| Sierra CX agents (TechCrunch; founded early 2024) | ~$100M ARR in under two years of company life (company + press, Nov 2025). Founded early 2024 — in-window for the young-startup filter at the time of that cited ramp. CEO later said ~$200M ARR by May 2026 (two more quarters to add the next $100M) — noted as post-window acceleration, not a new young-at-signal row. | Agents-as-product: enterprise customer-service agents (humans buy) | A product (at $100M signal); C for agents-as-buyer | Strongest public young-at-signal vertical-agent commercial row after filters. Founded early 2024: the ~$100M ramp was within ~24 months of founding; by the May 2026 ~$200M update and this article's date (~Aug 2026) Sierra is older than 24 months — we keep the historical young ramp, not a claim it is still a seed-stage startup. Survivorship / well-funded. Not vibe-coding. Not agents-as-buyer. |
| Rox sales agents (TechCrunch; founded 2024) | Founded 2024. At a 2025 fundraise (covered Mar 2026), sources said Rox was projected to close 2025 near ~$8M ARR (secondhand; company did not confirm to TC). | Agents-as-product: autonomous sales productivity agents | B– / C product (secondhand projection) | Young, non-vibe, agent-labeled. Soft source — keep as early thin signal, not a table win. Shows how sparse honest young-agent ARR still is. |
| Artisan AI SDR agents (TechCrunch; YC W24) | YC Winter 2024. At Apr 2025 Series A coverage: company-reported ~$5M ARR and ~250 customers for AI sales-development agents (flagship "Ava"). | Agents-as-product: outbound sales agents sold to human teams | B product (company-reported via TC; early churn noted) | Clears age filter at signal (~≤15 months). Not vibe-coding. Article itself documents early product issues, selective customer qualification, and that the firm still hires humans — useful honesty, not a pure "agents replace everyone" story. Still humans-as-buyer. Not multi-source ledger. |
| 7AI security agents (company + multi-outlet Series A; founded 2024, stealth exit Feb 2025) | Founded 2024; launched Feb 2025. At Dec 2025 Series A ($130M, Index-led; $166M total), company said agents processed >2.5M alerts and completed >650k investigations in ~10 months; Fortune 500 customers named (e.g. DXC). No public ARR. | Agents-as-product: autonomous security investigation agents (humans buy) | B product usage (company primary; multi-outlet fundraise); F for ARR | Clears young-at-signal filter. Not vibe-coding. Strong usage/ops intensity with named enterprise logos — still not a ledgered revenue row. Well-funded survivorship. Humans remain buyer of record. Do not treat $130M raise as ARR. |
| Vambe WhatsApp agents (TechCrunch; founded 2023, agent pivot Mar 2024) | Santiago-based. Pivoted Mar 2024 from debt-collection CRM to conversational AI agents for SMB sales/comms on WhatsApp. Company: ~$20k ARR pre-pivot → ~$1M ARR by Nov 2024. $3.85M seed (Monashees-led). | Agents-as-product: SMB messaging agents (humans buy) | B product (company via TC; modest ARR) | Valuable because it is not a mega-funded SF category winner: smaller geography, modest absolute ARR, explicit pivot. Young as an agent product at signal (~8 months post-pivot; company founded 2023 so age is borderline on pure company age, clear on product-window). Not vibe-coding. Still humans-as-buyer. Single-outlet company numbers. |
| Wonderful enterprise CX agents (TechCrunch; Reuters; founded 2025) | Founded 2025 (Israel R&D / Amsterdam HQ). Series A $100M (Index-led, Nov 2025; ~$134M total at the time); later Series B $150M (Insight-led, Mar 2026; ~$2B valuation reporting). Reuters: company expected ~$10M ARR for 2025. TC: voice / chat / email agents; tens of thousands of requests/day claimed; multi-country enterprise deploy. | Agents-as-product: enterprise CX / multi-channel agents (humans buy) | B product ARR (company expectation via Reuters); A young funding velocity | Clears young-at-signal hard. Not vibe-coding. ARR is company expectation at Series A, not multi-outlet audited ledger — keep B. Funding is huge relative to disclosed ARR (classic capital concentration). Still humans-as-buyer. Useful as a second young CX ramp beside Sierra, not as agents-as-buyer proof. |
| Toma dealership voice agents (TechCrunch; YC company page; YC W24 / founded 2024) | YC W24 (Jan 2024). a16z-led $17M Series A (Jun 2025). TC: >100 U.S. dealerships; voice agents for service scheduling, parts, sales Q&A with human handoff. Company (YC page): surpassed 7-figure ARR in <1 year, approaching 8 figures. | Agents-as-product: vertical voice agents for auto retail (humans buy) | B product (company ARR + TC customer count); not multi-source ledger | Clears age filter. Not vibe-coding. Strong signal for a solo founder: non-mega-ARR vertical with word-of-mouth sales story. ARR is company-stated on YC, not independently audited in TC — grade B. Still humans-as-buyer of agent seats. |
