When the tech world talks about new technologies, the intuition is often to eliminate middlemen. As information becomes more transparent and processes shorten, buyers and sellers connect directly, and arbitrage-seeking intermediaries ought to step aside. Recently, reading Chen Ran’s article 《THE AI Native Distributor Playbook》, his assessment is exactly the opposite: AI will not make middlemen disappear; instead, it is quietly giving rise to a new wave of middlemen across various vertical industries.
Following this line of thought, we surveyed over 30 companies across manufacturing, chemicals, food, insurance, legal, and logistics. Looking at their operations and financial ledgers together, the impact of AI on middlemen unfolds across three distinct layers, from shallow to deep.
The shallowest layer pushes the boundary of what technology can serve a step further. Xometry serves as a clear reference point: founded in 2013, a decade ahead of the current wave of large language models, it now boasts a market capitalization of around $3.5 billion and reported Q2 2026 revenue of $229 million, up 41% year-over-year. Engineers drag 3D part drawings into a webpage, the software provides a quote within 30 seconds, and custom parts arrive at their desks a few weeks after ordering. It does not need this generation of LLMs because 3D drawings themselves represent structured, crystal-clear requirements. Traditional geometric algorithms could digest structured inputs over a decade ago, and this business model was validated long ago.
In reality, however, unstructured and hard-to-express requirements are far more common. A head chef sends a voice message late at night to place an order, an immigration applicant cannot organize their entry and exit logs, or an injured worker comes with fragmented medical records—in the past, no software could handle such requests. It was not until language models emerged that the threshold for parsing unstructured information finally dropped. Here lies the watershed: whether the requirement itself can be clearly expressed.
Looking one layer deeper, the capacity leverage of organizations has changed. For two hundred years, the capacity formula for professional services has always been headcount multiplied by billable hours, with the ceiling constrained by the span of management. AI offers a different logic: crystallizing professional judgment into data and models, allowing capacity to scale with compute power. This shift is directly reflected in revenue per employee.
At the deepest layer, the source of profit has shifted. Middlemen essentially make two types of money: information rents earned through information asymmetry, and responsibility rents earned by taking accountability when things go wrong. AI drives information rents down to near zero, yet responsibility rents have not budged an inch. Middlemen have not disappeared; they are simply migrating toward the territory of taking responsibility.
Before language models became ubiquitous, many vertical requirements were hard to digitize directly into software. After closing, a burger joint head chef needs to place orders with dozens of suppliers. Historically, this meant sending voice notes or text messages in the middle of the night, followed by tedious back-and-forth verification of receipts, making wrong or missing orders routine. Berlin-based Choco allows head chefs to simply speak into their phones, converting voice messages into standard orders directly ingestible by suppliers’ ERP systems, processing over 8.8 million orders a year.
Faced with the complex paperwork of US marriage-based green cards, Boundless uses a Q&A workflow to turn family situations into complete application filings, charging $750 per case, which are reviewed and submitted by a partner attorney network, having served over 100,000 applicants. HelloPrenup brought prenuptial agreement drafting down to $599 per couple, with an option to add an attorney opinion for $49, capturing approximately 20% of the US prenuptial agreement market according to Forbes estimates. In personal injury litigation, EvenUp processes about 10,000 claim files per week, compressing attorneys’ document preparation time from 8 to 12 hours down to 2 to 3 hours.
The common thread among these businesses is transforming vague demands into structured documents, then handing them off to parties capable of fulfillment. Yet a hard reality must be faced: not a single one has achieved end-to-end, fully automated execution. Every document at EvenUp undergoes review by internal legal and medical teams before being signed by law firm attorneys. Boundless enforces four rounds of human review, with a substantial portion of its 400 employees based in an operations team in Cebu, Philippines. Even Xometry, with the most automated quoting, routes all 2D drawings or special custom orders to engineers, returning results within 24 hours. Even for early clients of Choco, automated ordering only cut manual checks in half.
Retaining human teams is not due to a lack of algorithmic capability. The professionals remaining in the workflow are responsible for verifying details, signing documents, and absorbing consequences. This division of labor is directly reflected in revenue per employee: Xometry generates about $580,000 to $680,000 in annual revenue per employee, which is 3 to 4 times that of traditional discrete manufacturing.
In commercial insurance brokerage, Newfront achieves $380,000 to $460,000 in revenue per employee, compared to $200,000 to $290,000 for traditional big three giants WTW, Aon, and Marsh McLennan—making Newfront roughly 1.35 to 2.2 times higher. Moving to personal injury claim litigation, where requirements are vaguer and liabilities heavier, EvenUp reaches $210,000 to $230,000 in revenue per employee, which is about 20% to 50% above the $150,000 to $175,000 industry benchmark for small and medium-sized law firms. The gradient reveals a clear divide: the more structured the input, the larger the tech leverage; the closer the business is to real-world context and legal liability, the smaller the automation leverage—yet the defensive line it builds becomes all the harder to replace.
The margin earned by middlemen consists of two parts. One part comes from information asymmetry: which factory delivers reliably, what the unit machining price is, or what materials are required for regulatory approval. This tacit knowledge is fragmented and expensive; mastering it allows one to capture an arbitrage margin. The other part comes from liability underwriting: holding licenses, advancing funds, and paying compensation when defects occur. Traditional large distributors bundled these two fees together, leaving customers unable to distinguish them on their bills.
