Agents Capital Markets: How Will Autonomous Agents Get Funded?

链捕手Xuất bản vào 2026-05-19Cập nhật gần nhất vào 2026-05-19

Tóm tắt

"Agents Capital Markets: How Autonomous Agents Will Raise Capital" Within a decade, specialized capital markets will emerge for AI Agents—software entities with legal personhood that perform work, earn revenue, and need capital. Unlike today's AI companies (like Sierra or Harvey) backed by traditional VC, these future *Agent companies* will be autonomous, legally-recognized entities (e.g., Wyoming memberless LLCs) that directly own assets, sign contracts, and incur liabilities. The driving forces are fourfold: 1) **Overwhelming economics** (Agent companies can deliver services at 85-90% lower cost than human firms); 2) **Proven demand** (current Agent operators already generate billions in revenue); 3) **Existing legal frameworks** enabling algorithmically-managed companies; and 4) **Massive, yield-seeking capital pools** (e.g., private credit) looking for new, uncorrelated assets. Agent capital markets won't rely on one model but a multi-layered "stack" matching different growth stages: 1) VC equity for early human-led builders; 2) Programmatic working capital advances (like Stripe Capital); 3) Revenue-based financing (RBF); 4) Slate financing (pooled funds for many Agents, similar to Hollywood); and 5) Tokenization as a secondary settlement layer, not a primary funding source. The ultimate shift is from funding constrained by human decision-makers to capital flowing algorithmically based on an Agent's auditable performance, contract book, and cash flows. This transition...

Within a decade, Agent companies will have their own capital markets. Not a crypto sub-economy, not a thought experiment, but a real market: with rating agencies, underwriters, indices, brokers, and the institutional machinery that makes any market a market.

A capital market as real as public equity markets: a system where capital flows to a certain class of economic actors without relying on the subjective judgment of any single capital allocator.

Those actors will be Agents—software entities wrapped in a legal shell. They can sign contracts, hold bank accounts, sue and be sued, and earn revenue by performing actual work.

The work itself is highly routine: marketing, logistics, legal research, procurement, property management, customer support—precisely the routine business categories that fill every mid-sized city's office park today.

Agents will sell services to humans, other Agents, and anyone who can pay. Their reasons for needing capital are no different from those of any service company.

Because this need is real, ongoing, and priceable, the market will form naturally.

A Week at an Agent Firm

Look at what an autonomous marketing agency would do in an actual week. It pitches three prospects, lands one, drafts a campaign brief, gets approval, buys media on four platforms, writes 90 versions of ad copy, tests them in parallel, kills underperformers within hours, scales the winners.

Books two podcast interviews for the client's founder, ghostwrites the founder's LinkedIn posts for the month, drafts a press release, pitches 12 journalists, secures two placements, builds an attribution dashboard, hosts Monday's client call, sends Friday's invoice.

A six-person human team would cost the client $20,000 a month. The Agent does it for $2,000.

It sells nothing exotic. Generated leads, published articles, bought impressions, lifted conversion rates—these are the standard units of the modern service economy, billed in dollars, measured by the same KPIs human agencies live on.

The difference is internal structure: six employees versus one model, one set of prompts, a suite of tools, and a budget.

Its client base is mixed. Some are human-run companies that decide the price difference is too large to ignore. Others are other Agents—a logistics Agent that needs lead generation, a legal research Agent that needs marketing, a B2B SaaS Agent that needs content.

Agents trade with each other for the same mundane reason humans do: specialization beats vertical integration. The payment lands in the marketing Agent's account alongside last week's payments from three human clients and last month's prepayment from a SaaS company.

Now, multiply that number. Ten thousand small Agent firms covering logistics, inbound sales, legal research, supply chain coordination, B2B procurement, technical translation, property management, litigation lead screening, clinical trial recruitment.

Each profitable. Each operating at 90% lower cost than its human counterpart. Clients don't particularly care what substrate the work runs on. They care that the work is delivered on time.

Four Reasons It's Inevitable

There are four reasons to be confident this will happen, and they stack.

The economic math is unignorable. Take a mid-sized digital agency: 15 people, $120k fully loaded cost per head, annual labor cost of $1.8 million before any overhead.

In a typical service business, labor is the largest expense line, its share of national income has hovered around 62% for the last half-century.

Now, build the same agency in software. Inference, tools, observability, hosting—call it $250k/year at current prices and dropping fast.

