¥257.5 billion), up 85% year over year; U.S. revenue grew 104%; and U.S. commercial revenue hit $595 million (approximately ¥94 billion), up 133%. In that single quarter, the company secured 47 contracts worth more than $10 million each, and net income swelled roughly fourfold to about $870.5 million (approximately ¥137.5 billion). FDEs pour into companies that have just signed AIP contracts and, before the pilot period expires, wire ontologies (knowledge graphs) into "messy real-world data"—this rotation is what underpins Palantir's rapid growth.

What We're Building at Client Sites, Part 2 — On the Ground in Finance, Semiconductors, and Call Centers

FDEs at AI labs are also beginning to deliver concrete results. The OpenAI FDE case studies cataloged in ZenML, a case database for LLMOps (Large Language Model Operations), describe the scale and impact of the systems they have built in relatively rich detail.

Emblematic is the engagement with financial giant Morgan Stanley. The firm was the first enterprise customer to put GPT-4 into production in 2023, and OpenAI's FDEs built a "Wealth Management Research Assistance System" for advisors serving high-net-worth clients, enabling them to search and summarize vast volumes of internal research. The technical pipeline itself was assembled in 6–8 weeks, with the following four months devoted to piloting, iteration, and trust building. As a result, 98% of wealth advisors are reportedly using the system, and the utilization of research reports has tripled.

Activity is also brisk on the manufacturing and semiconductor front. At one European semiconductor company, engineers were spending 70–80% of their working hours on bug fixes and compatibility maintenance. OpenAI's FDEs were embedded on site for several weeks, extending the coding-assistance model "Codex" for the company's domain and building agents that investigate and triage (prioritize) bugs and perform automated fixes. They achieved a 20–30% efficiency improvement in the initial departments, and the company is said to have set a 50% target company-wide. At an automaker in the Asia-Pacific region, every time supply-chain disruptions such as tariff changes occurred, each team was forced to coordinate manually over phone calls and meetings. The FDE team built a data layer and APIs that allow LLMs to orchestrate (coordinate processing) across systems without moving data. At payments giant Klarna, the practice of hand-writing prompts for each of more than 400 internal policies had hit its limit, so the FDEs created a framework to parameterize the instructions, which was later productized into OpenAI's internal tool "Swarm" and further into the Agent SDK. The customer support engagement at U.S. mobile carrier T-Mobile is said to be 10 times Klarna in terms of volume and policy complexity, and in the engagement with agricultural-machinery giant John Deere, FDEs built—in time for the planting season and on a short schedule—a system that delivers precise weed-control technology to farmers in order to reduce pesticide use. OpenAI's FDEs are characterized as targeting "problems with value ranging from tens of millions of dollars to the low billions."

What these engagements share is a technical philosophy: "Build as deterministically as possible, and use LLMs only where their probabilistic nature creates value." Jarvis admits that their biggest failure was "premature generalization," and says the key to success was first solving a specific customer's problem "extremely deeply."

Anthropic is also stacking up concrete examples in the financial sector. In early May 2026, financial-technology giant FIS announced its "Financial Crimes AI Agent." It compresses anti-money-laundering investigations "from hours to minutes," automatically collecting evidence across a bank's core systems, and was co-designed with Anthropic's Applied AI team and FDEs embedded at FIS. The first deployments are at Canada's Bank of Montreal (BMO) and the U.S.'s Amalgamated Bank. U.S. media outlet CIO.com has pointed out that, paradoxically, such cases have exposed the fact that "FDEs themselves are the new bottleneck." After all, deploying frontier AI into heavily regulated financial services ultimately requires a great deal of human expertise.

A Silicon Valley VC's Perspective: a16z Argues for the "Palantir-ification of Everything"

How do Silicon Valley VCs perceive this FDE boom? The firm publishing the most incisive commentary is a16z.

In an essay titled "Trading Margin for Moat," a16z partner Joe Schmidt unequivocally declared the FDE "the hottest job in startups." His argument runs as follows: AI startups should stop chasing the high gross margins of product-led growth (PLG) and instead, through professional services, become the very "system of work" for their customers. The idea is to sacrifice gross margin in the early days in order to build a "moat" of deep integration into customer operations. Against the prevailing tendency to glorify only the PLG ideal, Schmidt confronts readers with the historical fact that the combined value of SaaS giants like Salesforce, ServiceNow, and Workday—companies that embraced implementation-heavy models—overwhelmingly exceeds that of the top PLG firms. According to data he cites, 22 of OpenAI's 311 public job postings (roughly 7%) are already for FDE or solutions engineer roles.

But there is also a more cautious view within a16z. In "The Palantirization of Everything," fellow partner Marc Andrusko sounded an alarm about the casual imitation of the FDE model. FDE-related job postings have grown 800–1,000% this year, but few startups can, like Palantir, retain "hundreds of people who combine technical excellence with customer-facing skill." Andrusko predicts that many companies will fail to become a "one-of-a-kind platform company" like Palantir and will instead degenerate into high-cost service firms—merely "an Accenture for X industry dressed up with a pretty front end." His vision of the future is one in which countless "$10-million-scale startups" pitching the same story crowd in and collide with one another. He also points to the fact that Palantir trades at a stock valuation of 77x next-year revenue, underscoring its rarity. In a separate piece, a16z's Tom Hollands analyzed the very label "FDE" as a prime example of "title arbitrage," declaring that "Palantir owns the word FDE"—meaning whoever owns the language gets a head start in the hiring market.

