Developer Tools & Platforms — 2026年6月1日 週次レポート
重要な発見
重要な発見（10件）
- 1.Vercel reports weekly deployments have doubled in three months, with over 30% of deployments initiated by coding agents — up 1000% from six months ago. Claude Code accounts for 75% of agent-driven deployments, and projects deployed by coding agents are 20 times more likely to call AI inference providers than those deployed by humans. [1]
- 2.Vercel's AI Gateway production index — covering seven months of traffic from 200,000+ unique teams — shows Anthropic led AI model spend at 61% in April 2026, while Google led token volume at 38%. Agentic workloads now carry 59% of all token volume, up from 31.6% six months ago. [2]
- 3.OpenAI's spend share on Vercel's AI Gateway tripled from March to April 2026 following recent model updates, demonstrating rapid market share shifts tied to model release cycles. High-volume teams averaging 10M+ requests use 35 distinct models on average, making multi-model routing standard architecture. [2]
- 4.A new attack vector termed 'inference theft' emerged: on April 12, 2026, Vercel's docs AI chat endpoint was attacked via residential proxies, spiking to 1,300 requests per minute at peak — projecting an inference cost run rate of over ten thousand dollars per day. Vercel deployed BotID deep analysis, blocking more than ten thousand bot requests in the first minutes. [3]
- 5.Vercel open-sourced 'deepsec' on May 4, 2026, an agent-powered security harness that scales to 1,000+ concurrent sandboxes on Vercel infrastructure and uses Claude Opus 4.7 and GPT 5.5 to surface hard-to-find vulnerabilities in large codebases, with a reported false positive rate of roughly 10–20%.
- 6.Real-world agentic deployment case studies now corroborate Vercel's platform statistics: Superset generates roughly 600 preview deployments per day and 1,000–1,400 deployments per week, while General Intelligence — an 8-person team — runs over 4,000 branches in flight with approximately 100 parallel app versions at any moment. [5] [6]
- 7.Supabase became an official ChatGPT app on May 8, 2026, following its February 2026 Claude connector integration, and published npm supply chain attack guidance on May 26, 2026 — pairing AI ecosystem distribution expansion with proactive security communications. [7]
- 8.Atlassian introduced Cursor directly in Jira — enabling work items to be assigned to Cursor as an AI agent — and launched the Flex commercial model for the AI era, while reporting AI Alert Grouping in Jira Service Management saved 839 hours in 28 days internally. [8]
- 9.Cloudflare published details on May 28, 2026 about building Town Lake, a unified internal analytics platform, and Skipper, an internal AI agent running on top of it — demonstrating developer infrastructure companies deploying AI operationally on their own platforms. [9]
- 10.Stripe Atlas data (May 28, 2026) reports solo founders in the top decile generated 61 times the revenue of the median solo founder in their first six months in 2025, with solo founding at an all-time high — signaling that AI-powered productivity leverage is enabling micro-teams to compete at scale. [10]
エグゼクティブサマリー（8件）
- •Agentic deployments on Vercel have doubled week-over-week in three months, with over 30% of all deployments now agent-initiated — a 1000% increase in six months. New case studies from Superset and General Intelligence provide concrete production evidence that agentic-first development workflows are operating at scale in real organizations. [1]
- •Vercel's AI Gateway production index provides the market's most granular real-world view of AI model dynamics: Anthropic leads spend (61%), Google leads token volume (38%), and OpenAI's share tripled in one month. The data shows that no single provider dominates across all use cases, and multi-model routing at 35+ models is now standard for high-volume teams. [2]
- •Inference theft has emerged as a structurally new security threat for AI-powered developer platforms. The economic asymmetry — a single frontier model prompt costing approximately $2 versus standard HTTP requests at $2 per million — creates high-margin incentives for attackers and requires per-request bot verification rather than session-level authentication. [3]
- •Vercel's open-sourcing of deepsec introduces an agent-powered security tooling category: automated, large-scale vulnerability scanning using coding agents across parallel sandboxes. This repositions security tooling from static analysis toward agentic, concurrent investigation at infrastructure scale.