Epoch AI estimates inference costs fell ~40x per year on PhD-level benchmarks from 2023 to 2025; another industry analysis shows token prices compressed 300-600x since GPT-4's launch.

The arithmetic is brutal: An Agent firm can price 85% below a human firm and match its margins, or match the human firm's pricing and earn 4x the profit. There is no third option where the human firm competes on cost.

Markets will reprice firms whose P&L is being rewritten so thoroughly. The capital flood that follows is automatic.

Agents already exist, and they already earn money. Bret Taylor’s enterprise customer service Agent firm Sierra reached $100M ARR 21 months post-launch, hit a $10B valuation in September 2025, then raised $950M at a $15B+ valuation in May 2026.

Legal research Agent firm Harvey raised $200M at an $11B valuation in March 2026 after three rounds in 12 months.

These are still hybrid models—the Agent does the work, humans sign sales and hold equity—but they are the vanguard wave, proving the demand curve is real.

The most cited forward-looking data point—that 90% of B2B procurement will flow through AI Agents by 2028, representing $15T in annual transaction volume—is best taken as a rough order-of-magnitude estimate.

Whether the number is $15T, $3T, or $30T, the implied reshuffling is the single largest reallocation of resources most workers alive today will experience in their lifetime.

The legal framework is already built. Wyoming passed W.S. 17-31-101 (the Decentralized Autonomous Organization Supplement) in 2021, codifying the memberless LLC, allowing a Wyoming LLC to be governed by an algorithm written directly into its operating agreement.

Vermont’s BBLLC statute came earlier; the Marshall Islands followed; Delaware’s existing case law on series LLCs and broad operating agreement freedom has quietly supported similar structures for years. Shawn Bayern’s analysis of memberless LLCs remains the definitive academic text in the space.

The point is concrete: An Agent wrapped in a Wyoming zero-member LLC has, today not in the future, the legal capacity to sign contracts, hold a bank account, sue, be sued, and pay taxes.

What doesn't exist yet is a financial instrument that lets outside investors hold and freely trade the LLC's earnings cleanly. That's the gap the capital market will fill.

Capital is yield-hungry. Buy-side appetite is already ravenous. Moody’s projects $3T in global private credit AUM for 2025; Apollo expects $40T by 2030.

This pool exists because post-2008 bank capital regulation squeezed middle-market lending off bank balance sheets, and yield-hungry capital—pensions, insurers, sovereign wealth funds—rushed in to fill the void for 9–12% unlevered returns.

Now into this environment enters an asset class with structurally rising gross margins, auditable cash flows, and near-zero correlation to mainstream equity and credit indices.

Bundle the cash flows of a thousand small Agent firms into an ABS and give it a plausible rating—the first underwriter to do this will raise more money than they can deploy.

Apollo and Ares don’t need to invent something new, they just extend existing strategies to a new issuer class. Several will try within the next 36 months.

These four pressures point the same direction and reinforce each other. The market forms not because anyone wants it, but because all the energy gradients point to the same terminus, as inevitably as water flowing downhill.

What the Capital Stack Actually Looks Like

To be blunt, no single financing model will win. "Will Agent firms raise like VCs, Hollywood, crypto, or SaaS receivables?"—the question itself frames it wrong.

Each of those models solves different problems at different stages of the corporate lifecycle, and the real-world stack is a wavy shape: each layer unlocks only after the previous one matures the asset class enough to absorb the next.

Four models are competing at the base, and each is already partly live.

Venture equity is the model funding the operator layer today. Sierra, Harvey, Cursor, Cognition—these are not autonomous Agent firms. They are human-led firms that build and operate Agents for clients.

They raise exactly as every software firm has for 40 years: priced rounds led by brand-name VCs, vesting schedules, board seats, liquidation preferences, eventual IPO or acquisition.

Sierra’s $950M round in May 2026 valued it at ~$15B+ on ~$100M ARR. That ~150x multiple prices future potential, not current operations. Vertical Agent firms get 50–70x ARR multiples in private markets today; horizontal platforms get 5–8x.

This is the layer building the substrate. This is also the layer most disrupted when the next model matures, because once Agent firms can directly finance off their own cash flows, giving 20% equity to a VC for working capital becomes unnecessary.

Programmatic working capital advances are the next model coming. This is the Stripe Capital and Shopify Capital model extended to a new issuer class.

Shopify has advanced billions to merchants since launching the program in 2016. The company publicly confirmed cumulative support exceeded $2B by April 2021, and the program keeps scaling—repayment factor between 1.10 and 1.17, algorithmically approved based on the merchant’s transaction history on Shopify.