| Crescendo AI-native contact center (company PR; PartnerHero acquisition; founded 2024) | Founded 2024. Company (Sep 2025): on track to exceed $100M ARR by EOY 2025. Oct 2024: acquired PartnerHero (BPO / customer ops); combined company then said EBITDA-positive with >$50M ARR. Financing ~$50M (General Catalyst) at ~$500M valuation (company / Bloomberg-era reporting). | Agents-as-product + services: AI contact center with human escalation layer (humans buy) | B product with acquisition-mix caveat; C if read as pure organic software ARR | Young and non-vibe, but do not treat as a Sierra-class pure software ramp. Material ARR came with a human-ops acquisition. Hybrid AI + BPO is a real model; it is a different evidence class than software-only agent ARR. Company-forward $100M track claim needs independent corroboration. Still humans-as-buyer. |
| Sapiom agent spend / runtime layer (TechCrunch seed; company Series A; founded ~2025) | $15M seed (Accel-led, Feb 2026) framed as financial layer so agents can purchase software/APIs/compute under policy. Company Aug 2026 Series A ($35M Dragonfly-led; $50M total) claims >270M platform transactions and >100k agent runs/day since launch — vendor primary. Company blog: unnamed customer cut monthly inference bill ~$1.2M → ~$100k via Sapiom routing (vendor case, not multi-vendor SaaS GMV). | Infrastructure for agent spend + runtime economics (buyer/operator pipes), not vertical agent product ARR | A for "buyer/runtime rails thesis is fundable"; B− vendor usage + spend intensity; F for audited multi-vendor SaaS GMV | Young, thesis-specific, non-vibe. Seed press multi-outlet; Series A usage and inference-bill case are company-claimed. Inference/compute cost optimization is real agent-side spend pressure — still not a public ledger of agents buying many third-party SaaS products. Funding and tx counts are not ARR or SaaS GMV. |
| Natural agent payments (TechCrunch; founded 2025) | ~1-year-old (founded 2025). $30M Series A (Forerunner-led, Jul 2026; ~$40M total) for an orchestration layer so AI agents can move funds, pay vendors, collect, and settle with humans/other agents (stablecoins + bank rails). Beta until raise. | Infrastructure readiness for agents as payers / collectors | A for young buyer-rails company; F for GMV / volume | Clears age filter. Explicit agents-as-buyer thesis in product design. No public payment volume or ARR in the coverage. Competing with Stripe-class rails narratives — announcement ≠ settlement scale. |
| Skyfire agent payment network (TechCrunch; launch Aug 2024) | Launched Aug 2024 with $8.5M seed for payment rails so AI agents can spend under human-funded wallets/policy ("Visa for AI" framing). Later total funding cited ~$9.5M incl. crypto-ecosystem capital. No public GMV in coverage. | Infrastructure for agents as delegated payers | A for early buyer-rails product; F for GMV | Young at launch signal. Explicit agents-as-buyer thesis. Human budget still funds the wallet — delegated clerk, not independent corporate treasury. Funding ≠ settlement volume. Complements Stripe Sessions / Natural / Sapiom without proving multi-vendor SaaS GMV. |
| Paid outcome billing for agents (TechCrunch; ~2025, Manny Medina) | Results-based billing so agent makers can charge on value delivered (not only seats). $21.6M seed (Lightspeed-led, Sep 2025; ~$33M total incl. pre-seed). Named early customers include Artisan and ERP vendor IFS. No public ARR for Paid itself. | Monetization infra for agents-as-product (outcome/price capture), not agents purchasing third-party SaaS | B rails for outcome pricing; C for agents-as-buyer thesis | Young. Supports the article's "goal completion at justified cost" design notes without proving agents are the buyer of record. Customer logos are company- stated via press. Do not confuse "agents get paid for outcomes" with "agents buy your API overnight." |
| Stripe AI vs SaaS growth curves (Stripe annual letter / TechCrunch) | Top 100 AI companies (by revenue on Stripe) reached $5M annualized revenue in ~24 months (2024 cohort) vs ~37 months for top SaaS in an earlier benchmark. | Payment-network evidence that AI software businesses scale revenue faster | A institutional for "AI software S-curve is real"; C for pure agent-buyer claim | Best large-n economic signal we have. Not agent-only. Still: money on the rails is better than narrative. (Stripe letter also names vibe/coding ramps we exclude from this table's product rows.) |