- •Supabase's back-to-back integrations with Claude (February 2026) and ChatGPT (May 2026), combined with ISO 27001 certification and npm supply chain guidance, reflect a dual-track strategy: expanding AI assistant distribution while formalizing enterprise security credentials. [7] (company announcement — may reflect promotional framing)
- •Atlassian's introduction of Cursor in Jira represents a meaningful architectural shift — Jira is being repositioned from a project tracking tool to an agent orchestration platform where work items can be directly assigned to AI coding agents, extending the agentic workflow into enterprise project management. [8]
- •Cloudflare's internal deployment of Town Lake and Skipper signals a broader market pattern: developer infrastructure companies are now dogfooding AI agents operationally on their own platforms, simultaneously validating product capabilities and generating publishable case study content. [9]
- •Stripe Atlas data indicating solo founders in the top decile earned 61x the median solo founder's revenue in their first six months, alongside SaaStr's reporting that Anthropic runs GTM on conventional tools augmented by Claude as a connecting substrate, together suggest AI productivity leverage is both democratizing startup formation and pragmatically augmenting — not replacing — existing enterprise GTM stacks. [10] [11]
市場動向
Agentic Deployments Accelerate on Vercel Platform
The agentic infrastructure trend identified in the previous period continues to deepen with new production data. According to Vercel, over 30% of deployments are now initiated by coding agents — up 1000% from six months ago — with Claude Code accounting for 75% of agent-driven deployments, Lovable and v0 for 6%, and Cursor for 1.5%. Weekly deployments on Vercel have doubled in just three months. Vercel projects deployed by coding agents are 20 times more likely to call AI inference providers tha…
AI Model Spend Fragmentation: Anthropic Leads Cost, Google Leads Volume
New production data from Vercel's AI Gateway — covering seven months of traffic from over 200,000 unique teams — reveals a fragmented and rapidly shifting AI model market. By spend in April 2026, Anthropic took 61%, Google 21%, and OpenAI 12%. By token volume, the picture reversed: Google led with 38%, Anthropic 26%, OpenAI 13%, and xAI 10%. Agentic workloads now carry 59% of all token volume, up from 31.6% six months ago, and 22.2% of AI Gateway requests ended with a tool call in April 2026, up…
Inference Theft Emerges as Critical AI Security Threat
A new and distinct security trend has emerged in the current reporting period: inference theft, defined as the unauthorized use of paid AI inference for free consumption or downstream resale. Vercel disclosed that on April 12, 2026, traffic to its docs AI chat endpoint spiked to roughly ten times normal volume, reaching 1,300 requests per minute at peak — a rate that would have translated to an inference cost run rate of over ten thousand dollars per day. The attack came through residential prox…
Supabase Deepens AI Ecosystem Integration with ChatGPT and npm Security
Supabase's AI ecosystem integration strategy continues to expand in the current period. On May 8, 2026, Supabase announced it is now an official ChatGPT app, following its February 2026 announcement of becoming an official Claude connector. A new development in this reporting period is Supabase's May 26, 2026 publication on protecting projects from npm supply chain attacks, signaling that security posture is becoming a competitive differentiator for developer backend platforms as AI-generated co…
Atlassian Advances AI-Native Teamwork with Cursor Integration and Flex Pricing
Atlassian's current reporting period introduces two notable new developments beyond the previously identified Teamwork Graph and Rovo productivity gains. First, Atlassian announced the introduction of Cursor in Jira, enabling work items to be assigned directly to Cursor from Jira, positioning Jira as an agent orchestration platform. Second, Atlassian introduced Flex, described as a flexible commercial model for the AI era, built for enterprise agility to help customers keep pace with AI-driven c…
Cloudflare Builds Unified Analytics Platform with Internal AI Agent