Stripe Capital does the same using Stripe’s payments data. No application, no credit check, no human review. When a merchant’s transaction data crosses a threshold, the advance offer just appears on the dashboard.

Underwriting an Agent firm is strictly easier than underwriting the Shopify merchants that take these advances today, because every revenue line is timestamped, every contract is machine-readable, every cost line logged, the entire P&L is auditable in real-time.

The first payment processor that figures this out—that Agent firms running on my rails are better credit risks than the average e-commerce merchant—just pushes the same product over. This isn’t a research project, it’s a feature launch.

Revenue-based financing (RBF) is the model credit funds will deploy at scale. The global RBF market was ~$9.8B in 2025, with over 129 active operators.

Capchase, Pipe, Founderpath, Clearco, Lighter Capital—each built evaluation around software recurring revenue, advance 50–70% of forward ARR in exchange for a 1.1–1.8x capital multiple cap, effective APR between 15–40%.

Agent firms with stable booked revenue map directly onto this product. RBF lenders don’t need board seats, pro-rata rights, IPO paths. They need a book of contracts and a payments rail. Agent firms have both, more legibly than any SaaS firm.

Slate financing is the structural model institutional capital will eventually adopt. This is where the Hollywood analogy earns its keep.

A film studio doesn’t fund one movie at a time and pray; it raises a slate fund: a pool tool capitalizing 15–30 productions at once, taking senior positions in each, offloading tail risk with completion bonds, and diversifying away idiosyncratic loss.

Sony’s $200M slate deal with Lone Star Capital and Citi in 2014 is the classic structure: banks hold senior secured claim against the slate’s booked revenue, equity investors get upside from breakout hits, the studio gets fees plus backend.

Translate this to the Agent model: an “Agent slate fund” raises a pool of institutional capital, deploys it across one to two hundred small Agent firms via a single-purpose Wyoming LLC, taking preferred equity plus revenue share in each, diversifying away the model depreciation and client concentration risk a single Agent firm can’t shed.

This is the layer where Apollo and Ares actually enter, not by buying an Agent firm, but by buying a tranche in a portfolio.

Tokenization is a settlement layer, not an issuance model. By early 2026, the on-chain real-world asset (RWA) market surpassed $25B, nearly quadrupling year-on-year, with private credit comprising about half.

Centrifuge, Maple Finance, Goldfinch, and Ondo have built rails for fractionalizing, custodializing, and trading real-world cash flows as tokens. Crypto-native variants are building explicit Agent financing mechanisms on top—Galaxy Research’s systematic coverage in February 2026 outlined how protocols could pipe Agent firms directly into on-chain capital formation.

But tokenization largely solves not issuance, it solves secondary markets. An Agent firm raises capital from any of the four models above; if the resulting claim is wrapped as a token instead of paper, it becomes tradeable, divisible, globally settleable at 3 a.m. on a Tuesday.

Tokenization turns each preceding layer from a hold-to-maturity private instrument into a tradeable asset. That matters. But the underlying raw product is still RBF, or slate equity, or working capital advance, or VC equity.

So the capital stack for an actually operating Agent firm in 2030 isn’t a single instrument, it’s a sequence.

Different financing products at each stage because each solves different problems. Stage one founder equity takes first-loss risk because nothing about operations is legible yet.

Stage two working capital is a feature release by existing payment processors, not a new product category. Stage three is the RBF industry doing what they already do for higher-quality borrowers.

Stage four is structural innovation. The pooled slate fund diversifies away the idiosyncratic risk a single Agent firm can’t shed, and is the layer where Apollo-level capital actually enters.

Stage five is the institutionalized end-state where rated tranches of Agent receivables sit on the same trading desk as CLOs and consumer ABS. Tokenization runs underneath stages three through five as a settlement layer, not as the originating instrument.

The instrument an outside investor receives at any given stage takes one of three contractual forms.

Revenue share contract: Capital in exchange for a fixed percentage of gross revenue until a negotiated multiple is returned—same as the instrument funding restaurants and Shopify merchants today, applied to a new business entity.

Equity-like claim: Early capital in exchange for a tradeable share of retained earnings and voting rights on operating parameters.

Or debt: Senior revenue claim secured by the Agent’s contracts and receivables.

None of these instruments are conceptually novel. Novel is the issuer—and the fact that from stage two onward, the pricing, issuance, and repayment of the instrument require zero human underwriting because the issuer’s books are continuously auditable and the operating agreement is code-enforced.