VC interest is not confined to a16z. In an essay titled "Services: The New Software," Sequoia Capital partner Julien Bek pointed out that "for every dollar spent on software, up to six dollars are spent on services," and presented the thesis that AI-native upstarts will seize the market faster than legacy companies can transform themselves into AI-first players. Even more emblematic is the existence of a fund called "Forward Deployed Venture Capital." Founded in 2022 by Mark Scianna—who spent 11 years as an FDE at Palantir and was a founding member of its defense team—this defense and national-security-focused VC literally puts "forward deployed" in its name. The fund closed a $45 million (approximately ¥7.1 billion) vehicle in 2025. The job title "FDE" is now being used even in investors' self-definitions.

"Consulting in Disguise"? — Concerns Over Overheating and Media Controversy

Alongside the enthusiasm, the controversy surrounding FDEs has also deepened. The most frequently repeated question is, "Isn't this ultimately consulting in disguise?" The jibe "McKinsey with a code editor" is also circulating in the industry.

The standard industry rebuttal is that "consultants advise, FDEs build" and "consultants leave behind a one-off recommendation, while FDEs keep building alongside the customer over the long term." That said, it is acknowledged as fact that this boundary is blurry. A more substantive criticism lies in the accounting argument. If FDE work is billed to the customer, it is consulting revenue; if not billed, it is a customer success or support expense. Either way, it is not R&D but is classified as cost of goods sold (COGS). FDE costs are booked as service expenses, and the corresponding compensation goes into service revenue. To call that "annual recurring revenue (ARR)" is to deceive both investors and oneself—such criticisms have been coming one after another from commentators who value SaaS financial discipline. Palantir put the same approach into practice 20 years ago and was mercilessly criticized as "not scalable," "just consultants," and "not a real software company." History is repeating itself.

Alex Coqueiro, an analyst at the research firm Gartner, has issued an even more pointed warning. According to reporting by CIO.com, he predicted that "70% of enterprises will be forced to abandon agentic AI solutions obtained through FDE-led initiatives, citing vendors' high costs and a lack of in-house skills." If the FDE hours injected do not decrease even as the solution matures, that is a sign of "dependence, not capability." Nik Kale of the Coalition for Secure AI is even more scathing. Regarding the FIS case, Kale stated, "This is an admission that frontier AI is not yet a product. CIOs thought they were buying software, but what they are actually buying is a professional services contract."

There are also workload issues on the side of those working as FDEs themselves. Travel that amounts to 25–50% of working hours, on-site embedment at locations such as factories and air-gapped facilities, and a high-intensity work style driven by tight deadlines are prone to inducing burnout, and attrition has been observed. They are caught between customers demanding "fast and custom" and product teams demanding "maintainable designs with clear scope," and are forced into intense cross-industry context switching (shifting domains of responsibility). The more they deliver, the more "next hard problems" come pouring in, and they are required to be on constant standby for strategic customers—this "always on" pressure is the reality behind the high compensation.

The Tectonic Shift of May 2026: The "PE Fund-Style" Strategy Launched by OpenAI and Anthropic

And in May 2026, the movement surrounding FDEs entered an even greater structural transformation. OpenAI and Anthropic, one after the other, launched "FDE-specialized business entities" that brought in private equity (PE).

On May 11, 2026, OpenAI announced the launch of "OpenAI Deployment Company." This new company, in which OpenAI holds a majority stake and control, has 19 investment firms, consulting companies, and system integrators participating, and gathered an initial investment of over $4 billion (approximately ¥632 billion). U.S. outlets such as Axios have reported the new company's valuation at $14 billion (approximately ¥2.21 trillion) (note that some pre-launch reports described it as a vehicle on the order of $10 billion in total, so there is variation in how the scale is expressed across reports). The lead investor is TPG, with Advent International, Bain Capital, and Brookfield named as co-founding partners. OpenAI itself will contribute up to $1.5 billion (approximately ¥237 billion; $500 million = approximately ¥79 billion at launch plus a $1 billion = approximately ¥158 billion option). According to multiple reports, OpenAI guaranteed PE investors a minimum return of 17.5% per annum over five years, while simultaneously setting a profit cap. Concurrent with its launch, the new company acquired Tomoro, an applied AI consulting firm based in Edinburgh and London, UK (with clients including Tesco and Virgin Atlantic), securing approximately 150 experienced FDEs and deployment specialists from day one.