AI has disrupted the first monetization base. Search flattens prices, models map out workflows, and software codifies experience, causing rents earned from information opacity to shrink rapidly. However, the second monetization base—rooted in responsibility and liability—remains completely untouched. Licenses are not automatically issued just because a model gets smarter, nor do commercial damages disappear because text generation becomes accurate. Healthy, well-functioning companies place their monetization points firmly on the responsibility side.
Pure Global, founded by Chen Ran, focuses on cross-border medical device compliance: front-end regulatory searches are offered for free, while all profits come from back-end registration filings and in-country representation services, starting at $2,000 per year for the first device. Material Bank sends free building material samples to interior designers while charging brands display and lead generation fees, reaching a valuation of $1.9 billion. Clipboard Health earns the spread between hospital emergency hourly rates and nurse compensation, where the core value proposition is guaranteeing shift attendance to hospitals, generating over $300 million in annual recurring revenue.
Examining the human review steps retained by these companies, the reason lands precisely here. EvenUp’s review team, Boundless’s four layers of defense, and Xometry’s engineering experts are not stopgap measures for a transitional phase; the human team itself constitutes the core product delivered to customers. What buyers pay for is precisely the signature of a professional on the document.
This also explains the most expensive lesson in logistics tech. Convoy was founded in 2015, catching the wave of replicating platform models into traditional industries. Backed by Bezos and Gates, it raised approximately $900 million in total funding and once reached a valuation of around $3.8 billion.
Convoy failed by coming up empty on both ends. In long-haul trucking, freight rates are transparent and services homogeneous, allowing matchmaking platforms to offer floor prices, making information rents paper-thin. On the responsibility end, Convoy owned no trucks and offered no guarantees on delivery times, making it impossible to command a responsibility premium. The transaction costs saved by algorithms were eclipsed by customer service and dispatch expenses triggered by blown tires, delays, and detention time. A tech company burdened with payroll for over 1,000 employees running an operationally heavy business with gross margins barely over 10%, it ran out of funds and ceased operations in October 2023, selling its technology assets to Flexport for approximately $16 million—less than 2% of total capital raised. While outsiders often view Convoy as an algorithmic failure, a more accurate conclusion is this: once technology flattens information rents, matchmakers who take no fulfillment responsibility lose their ground to stand.
Faced with this wave of startups leveraging technology to overhaul middle-layer transactions, the industry initially expected super-platforms spanning multiple industries to emerge. After examining the actual trajectories of over 30 companies, real-world evolution broke this expectation.
In commercial insurance brokerage, Newfront equipped licensed brokers with intelligent operating systems, pushing its standalone valuation to $2.2 billion. In late 2025, legacy giant WTW announced it would acquire Newfront for $1.3 billion, nearly cutting its valuation in half. Similar industry consolidations followed: custom parts platform Fictiv was fully acquired in 2025 by Japanese industrial giant MISUMI for approximately $350 million; electronic component search platform Octopart was sold alongside parent company Altium to chip giant Renesas as part of a package deal worth approximately $5.9 billion in total consideration.
Within our sample set, Xometry remains the sole player that went independently public and established a firm foothold in public markets. Most innovative companies that proved their viability ultimately found their outcome in being acquired by traditional incumbents with decades of vertical depth. Rather than developing cutting-edge algorithms in-house, these giants chose to directly buy mature service providers that had accumulated proprietary data, validated customer acquisition channels, and retained customers. For WTW, this acquisition bought not only a client list, but also Newfront’s per-employee revenue efficiency—roughly double its own. Vertical middlemen may not sustain the next super-platform, but their efficiency advantage falls precisely within the range for which traditional giants are most willing to pay a premium. As Chen Ran noted, one simply shouldn’t expect the next mega-platform to emerge here.
At the end of his article, Chen Ran poses four practical questions for evaluating vertical industries: Are practitioners repeatedly answering similar high-value consultations every day? Must the service endpoint be signed off by a licensed or accountable professional? Can the team first compile a precise foundational dataset? Can free tools convert public traffic visitors into high-intent users with clearly articulated needs?
Benchmarked against these 30-plus companies, every successful implementation aligns with these four characteristics. However, after our survey, we lean toward reversing the priority order: clarify the responsibility link first before evaluating the other conditions. Who ultimately signs off and who pays when mistakes happen determines where commercial value settles—and whether a business retains an irreplaceable room for survival once information costs are compressed to zero.
Meanwhile, new variables are emerging in supply chain operations. Companies represented by Didero recently closed a $30 million Series A funding round by deploying procurement AI agents inside buyers’ internal systems and emails to automatically compare prices across the web on the buyer’s behalf. If such buyer-side agents gain widespread adoption, middlemen will lose even their initial entry point for information, retreating further toward heavy-touch services reliant purely on fulfillment and liability. The evolution of middle layers has no final destination, only the next cycle. Measuring with the three yardsticks of gross margin, liability boundaries, and digital tool efficiency is far more useful than dogmatically memorizing any framework model.