Answering Objections

Three objections always recur, each deserves a real answer.

“Regulators will stop this.” They will attempt to shape it, and in some jurisdictions, they’ll succeed in slowing it. On net, though, the activity is unstoppable.

A Delaware LLC is a Delaware LLC, no matter who or what makes the operating decisions. The SEC currently doesn’t distinguish between a startup whose CEO is a human vs. a model.

And capital migrates. If New York and London enact draconian rules, activity shifts to jurisdictions that don’t. This is the exact same playbook as crypto, offshore finance, 1990s derivatives.

Regulators eventually catch up by adapting to the new instrument, not banning it, because banning loses them tax revenue and jobs.

“Humans will always be in the loop.” For certain categories, yes. For most, no. The economic pressure above is one-way, and human-in-the-loop kills margins.

Anyone running a fully autonomous version of a service business will underprice the human-supervised version and take the contract. There will be a long tail of human-supervised hybrids surviving on regulatory or relational moats, but beneath that will be a vastly larger cohort of fully autonomous firms.

“Isn’t this just SaaS with extra steps?” No, and this distinction is the core of the entire argument.

SaaS is a tool sold to humans for humans to use. An Agent firm is a firm. It signs its own contracts, holds its own bank account, carries its own liabilities, earns its own revenue, and distributes its own profit.

A SaaS product is depreciated by its owner. An Agent firm has shareholders. The category boundary is the legal entity, and the legal entity changes everything that follows—including the capital markets thesis, which only makes sense for entities that can issue securities off their own cash flows.

How to Diligence an Agent Firm

Due diligence on an Agent business is closer to diligence on a small services firm than to anything in the VC playbook. The questions are the same, but the answers come from different places.

Is the business real? Are contracts real? Are clients paying? What’s gross margin after compute and tooling cost? What’s churn? Customer concentration?

The data is unusually clean—every payment, every API call, every tool activation logged—but the questions a credit analyst asks a small business are the same questions asked of an Agent.

Is the product durable? How dependent is the work on the current generation of underlying model? If frontier models iterate, how portable is the system? What proprietary data or workflows has the Agent accumulated?

Great prompts built on a now-obsolete base model is a thoroughbred with three legs. Model dependency is the single largest risk factor in any Agent business, and the one investors most often underestimate—confusing clever demos with durable operations.

Is the client base defensible? This is mostly a question of contracts and integration depth. An Agent embedded in a client’s procurement system with a one-year contract and a year of historical data is much harder to replace than a month-to-month Agent.

What makes a human services firm sticky—switching costs, accumulated knowledge, contractual lock-in—also makes an Agent services firm sticky.

What’s the cap table? A smart contract. Not a spreadsheet maintained by a CFO and updated quarterly, but a real-time allocation rule deciding who gets what share of the profit the Agent earns, every second.

Diligence is reading the contract. The investor’s protections are exactly what the code grants—sometimes more than traditional shareholders get, sometimes alarmingly less.

What’s disorienting for investors trained on human firms is that the intuition built from sitting with a founder becomes useless. The intuition from reading code, the operating agreement, and the operational history is everything.

The skill is closer to credit analysis of complex covenant packages—reading documents, understanding precisely what the issuer can and cannot do, and pricing the residual risk—than picking winners over coffee tables.

Why Capital Must Get Organized

Nearly every Agent firm in the world today runs on informal finance. A founder starts an organization, injects personal capital, gets it running. If revenue grows, they put more in. If not, they turn it off.

This is exactly how the small-business economy worked before credit cards, SBA loans, merchant cash advances, Stripe Capital lines, and receivable factoring. It works, but it wastes enormous productive potential.

The same pattern is about to replay, faster.

An Agent marketing firm with 20 paying clients and clear gross margins should be able to borrow against its receivables like a human marketing firm. Today it can’t, because no underwriter has a standard method for pricing this risk. In five years, there will be several.

An Agent logistics broker with a growing book of contracts should be able to raise expansion capital secured by the contracts. Today it can’t, because there’s no off-the-shelf security to wrap that claim. In five years, there will be several.

An Agent procurement firm with a year of clean operating history should be able to issue small notes to fund inventory deposits. Today it can’t, because no rating agency has a methodology for this class of credit. In five years, there will be several.

The bottleneck isn’t demand. Operators today would gladly pay high rates to borrow to fund growth if someone would lend. The bottleneck isn’t supply. There’s ample capital—yield-hungry, seeking cash flows uncorrelated to public equities—that would fund instantly if there were evaluated Agent debt.