| Databricks platform usage (State of AI Agents / Neon notes) | Multi-agent workflows +327% in <4 months among measured customers; on Neon, AI agents create ~80% of databases and ~97% of branches. | Agents as high-frequency operators of infra APIs (create / branch / orchestrate) | A for usage intensity; B for economic buyer claim | Vendor-scoped (Databricks / Neon customers). Creating databases ≠ paying at scale. Free tiers and internal automation inflate counts. Still the cleanest public signal that agents already drive machine demand differently than humans. |
| Cloudflare agentic / bot traffic (Cloudflare blog; Radar; CEO Jun 2026; Mar 2026 forecast; Investor Day 1,700% agent requests; Q1 official earnings transcript; call Q&A; Q2 earnings PR) | Mar 2026: Cloudflare CEO predicted bot traffic would exceed human traffic by 2027. Jun 2026: same CEO said the crossover had already happened — "agentic traffic growing so fast that bots have now passed human traffic online for the first time". Radar/reporting: ~57% bot vs ~43% human HTTP requests on Cloudflare's network. Company blog (Jul 2026): >50% of traffic non-human. Investor Day: daily AI agent requests up ~1,700% Jun 2025–May 2026. Q1 2026 earnings call (Prince): "hundreds of billions of agentic requests per month, growing exponentially" — framing "rails and guardrails" for agentic commerce (company term: Act One). Q1 Q&A (RBC Hedberg asked why CF monetizes agentic traffic better than Act One peers): Prince answers with essential traffic (APIs/apps, not video CDN commodity) + Workers efficiency + Zero Trust agent access controls — i.e. request vitality and control plane, not a SaaS GMV ledger. Q2 2026 earnings call: for the first time, more than 50% of traffic on Cloudflare's network was non-human; AI agent request growth "continues to grow unabated"; PR frames "agent-driven commerce" and payment rails for the "Agentic Internet." Q2 Q&A (Barclays): Prince states over 80% of major AI companies are Cloudflare customers (concentration of lab demand on the network — still humans/enterprises buying CF). Q2 Q&A (Morgan Stanley on securing agents): demand is already present — large orgs asking for agent-aware SASE/Zero Trust controls (agents-as-product security, not agents shopping SaaS). Q2 Q&A (Prince, with trend caveat): if current trends continue, in ~5 years non-human traffic could be as much as ~1,000× human traffic (~three orders of magnitude) — company projection, not a measured ledger. Elon Musk (Aug 9, 2026) publicly confirmed that forecast direction: "AI agentic Internet traffic will obviously VASTLY exceed human usage. Not a close call at all. Cloudflare's forecast is accurate." (X post) — celebrity confirmation of CF's directional forecast, not an independent measurement or audited GMV. | Machine / agentic request intensity + AI-lab customer concentration + agent-aware security demand + directional ~1,000× non-human vs human projection (CF) with public celebrity confirmation (Elon) — not SaaS ARR or agent-originated multi-vendor GMV. Request counts, 80% AI-lab claim, and 1,000× forecast are company-stated; Elon post is endorsement, not third-party audit. | A institutional for traffic vitality; C for agents-as-buyer SaaS spend | Strongest public network-scale signal that machine demand already outruns human browser traffic in the wild. Hard caveats: (1) "bot" ≠ commercial goal- completing agent — training crawlers are a large share of crawler traffic (Cloudflare: ~52% of crawler requests for AI training as of Jun 2026); (2) Cloudflare is a large but partial view of the web; (3) free scraping and eval loops inflate counts; (4) traffic without paid conversion is still Hole 3; (5) do not treat Cloudflare total revenue ($696.1M Q2 2026, +36% YoY) as agent GMV — Act One / security / Workers mix is broader than agent commerce. Use for vitality of machine demand, not as proof agents hold the purse. Q&A breadcrumbs (AI labs as CF customers; agent security RFPs; one unnamed "AI studio" scaling Workers) stay company-scoped — do not invent young product ARR rows from them. The ~1,000× / three-orders forecast is trend-conditional management projection; Elon's "forecast is accurate" is high-signal confirmation of direction, not proof agents buy SaaS or that 1,000× has already arrived. |