In the current reporting period, Cloudflare published details on May 28, 2026 about how it built Town Lake, its unified analytics platform, alongside Skipper, an internal AI agent running on top of it. This is a new development distinct from the previous period's Cloudflare-Claude Compliance API integration. The publication signals a broader trend of developer infrastructure companies building internal AI agents on top of their own data platforms — a form of dogfooding that simultaneously valida…
Solo Founders and Agentic GTM Stacks Reshape SaaS Go-to-Market
Two converging signals in the current reporting period point to a structural shift in how SaaS businesses are being built and operated. Stripe Atlas data published on May 28, 2026 reports that in 2025, solo founders in the top decile generated 61 times the revenue of the median solo founder in their first six months, with solo founding described as being at an all-time high. Separately, SaaStr reported in the current period that Anthropic — one of the most prominent AI companies — runs its go-to…
競合動向
Vercel AI Gateway Production Index Reveals Fragmented Multi-Model Market
Vercel published a detailed production data report from its AI Gateway on May 12, 2026, offering new market intelligence on real-world AI model usage. According to the report, Anthropic led in spend at 61% in April 2026, while Google led in token volume at 38%. OpenAI's spend share tripled from March to April following recent model updates. Agentic workloads now carry 59% of all token volume, up from 31.6% six months prior, and 22.2% of requests ended with a tool call, up from 11.4% in October 2…
Vercel Launches Open-Source AI Security Tool and Inference Theft Defense
Two new security-focused developments emerged from Vercel in the current reporting period. On May 4, 2026, Vercel open-sourced 'deepsec,' a security harness powered by coding agents that scans large codebases for vulnerabilities using Claude Opus 4.7 and GPT 5.5, with support for scaling to 1,000+ concurrent sandboxes via Vercel Sandboxes. On May 29, 2026, Vercel published guidance on defending against 'inference theft' — the unauthorized use of paid AI inference endpoints — describing a real at…
Supabase npm Supply Chain Attack Response Signals Growing Platform Security Focus
Supabase published a post on May 26, 2026 addressing npm supply chain attacks, detailing how the company is responding and providing practical steps for developers to reduce risk. This is a new development in the current reporting period, extending the previously observed trend of Supabase deepening its platform trust and security posture — which previously included ISO 27001 certification (April 22, 2026) and the introduction of PrivateLink (January 27, 2026). The npm supply chain attack respon…
制度・規制動向
AI Inference Theft Emerges as Critical Security Threat for Developer Platforms
Vercel has published detailed guidance on a growing attack vector it terms 'inference theft'—the unauthorized use of paid AI inference endpoints for free consumption or downstream resale. According to Vercel's blog, a real attack on April 12, 2026 caused traffic to their docs AI chat endpoint to spike to roughly ten times normal volume, reaching 1,300 requests per minute at peak, which would have translated to an inference cost run rate of over ten thousand dollars per day. The attack exploited …
Agentic Deployment Share Doubles on Vercel; Multi-Model Routing Becomes Standard
Vercel's AI Gateway production index, published May 12, 2026, reveals that agentic workloads now carry 59% of all token volume through its platform, up from 31.6% six months prior—roughly doubling in half a year. The report, based on seven months of production traffic from over 200,000 unique teams, also shows that 22.2% of AI Gateway requests ended with a tool call in April 2026, up from 11.4% in October 2025. At scale, multi-model routing has become standard: teams processing 10 million or mor…
Supabase Expands Security and Compliance Posture with ISO 27001 and Supply Chain Guidance
Supabase has made two notable compliance and security moves in the current reporting period. The company announced it is now ISO 27001 certified as of April 22, 2026, a formal information security management standard relevant to enterprise developer platform adoption. Additionally, on May 26, 2026, Supabase published guidance on protecting projects from npm supply chain attacks, outlining practical steps for developers to reduce risk from compromised open-source packages. These developments refl…
ソース活動
重要な変化の整理
Vercel Agentic Infrastructure Platform