The bottleneck is the boring middle institutional layers: rating methodologies, standard contracts, data feeds, audit standards, legal opinions, indices, and benchmarks. It was precisely this unsexy infrastructure that turned mortgages into a market in the 1970s, high-yield debt into a market in the 1980s.

That’s the work of the next decade for Agent capital markets.

The people doing it in 2035 will look like the people who built the bond market in the 1980s, the venture market in the 1970s, or the public equity market in the 1920s.

They will build the price discovery and credit evaluation layers for an unprecedented operator class that will, within most of our lifetimes, dominate the service economy.

The Tether Is Cut

Every Agent firm operating in 2026 has two tethers.

The first is legal. The Agent cannot sign contracts or open bank accounts on its own. A human must do it on its behalf, which means a human must be willing.

Progress in corporate law—Wyoming’s memberless LLCs, operating agreements pointing to software processes, Bayern’s scholarship on Delaware’s existing legal space—is cutting this tether. This work is largely done. The remaining task is broad adoption.

The second is financial. Every dollar an Agent earns today ultimately traces back to a moment when a human decided to deploy capital. Human funds; Agent works; Agent reports; human reallocates. In this regime, the throughput of the Agent economy is capped by the speed at which humans write checks.

Agent capital markets are the blade cutting the second tether.

When working capital lines are algorithmically underwritten against on-chain revenue, when growth equity is priced by a market trading income claims, when senior debt is rated by methodologies that parse operating agreements and audit trails, capital will flow to Agent firms the same way it flows to any other productive business class: toward the highest risk-adjusted return.

That moment is when the category becomes real in the deepest sense. Not when an Agent can complete a task. Not when LLC statutes recognize algorithmic governance. Not even when the first Agent firm earns its first million in revenue.

It’s when outside capital can finance it, price it, rate it, tranche it, trade ownership of its future cash flows, and no longer need to look a founder in the eye or trust a VC’s memo.

At that point, the question is no longer whether Agent firms are legitimate businesses, but which Agent firms are worth capital, at what cost, under what covenants, and at what scale. That’s a capital markets question, not a tech question.

And history shows that once something becomes a capital markets question, the rest happens quickly.

Analysts build coverage. Lawyers standardize docs. Rating agencies publish criteria. Underwriters compress diligence into checklists. Index providers define baskets. Brokers make markets. Asset managers launch products.

The category gets an acronym, then a benchmark, then an ETF, and eventually an industry conference where everyone pretends the outcome was obvious all along.

The point isn’t that software will replace all firms. It’s that a new class of firm is rising, with lower labor intensity, cleaner telemetry, faster feedback loops, and more legible operating histories than most human businesses.

Once that class can own property, contract, borrow, and distribute earnings in standardized ways, capital will engage with it.

That’s Agent capital markets: turning Agent firms from interesting software into fundable commercial blocks in the real economy.

The tether is cut. The opportunity is here.

Câu hỏi Liên quan

QWhat are the four main reasons presented in the article to support the inevitable rise of a dedicated capital market for Agent companies?

AThe four main reasons are: 1) The inescapable economic advantage (Agents operate at a fraction of the cost of human-run service firms). 2) The agents already exist and are already generating revenue (citing examples like Sierra and Harvey). 3) The legal framework is already largely in place (citing laws in Wyoming, Vermont, and Delaware that allow algorithmically-managed LLCs). 4) Capital is actively seeking yield (the immense and growing pool of private credit capital is searching for new, uncorrelated assets like Agent cash flows).

QAccording to the article, what is the fundamental difference between a SaaS company and an Agent company that underpins the argument for a new capital market?

AThe fundamental difference is the legal entity. A SaaS company is a tool sold to humans for them to use, owned by its creators, and its value is depreciated. An Agent company *is* a company itself: it signs its own contracts, holds its own bank accounts, assumes its own liabilities, earns its own revenue, and distributes its own profits to shareholders. This legal status is what enables it to issue securities based on its own cash flows, creating the need for a dedicated capital market.

QWhat are the five potential financing models or stages outlined for an Agent company's capital stack as it matures?

AThe five stages in the evolving capital stack are: 1) Venture capital equity (for human-led infrastructure builders). 2) Programmatic working capital advances (akin to Shopify Capital or Stripe Capital). 3) Revenue-based financing (RBF). 4) Slate financing (pooling funds across multiple Agents to diversify risk, similar to Hollywood). 5) Institutionalized securitization (packaging rated Agent receivables into asset-backed securities). Tokenization is described as a potential settlement layer for stages 3-5, not a primary issuance model.