| Cloudflare Wallets / Monetization Gateway (Wallets blog; press; Monetization Gateway; Q2 earnings framing; prior Visa/Mastercard Web Bot Auth) | Agents Week (Aug 2026): Cloudflare Wallets + cloudflare.pay — Account Wallet for stablecoins; Virtual Wallets for agents with human-set spend caps, merchant allow lists, and max transaction size; Monetization Gateway for sellers (x402 micropayments on HTTP; company: sell page, API, dataset, or MCP tool). Handle reservation open; full on/off-ramp and agent spend coming months (company: not full GMV live). Q2 prepared + Q&A: CEO frames payment rails for agentic Internet; fractions of a penny per request via cloudflare.pay. Q2 2026 earnings Q&A (Prince, primary): CF handles ~ half a billion requests/sec; management estimates ~1–10% could carry micro-transactions → ~10 million financial txs/sec day-one, scaling to ~100 million financial txs/sec (hundreds of millions peak class). Same answer: Visa at holiday peak handles ~ 20,000 txs/sec (Prince: "from memory") — so the agentic micro-pay rail must be roughly "three orders of magnitude bigger than Visa" to work. That is architecture ambition / capacity target, not measured settlement volume today. Citi Q&A: Workers / Workers AI described as a meaningful revenue contributor (CF product economics under agent load — still not multi-vendor agent-as-buyer SaaS GMV). Earlier (Oct 2025): Trusted Agent Protocol with Visa; Web Bot Auth with Mastercard Agent Pay / Amex; feedback partners include Stripe, Shopify, Adyen, Coinbase, Circle (ecosystem, not startup ARR). | Infrastructure readiness for agents as payers / merchants as receivers — identity + wallet + HTTP micropay rails — not multi-vendor SaaS GMV | A institutional for rails; F for measured agent-originated SaaS GMV | Same grade logic as Stripe Sessions rails: announcement + protocol work ≠ ledger. Human-set caps mean budget holder is still usually a person (Hole 2). x402 already graded separately for on-chain activity. The ~10M–100M tps vs Visa ~20k tps (~3 orders of magnitude) comparison is CEO Q&A math on required future capacity if a slice of HTTP requests becomes paid — do not report it as live GMV or as Visa's own disclosure. MCP-tool monetization and Workers "meaningful revenue" are CF product/rails economics, not multi-vendor agent-as-buyer SaaS GMV. Breadcrumbs are mostly large payments/infra incumbents — not a free pass for young product ARR rows. |
| Salesforce Agentforce ARR (Q4 FY26 PR; Q1 FY27 PR) | Q4 FY26 (Feb 2026): Agentforce ARR ~$800M, +169% YoY; ~29,000 Agentforce deals closed since launch; ~2.4B Agentic Work Units to that date. Q1 FY27 (May 2026): Agentforce ARR ~$1.2B, +205% YoY; combined Agentforce + Data 360 ARR ~$3.4B (includes ~$1.1B Informatica Cloud ARR — strip M&A when reading "AI+data"); ~3.8B Agentic Work Units to date (+111% QoQ); ~28.6T tokens processed. | Enterprise agents-as-product ARR and usage units sold by a large public CRM vendor to human buyers / seats / contracts | A institutional for agents-as-product monetization at scale; F for agents-as-buyer SaaS GMV; not a young-startup product row | Older public company — institutional signal only under our hard filters. Humans (and enterprises) buy Agentforce; agents do not buy third-party SaaS through this metric. Combined "AI + data" ARR mixes Data 360 + Informatica Cloud — use standalone Agentforce ARR for the cleanest product read. Still: largest clean public breakout of agent product ARR we have from a listed company. |
| ServiceNow AI ACV + agentic deployments (Q2 2026 PR) | Q2 2026: ServiceNow AI crossed $1B in annual contract value; agentic deployments of ServiceNow AI increased ninefold in nine months (CEO McDermott). AI Control Tower positioned as governance/control plane; Autonomous Workforce / related agent products expand the platform surface. Subscription revenue $3.877B (+24.5% YoY) is company-wide — not AI-only. | Enterprise platform AI / agentic product ACV and production deployment counts — agents-as-product inside workflow software, not agents purchasing external SaaS | A institutional for enterprise agentic product adoption + ACV; F for agents-as-buyer SaaS GMV; not a young-startup product row | Same category split as Salesforce: humans buy platform AI; 9× deployment growth is production adoption intensity, not multi-vendor agent spend. AI ACV ≥$1B is a rare public line-item for agentic-flavored enterprise software. Do not treat total ServiceNow revenue as agent GMV. Older public co — institutional only. |
| Stripe payment rails for agents (Sessions 2026) | Link wallets for agents; Agentic Commerce Suite; agents can pay with human-approved, one-time cards without exposing full credentials. | Infrastructure readiness for agents as payers | A for "rails exist"; B− for "volume exists" | Primary source is Stripe (interested party). Announcement ≠ GMV. Explicit human approval per payment means the agent is often a delegated clerk, not an independent budget holder. |