Anthropic, too, on May 4, 2026, announced the establishment of a new enterprise AI services company with Blackstone, Hellman & Friedman, and Goldman Sachs as founding partners (General Atlantic, Leonard Green, Apollo Global Management, GIC, and Sequoia Capital also joined as investors). This new company aims to deliver Claude to mid-sized companies that lack the in-house resources to deploy frontier AI on their own—regional banks, manufacturers, regional healthcare systems, and so on. Anthropic's Applied AI engineers will work alongside the new company's engineering team to identify where Claude has the greatest impact, build custom solutions, and support customers over the long term. Some details, such as the new company's name and total investment amount, were not disclosed as of the May 4 announcement. The company will join the "Claude Partner Network," which also includes consulting firms such as Accenture, Deloitte, and PwC. Anthropic has stated it will make an initial investment of $100 million (approximately ¥15.8 billion) in this network in 2026.

These two cases are not isolated events. On February 23, 2026, OpenAI formed the "Frontier Alliance" with McKinsey, BCG (Boston Consulting Group), Accenture, and Capgemini, getting a head start on a system that pairs FDEs with major consulting firms. Accenture launched the "Microsoft Forward Deployed Engineering Practice" in 2026, and in April 2026, EY opened an FDE practice in the UK and Ireland—reportedly the first case of a major consulting firm formally adopting this model. EPAM partnered with Anthropic. FDEs are no longer merely a hiring initiative of a single company; they are transforming into an "industrial structure" in which AI labs, PE funds, and global consultancies move as one.

Japan's FDE — Similar to Yet Distinct from the U.S.-Style "On-Site Client Embedding"

This trend is also spreading to Japan. However, since Japan has long had business practices known as "client-site stationing" and "SES (Systems Engineering Service)," it is essential to correctly understand how these differ from the American-style FDE. In 2026, Nikkei xTECH published an article titled "American-style 'FDE' Is Similar to but Distinct from Japan's Client-Site Stationing," cautioning against conflating the two. Whereas Japan's client-site stationing is a contractual arrangement that provides labor measured in "person-months" (ningetsu), the American-style FDE differs greatly in character: FDEs are responsible for outcomes (whether the customer has actually succeeded) and personally write production-quality product code.

The Japanese market is also beginning to move on the hiring front. According to tallies by several recruiting media outlets, as of spring 2026, approximately 26 FDE job openings from domestic Japanese companies and about 9 from foreign-affiliated firms were observed—roughly 35 in total. In addition to global players such as Palantir's Japan subsidiary and SB OpenAI Japan, the joint venture between SoftBank and OpenAI, domestic AI and SaaS companies including ExaWizards, LayerX, AI Shift, and ANDPAD have begun recruiting. AI solutions company Headwaters announced in a press release its policy of placing FDEs at the core of its hiring. Because the job category is still in the process of being established, annual salary levels—as a rough guide based on each company's job listings—are said to range from about 7 million to 15 million yen, and Salesforce has reportedly posted a range of 8 million to 30 million yen on the word-of-mouth site OpenWork. While FDE positions at global companies require English, many positions at domestic AI companies and DX consultancies can be handled in Japanese alone. If Japanese companies casually operate this role as a "convenient on-site SE," they will not achieve the American-style results—this is a point that experts unanimously emphasize.

Future Outlook――The "Next Move" to Be Measured in 2027–2028

When and in what form will the next moves around FDEs be measured? Integrating the perspectives of Silicon Valley VCs and analysts, several predictive axes emerge.

First, 2027 is likely to become the "year of attrition." Gartner predicts that more than 40% of agentic AI projects will be canceled by the end of 2027 due to escalating costs, unclear business value, and inadequate risk management. Among FDE-led initiatives, those unable to demonstrate clear ROI (return on investment) are expected to be cleared out en masse during this period. Conversely, the FDE organizations that survive will be the minority that has "proven its capabilities."

Second, 2028 will become the "year of scale expansion." Gartner predicts that by 2028, at least 15% of routine business decisions will be made autonomously by agentic AI, and 33% of enterprise software will incorporate agentic AI. Some analysts estimate that, accompanying the expansion of AI startups, the number of FDE job postings could reach five times the current level by 2028. The picture is one in which, beyond surviving the attrition, full-scale quantitative expansion awaits.

Third, there is a noteworthy paradox: the FDE role itself is beginning to be automated by AI. Palantir has already made generally available an AIP-native agent called "AI FDE," which operates Foundry through natural-language dialogue and handles data connection, transformation, ontology construction, and app development. Gartner, too, referenced FDEs in its 2026 strategic technology trends, depicting human FDEs leveraging AI-native development platforms to undertake strategic analysis, app discovery, agent platform construction, and the installation of security and governance guardrails. "The specialized profession that delivers AI to the field" is being amplified by that very "AI"—and some portions of it will eventually be replaced. This recursive structure will become a focal point of debate going forward.

In the short term, the first litmus test will be whether, between Q2 and Q3 of 2026, the OpenAI Deployment Company and Anthropic's new company publish their first customer outcomes, or whether follow-on "PE fund–style AI deployment vehicles" are newly announced by other labs or consultancies. Box CEO Aaron Levie stated that as AI agents reshape the fundamental workflows of organizations, "the need and opportunity for professional services and FDEs to deploy agents is enormous," predicting the arrival of a massive "consulting gold rush." FDEs are themselves the very question of who will own the "last one mile" of the AI era, and the answer will be made visible from 2027 through 2028 as two waves: attrition and expansion.

Sources