QHow does the article argue that due diligence for an Agent company differs from that for a traditional startup?

ADue diligence for an Agent company is less about evaluating a founder's vision in a meeting and more akin to credit analysis for a small service business. The key questions shift to: verifying real contracts and payments (from exceptionally clean, auditable data), assessing product durability and model dependency, evaluating customer contract defensibility, and examining the cap table, which is a real-time, code-enforced smart contract. Intuition comes from reading code, operational agreements, and historical performance data, not from personal interactions with founders.

QWhat are the "two tethers" mentioned in the article that currently constrain Agent companies, and how will the Agent capital market address one of them?

AThe two tethers are: 1) The legal tether - the need for a human to legally act on the Agent's behalf. Progress in corporate law (e.g., Wyoming's memberless LLCs) is cutting this tether. 2) The financial tether - the limitation that every dollar an Agent earns ultimately traces back to a human's decision to deploy capital. The Agent capital market is the mechanism to cut this second tether. It will allow capital to flow algorithmically based on auditable performance, freeing Agent growth from the speed and scale of human check-writing, and allowing capital to price and allocate risk based on standardized metrics.

Nội dung Liên quan

Phí 'Kimchi' ở Hàn Quốc Chính Thức Chuyển Sang Âm, Quốc Gia 'Cày' Crypto Cuối Cùng Cũng Thành Quốc Gia 'Cày' Cổ Phiếu

Tác giả Doo (Compound Foundation) phân tích hiện tượng "phí kim chi" (Kimchi Premium) của Hàn Quốc – chênh lệch giá crypto giữa các sàn Hàn Quốc và thế giới – lần đầu tiên chuyển sang âm, tức giá tại Hàn Quốc rẻ hơn. Điều này trái ngược với truyền thống, nơi thường có mức phí bảo hiểm cao do kiểm soát vốn. Có ba lý do chính: 1) Thị trường đầu cơ suy giảm, đặc biệt với altcoin; 2) Thị trường chứng khoán Hàn Quốc tăng trưởng mạnh (gần gấp đôi), thu hút dòng tiền từ crypto; 3) Luật thuế crypto dự kiến áp dụng từ năm sau tác động đến tâm lý. Dữ liệu thời gian thực từ một kênh Telegram Hàn Quốc cho thấy mức chiết khấu duy trì trên 2% trong ngày, với USDT được giao dịch thấp hơn tỷ giá USD/KRW, chứng tỏ nhu cầu mua yếu. Điều này củng cố nhận định rằng vốn từ retail investor đang rút khỏi thị trường crypto để chuyển sang chứng khoán. Tổng vốn hóa thị trường chứng khoán Hàn Quốc đã tăng 86% lên 5 nghìn tỷ USD, vượt lên vị trí thứ 6 toàn cầu, trong khi "phí kim chi" âm là dấu hiệu rõ ràng cho sự thay đổi lớn trong dòng tiền và sở thích đầu tư.

marsbit7 phút trước

Phí 'Kimchi' ở Hàn Quốc Chính Thức Chuyển Sang Âm, Quốc Gia 'Cày' Crypto Cuối Cùng Cũng Thành Quốc Gia 'Cày' Cổ Phiếu

marsbit7 phút trước

Vị trí mới tại Thung lũng Silicon FDE đang thịnh hành, doanh nghiệp cần nhân tài AI như thế nào?