継続監視Vercel's agentic infrastructure strategy continues as a dominant market signal. The platform reports weekly deployments have doubled in three months, with over 30% of deployments initiated by coding agents — up 1000% from six months ago. Claude Code accounts for 75% of agent-driven deployments, and Vercel projects deployed by coding agents are 20 times more likely to call AI inference providers than those deployed by humans. The platform unifies AI SDK, AI Gateway, Fluid compute, Workflows, Queu…
Vercel AI Gateway Production Index: Anthropic Leads Spend
新規Vercel published a production AI model usage report (May 12, 2026) based on seven months of data from 200K+ unique teams using AI Gateway. Anthropic led in spend at 61% in April 2026, while Google led in token volume at 38%. OpenAI's spend share tripled from March to April following GPT-5.4/5.5 releases. Agentic workloads now carry 59% of all token volume, up 2x over six months. At scale, teams averaging 10M+ requests use 35 distinct models on average, making multi-model routing standard archite…
Vercel Launches deepsec Open-Source Security Tool
新規Vercel open-sourced 'deepsec' on May 4, 2026, a security harness powered by coding agents designed to surface hard-to-find vulnerabilities in large codebases. It uses Claude Opus 4.7 and GPT 5.5 for investigation and supports optional fanout to Vercel Sandboxes for parallel execution, scaling to 1,000+ concurrent sandboxes. The tool reports a false positive rate of roughly 10-20%. Vercel used it on its own monorepos to identify subtle edge cases in auth conditions. [4]
Supabase npm Supply Chain Attack Response
更新Supabase published guidance on May 26, 2026 addressing npm supply chain attacks, detailing how the platform is responding and providing practical steps for developers to reduce risk. This is a new security-focused development since the previous period, which highlighted Supabase's AI integration expansion (ChatGPT app, ISO 27001 certification). The platform continues to expand integrations, with Branching Without Git now set as the default (4 May 2026) and @supabase/server introduced (6 May 2026…
Atlassian Cursor in Jira and AI-Native Org Push
継続監視Atlassian continues its AI-native organization narrative, now introducing Cursor directly in Jira, allowing work items to be assigned to Cursor as an agent orchestration platform. The blog also highlights Rovo Dev delivering 120 PRs in two weeks, AI Alert Grouping saving 839 hours in 28 days in Jira Service Management, and the Atlassian Teamwork Graph described as a context engine powering AI across tools, teams, and agents. The Flex commercial model for the AI era was also introduced. [8]
示唆・見るべき論点（8件）
- 1.Vercel's AI Gateway data revealing Anthropic leads spend (61%) while Google leads token volume (38%) indicates that AI model selection is bifurcating by workload type — Anthropic concentrates on high-stakes, high-cost agentic tasks while Google captures high-volume, lower-cost consumer-facing workloads. Developer platform vendors should design multi-model routing infrastructure rather than betting on a single provider relationship. [2]
- 2.The inference theft attack vector — exploiting the $2-per-prompt cost asymmetry of frontier models — creates a new product category imperative: per-request AI endpoint verification. Developer platforms offering AI inference capabilities without bot-detection layers at the request level face potentially unlimited cost exposure from residential proxy-based attacks. [3]
- 3.OpenAI's spend share tripling in one month following model updates demonstrates that AI model market dynamics are unusually sensitive to release cycles — developer platform vendors using AI Gateway-style multi-model routing can capture these shifts dynamically, while those locked into single-provider integrations face higher switching costs during rapid market moves. [2]
- 4.The General Intelligence case study — an 8-person team running 4,000+ branches and 100 parallel app versions — quantifies a new organizational design possibility: agentic infrastructure can decouple team headcount from engineering throughput, which will pressure traditional staffing models for developer tool companies and their enterprise customers alike. [6]
- 5.Supabase's sequential integration into both Claude and ChatGPT interfaces within three months establishes AI assistant ecosystems as a primary distribution channel for developer backend infrastructure. Vendors without equivalent AI assistant partnerships risk losing discovery advantage as developers increasingly interact with infrastructure through conversational interfaces rather than documentation. [7]