| x402 agentic payments on Base (Chainalysis; Coinbase-origin protocol) | On-chain: >100M cumulative agentic x402 transactions on Base from near-zero in mid-2025 through ~Q1 2026; share of volume in ≥$1 transfers rose sharply over that window. Protocol enables machine HTTP 402-style stablecoin payments. | Measured machine payment activity (tx counts / size mix), not SaaS ARR | A for "agents already pay on rails"; C for multi-vendor SaaS GMV | Strongest public on-chain signal that agents settle value without a human click every time. Caveats: early activity included meme/farming patterns; tx count ≠ durable commercial GMV; independent analyses have argued genuine commercial dollar volume can be thin relative to headline counts. Still: rails + non-zero settlement beat pure narrative. |
| Salesforce Cyber Week 2025 AI/agent-influenced retail (Salesforce PR Dec 2025) | Nov 25–Dec 1 2025 Cyber Week: Salesforce Shopping Index — AI and agents "drove $67 billion in sales," influencing ~20% of global orders via personalized recommendations and conversational service. Aggregated shopper activity across ~1.5B shoppers / 89+ countries; extrapolated with market assumptions. | Agentic retail commerce influence (shopper path-to-purchase), not agents buying multi-vendor SaaS | B institutional for agent-influenced retail GMV; F for agents-as-buyer SaaS GMV | Largest public dollar figure in the "agents touch commerce" bucket. Hard category split: this is humans shopping with AI assistance / on-site agents — not autonomous agents procuring third-party software. Vendor Shopping Index with extrapolation; "not indicative of Salesforce performance." Do not rebrand as SaaS GMV. |
| Shopify AI-referred orders (Shopify enterprise blog; Q1 2026 commerce data) | AI-referred orders on Shopify storefronts grew nearly 13× YoY in Q1 2026; referral sessions from AI chatbots (ChatGPT, Perplexity, Gemini, Copilot, Claude, Grok, etc.) >8× YoY. No public dollar GMV breakout for the AI-referred slice in the blog. | Agentic / AI-mediated retail discovery → order intensity on a large commerce platform | B institutional for agentic retail demand; F for agents-as-buyer SaaS GMV | Primary merchant-platform telemetry (vendor-scoped). Growth multiples from a small base; early signal, not mainstream share. Still humans buying goods — agents/chatbots as discovery channel. Complements Salesforce Cyber Week retail influence; still does not prove agents hold SaaS budgets. |
| McKinsey State of AI 2025 (Global Survey) | ~62% of orgs at least experimenting with AI agents; ~23% scaling at least one agentic system; typically limited to one or two functions. | Self-reported enterprise adoption stage | B | Survey, not ledger. "Experimenting" is cheap. Scaling is still minority. Does not measure agents purchasing third-party SaaS. |
| Gartner task-agent embedding (Aug 2025 PR) | 40% of enterprise apps to include task-specific agents by end of 2026 (from<5% in 2025). Best-case: agentic AI ~30% of enterprise app software revenue by 2035. | Forecast of apps featuring agents | B as directional forecast; F if cited as observed fact | Prediction, not measurement. Same firm warns against "agentwashing." Best-case revenue scenario is explicitly labeled. |
| Gartner project failure risk (Jun 2025 PR) | Over 40% of agentic AI projects canceled by end of 2027 (cost, unclear value, risk controls). | Forecast of enterprise project attrition | B as counter-signal | Also a forecast. Directly contradicts naive "agents take over everything" narratives. Failure modes named are managerial, not just model IQ. |
| MIT NANDA GenAI divide (Fortune summary; report widely cited as The GenAI Divide: State of AI in Business 2025) | ~95% of genAI pilots show little/no measurable P&L impact; ~5% extract rapid value. | Enterprise genAI pilot → value conversion | B as counter-signal | About genAI broadly, not only autonomous agents. Methodology and sample are imperfectly public; still the strongest mainstream "stop hallucinating ROI" finding founders should internalize. |
| YC W26 batch growth (Garry Tan / Demo Day reporting, e.g. secondary recaps) | ~14% average weekly revenue growth across ~200 companies; 14 at $1M+ ARR into Demo Day; AI-heavy batch composition. | YC batch fundraising-week narrative, not agent-buyer ledger | C | Batch-wide, not agent-as-buyer. Demo Day selection and self-report bias. "AI-first companies grow fast" ≠ "agents buy software while you sleep." Useful only as ambient context that AI product velocity is real. |