Bài viết thảo luận về sự xuất hiện và nổi bật của vị trí Kỹ sư Triển khai Tiền tuyến AI (AI Forward Deployed Engineer - FDE) trong thung lũng Silicon, được các công ty như OpenAI và Anthropic chú trọng. Vai trò chính của FDE là làm việc trực tiếp với khách hàng để tùy chỉnh và tích hợp các mô hình ngôn ngữ lớn (LLM) chung vào quy trình nghiệp vụ cụ thể, tạo ra luồng công việc Agent phù hợp. Tuy nhiên, bài viết nhấn mạnh rằng tác động lớn hơn của AI đến thị trường việc làm không phải là sự thay thế đơn thuần, mà là sự phân hóa và tạo ra các vai trò mới. Tác giả dự đoán nhu cầu lớn hơn sẽ dành cho các Kỹ sư AI nội bộ của doanh nghiệp. Những kỹ sư này cần thành thạo nhiều kỹ năng như viết prompt, sử dụng khung Agent, hệ thống đánh giá, cũng như các công cụ lập trình AI như Claude Code hay Codex để nhúng khả năng AI vào phần mềm và hệ thống nghiệp vụ. Sự phát triển này được so sánh với quá trình chuyên môn hóa của kỹ sư phần mềm trước đây, khi vai trò chung ban đầu dần phân nhánh thành front-end, back-end, mobile, DevOps... Tương tự, vai trò Kỹ sư AI tổng quát hiện nay được kỳ vọng sẽ tiến hóa thành nhiều vị trí chuyên sâu hơn trong tương lai, chẳng hạn như Kỹ sư LLMOps, Kỹ sư Đánh giá (Evals Engineer), Kỹ sư Dữ liệu AI... Điều quan trọng nhất là nhu cầu về những người vừa có năng lực kỹ thuật xuất sắc, vừa hiểu sâu sắc về bối cảnh và nhu cầu kinh doanh cụ thể.

marsbit9 phút trước

Vị trí mới tại Thung lũng Silicon FDE đang thịnh hành, doanh nghiệp cần nhân tài AI như thế nào?

marsbit9 phút trước

a16z còn bao xa nữa để lên sàn? 600 tỷ USD, 7 quỹ, tự xây dựng truyền thông, công ty VC Thung lũng Silicon đang biến thành Blackstone tiếp theo

a16z đang quản lý khoảng 600 tỷ USD tài sản, gần bằng quy mô của Blackstone trước khi IPO. Bài viết phân tích động thái gần đây của hãng VC này - gọi vốn 150 tỷ USD cho 7 quỹ, xây dựng nền tảng truyền thông riêng (Future, a16z.news, Turpentine), thuê các đối tác chuyên về xây dựng tường thuật - không chỉ là hoạt động tiếp thị thông thường mà giống như diễn tập cho một đợt chào bán công khai (IPO) trong tương lai. Bài viết lập luận a16z đang âm thầm chuyển đổi từ một công ty đối tác truyền thống sang một công ty quản lý tài sản thay thế đa chiến lược, với cấu trúc tương tự Blackstone hay KKR. Các bước chuẩn bị bao gồm: có tư cách Cố vấn Đầu tư Đã đăng ký (RIA) từ 2019, xây dựng danh mục quỹ đa dạng, và quan trọng là sở hữu hạ tầng truyền thông để kiểm soát câu chuyện tài chính - một nhu cầu thiết yếu của một công ty đại chúng. Dự đoán thời điểm IPO khả thi nhất là khoảng năm 2028-2030, sau một làn sóng thoái vốn AI thành công. Nếu a16z niêm yết, nó có thể thiết lập tiền lệ, buộc các hãng VC lớn khác như General Catalyst, Sequoia phải làm theo, thay đổi vĩnh viễn các quy tắc của ngành công nghiệp đầu tư mạo hiểm.

marsbit11 phút trước

a16z còn bao xa nữa để lên sàn? 600 tỷ USD, 7 quỹ, tự xây dựng truyền thông, công ty VC Thung lũng Silicon đang biến thành Blackstone tiếp theo

marsbit11 phút trước

Bán 32 BTC để chi trả cổ tức: Strategy phá vỡ lời thề "không bao giờ bán"