- 6.Atlassian's Cursor-in-Jira integration signals that enterprise project management platforms are becoming agent orchestration layers. This creates a competitive dynamic where whichever project management vendor most deeply integrates with leading AI coding agents will accumulate organizational workflow lock-in beyond traditional feature competition. [8]
- 7.Cloudflare's internal Skipper AI agent built on its own Town Lake analytics platform illustrates an emerging competitive signal: infrastructure vendors that operationalize their own products via AI agents generate compounding product feedback loops — internal usage at scale surfaces edge cases and produces authentic case studies that marketing-only approaches cannot replicate. [9]
- 8.Stripe Atlas data showing top-decile solo founders generating 61x the median founder's revenue, combined with SaaStr reporting that AI-native companies like Anthropic use conventional GTM tools augmented by AI rather than AI-native replacements, suggests the near-term opportunity for developer tool vendors is augmentation-layer tooling — not wholesale GTM replacement products. [10] [11]
信頼度サマリー
今週追跡された 11 件のソース15 件の監視対象 URL から、期間中に新着・更新が検出された記事数。
各ソースは信頼度レベルに応じて重み付けされています。単独ソースの主張は AI 合成時に未検証としてフラグ付けされます。
ソース
Vercel published quantitative data on its agentic infrastructure positioning: over 30% of deployments are now agent-initiated (up 1000% in six months), weekly deployments doubled in three months, Claude Code accounts for 75% of agent deployments, and agent-deployed projects are 20x more likely to call AI inference providers. Company announcement — may reflect promotional framing.
関連: marketVercel's production AI model usage report covering seven months of data from 200,000+ teams: Anthropic led spend at 61%, Google led token volume at 38%, OpenAI's spend share tripled March-to-April. Agentic workloads carry 59% of token volume (up from 31.6%), and high-volume teams average 35 models. Company announcement — may reflect promotional framing.
関連: marketVercel disclosed an inference theft attack on April 12, 2026 that spiked its docs AI chat to 1,300 requests/minute at peak via residential proxies, projecting over $10,000/day in inference costs. Mitigation used BotID deep analysis, blocking 10,000+ bot requests in the first minutes.
関連: marketVercel open-sourced deepsec, an agent-powered security harness using Claude Opus 4.7 and GPT 5.5 that scales to 1,000+ concurrent sandboxes on Vercel infrastructure with a ~10-20% false positive rate.
関連: marketSuperset generates roughly 600 preview deployments per day and ships 1,000–1,400 deployments per week on Vercel, with 57–64% week-over-week DAU growth, serving as a production case study for agentic infrastructure at scale.
関連: marketGeneral Intelligence, an 8-person team, runs over 4,000 branches in flight with ~100 parallel app versions at any moment, with engineers averaging 10 PRs per day — demonstrating agentic infrastructure decoupling headcount from engineering throughput.
関連: marketSupabase became an official ChatGPT app (May 8, 2026), published npm supply chain attack guidance (May 26, 2026), introduced @supabase/server (May 6), set branching without Git as default (May 4), and previously announced ISO 27001 certification (April 2026). Company announcement — may reflect promotional framing.
関連: marketAtlassian introduced Cursor in Jira for agent orchestration, launched the Flex commercial model for the AI era, reported AI Alert Grouping in Jira Service Management saved 839 hours in 28 days, and repositioned its Design System as a full-stack AI context engine.
関連: marketCloudflare published details on building Town Lake, its unified internal analytics platform, and Skipper, an internal AI agent operating on top of it — demonstrating internal dogfooding of AI agents on developer infrastructure.
関連: marketStripe Atlas data (May 28, 2026) reports solo founders in the top decile generated 61x the median solo founder revenue in their first six months in 2025, with solo founding described as at an all-time high.
関連: marketSaaStr reported that Anthropic runs its GTM on conventional tools augmented by Claude as a connecting substrate rather than AI-native replacements, illustrating pragmatic AI integration over wholesale GTM rebuilds.
関連: market