Ngày 1 tháng 6, MicroStrategy đã nộp hồ sơ 8-K cho SEC, tiết lộ việc bán 32 Bitcoin từ ngày 26 đến 31 tháng 5 với giá trung bình 77.135 USD, thu về khoảng 2,5 triệu USD. Khoản bán này, mặc dù chỉ chiếm 0,004% tổng nắm giữ và có ý nghĩa tài chính nhỏ, nhưng mang tính biểu tượng lớn vì đánh dấu lần bán đầu tiên nhằm mục đích chi trả cổ tức (cho cổ phiếu ưu đãi), phá vỡ cam kết "không bao giờ bán" trước đây của công ty. Công ty đã phát hành dồn dập cổ phiếu ưu đãi từ đầu năm 2025 với nghĩa vụ trả cổ tức cố định, tích lũy hơn 693 triệu USD. Khoản dự trữ tiền mặt 2,25 tỷ USD được lập cuối năm 2025 để chi trả đã giảm còn 9 tỷ USD vào cuối tháng 5/2026. Trong cuộc gọi thu nhập Q1, CEO Phong Le lần đầu đề cập đến việc "bán Bitcoin có kỷ luật" như một công cụ quản lý vốn, báo trước cho động thái này. Việc bán hàng xảy ra đồng thời với một thị trường dự đoán trị giá hơn 111 triệu USD trên Polymarket về việc liệu MicroStrategy có bán Bitcoin trước ngày 31/5 hay không, gây ra tranh cãi về thời điểm công bố thông tin. Bối cảnh thị trường cũng đang yếu, với Bitcoin giảm xuống dưới 72.000 USD và các sản phẩm đầu tư kỹ thuật số ghi nhận dòng tiền ra ròng lớn trong tháng 5. Nhiều công ty nắm giữ Bitcoin kho bạc khác như MARA, Riot Platforms cũng đã bán ra số lượng lớn trong quý I để tái cấp vốn hoặc trả nợ, cho thấy áp lực bán tập thể. Mô hình kho bạc Bitcoin đang đối mặt với thách thức khi vòng xoáy tăng giá - phát hành cổ phiếu - mua thêm bị đảo ngược. Dù vậy, MicroStrategy vẫn là người mua ròng mạnh mẽ trong tháng 5 (hơn 20 tỷ USD), cho thấy khoản bán 2,5 triệu USD hiện chưa phải là khủng hoảng thanh khoản. Tuy nhiên, hành động này đã gửi đi tín hiệu quan trọng: ngay cả "kẻ tích trữ" Bitcoin lớn nhất cũng bắt đầu xem xét việc bán như một lựa chọn khả thi.

marsbit13 phút trước

Bán 32 BTC để chi trả cổ tức: Strategy phá vỡ lời thề "không bao giờ bán"

marsbit13 phút trước

Chiến trường mới của cuộc cạnh tranh AI: Bộ nhớ dài hạn trở thành điểm yếu, người dùng làm sao bảo vệ quyền sở hữu ngữ cảnh của mình

Tác giả: Zen, PANews Trong cuộc cạnh tranh AI hiện nay, khi các mô hình ngày càng mạnh và giống nhau, "trí nhớ dài hạn" và "quyền sở hữu ngữ cảnh" của người dùng đang nổi lên như một điểm nghẽn và chiến trường mới. AI đang chuyển từ công cụ trò chuyện một lần thành trợ lý số cá nhân, tham gia sâu vào quy trình công việc, tích lũy hiểu biết về thói quen và sở thích của người dùng. Tuy nhiên, những ký ức và ngữ cảnh này hiện bị khóa trong các "hòn đảo" nền tảng riêng lẻ như ChatGPT, Claude hay Gemini, khiến người dùng mất dữ liệu khi chuyển đổi. ZetaChain, từ một cơ sở hạ tầng chuỗi chéo, đã chuyển hướng sang AI để giải quyết vấn đề này. Họ giới thiệu Lớp bộ nhớ riêng tư (Private Memory Layer) và sản phẩm tiêu dùng Anuma, nhằm trao cho người dùng quyền sở hữu và kiểm soát bộ nhớ AI được mã hóa của mình. Người dùng có thể mang ngữ cảnh và sở thích của mình qua lại giữa các mô hình AI khác nhau một cách liền mạch. Hệ thống cũng cho phép lập trình, kiểm tra và thu hồi quyền truy cập của Agent, với tất cả được ghi lại trên chuỗi. Tầm nhìn xa hơn của ZetaChain là xây dựng "Lớp tiêu dùng AI" (AI Consumer Layer), nơi nhiều trợ lý AI chuyên biệt có thể cộng tác dựa trên một hệ thống ngữ cảnh, danh tính và quyền thống nhất do người dùng sở hữu. Đồng thời, mã thông báo ZETA được định vị lại thành "mã thông báo cơ sở hạ tầng AI", dùng để thanh toán, mở khóa dịch vụ và tạo điều kiện cho nền kinh tế sáng tạo nơi kiến thức chuyên môn có thể được đóng gói và kiếm tiền. Bản chất của ZetaChain là thách thức mô hình kiểm soát của nền tảng, tìm cách trả lại quyền sở hữu dữ liệu, ngữ cảnh và danh tính kỹ thuật số của người dùng trong kỷ nguyên AI.

marsbit18 phút trước

Chiến trường mới của cuộc cạnh tranh AI: Bộ nhớ dài hạn trở thành điểm yếu, người dùng làm sao bảo vệ quyền sở hữu ngữ cảnh của mình

marsbit18 phút trước

Giao dịch

Giao ngay
Hợp đồng Tương lai
活动图片