Now Claude Managed Agents Threaten Edge Networks

Add edge networks such as Akamai, Fastly and Cloudflare to the list of enterprise software of service providers that might be disrupted or helped by growing use of language models and especially agents. 

And blame Anthropic for the recent concern. Claude Managed Agents, launched April 8, 2026 in public beta, handles sandboxed code execution, checkpointing, credential management, scoped permissions, and end-to-end tracing.

Investors seem to be pricing in only danger, but upside arguably also exists.

The core tailwind is simply volume. Cloudflare's CEO noted that if a human were shopping for a digital camera and might visit five websites, an AI agent doing the same task might visit 5,000. 

In fact, AI bot traffic across the Akamai network surged more than 300 percent since Akamai began tracking it, with AI training crawlers accounting for the largest share. 

One might point to earlier concerns about Akamai capital spending that has rattled equities of many types, but that does not seem to be the impetus for the declines of April 10. 

Fastly has noted that inference operations likewise are providing a big boost to its revenue. 

So why did Akamai stock sell off sharply, about 17 percent, on the news? Cloudflare sold off 13.5 percent on April 10. And Fastly sold off nearly 22 percent on the same day. 

Simply, Claude Managed Agents threatens to move much of that former “outsourced” or external traffic “in house.” Claude is now arguably getting into the infrastructure layer for agents, as a potential competitor. 

Impact area	Mechanism	Direction	Effect on CDNs	Who benefits	Source
Explosive growth in bot & agent traffic	Claude Managed Agents spawn long-running sessions that browse the web at machine speed — the Cloudflare CEO noted agents may visit 1,000× more sites than a human doing the same task. Akamai tracked a 300% rise in AI bot traffic across its network in 2025.	tailwind	More requests, bandwidth, and cache-hit pressure — CDNs must handle vastly more non-human HTTP volume.	All three; Cloudflare leading	TechCrunch
Edge inference & AI Gateway demand	Anthropic routes MCP tool calls through third-party services; CDNs like Cloudflare offer AI Gateway products that proxy and cache LLM API calls at the edge. Workers AI inference requests grew 4,000% YoY in Q1 2025.	tailwind	New high-margin inference and gateway revenue streams open up for CDNs that offer edge compute.	Cloudflare (Workers AI, AI Gateway); Fastly (semantic caching)	Klover.ai
MCP server hosting & orchestration	Claude Managed Agents connects to external services through MCP servers. Cloudflare extended its Workers platform with MCP Server Portals; Fastly offers edge orchestration for MCP-style workloads.	tailwind	CDNs become the natural 'control plane' for routing agent tool calls with low latency and built-in auth.	Cloudflare (MCP Server Portals); Fastly	Cloudflare blog
Bot management & agent identity	Agents generate traffic that is hard to distinguish from malicious bots. Akamai and Cloudflare must evolve bot management to allow legitimate agent traffic while blocking bad actors. Cloudflare & GoDaddy partnered specifically to separate trusted agents from scrapers.	tailwind	Increased demand for bot management, agent identity verification, and WAF rule updates.	Akamai (Bot Manager); Cloudflare (Bot Fight Mode)	The Register
DDoS & security complexity	AI agents can be weaponised or exploited. Cloudflare mitigated a 31.4 Tbps DDoS attack in Q4 2025; hyper-volumetric attacks grew 700% that year. Managed agent traffic is a growing attack surface.	tailwind	Higher security service revenue; CDNs are critical infrastructure for absorbing AI-amplified attack volumes.	All three; Cloudflare most exposed	Cloudflare 2026 report
Semantic caching of agent queries	AI agents issue repetitive or semantically-similar requests. Fastly's 'AI Accelerator' uses semantic caching to avoid redundant inference passes. Cloudflare AI Gateway adds response caching across providers.	tailwind	New caching product category; CDNs monetise the gap between raw inference cost and cached delivery.	Fastly (AI Accelerator); Cloudflare (AI Gateway caching)	Futuriom
Anthropic runs its own agent infrastructure	Claude Managed Agents runs on Anthropic's own infrastructure — sandboxed containers, state management, and orchestration are all handled in-house. This keeps a large slice of compute spend inside Anthropic's cloud, not CDN PoPs.	headwind	Enterprises that previously self-hosted agents (and used CDN edge compute for inference) may shift workloads onto Anthropic's platform instead.	Cloudflare Workers AI; edge compute providers	The New Stack
Disintermediation of traditional CDN caching	AI agents retrieve answers from Claude rather than browsing cached web pages, reducing human-driven page requests. If AI mediates content discovery, fewer end-user requests reach CDN-cached origin servers.	headwind	Long-term erosion of traditional web caching revenue as human page views decline relative to bot/agent queries.	Akamai (most exposed, hardware-intensive legacy CDN model)	FinancialContent
AI-driven content disengagement	Akamai observed agentic traffic declining post-Cyber Week 2025 because many sites lack the structured data agents need. Agents disengage from unoptimised sites, shifting traffic away from poorly-structured CDN customers.	headwind	CDN customers in retail/media face traffic shifts if they don't optimise for agent consumption, reducing effective CDN utilisation.	Akamai customers in retail & media	Akamai AI Pulse
Capital expenditure burden (Akamai-specific)	Akamai is investing heavily in physical GPU inference infrastructure (Akamai Inference Cloud). CapEx is projected at 23–26% of revenue in 2026, weighing on margins even as AI-adjacent revenue grows 45%.	mixed	AI agents drive new revenue but at the cost of high capital intensity for hardware-focused CDNs — software-defined competitors benefit more.	Akamai (headwind); Cloudflare (tailwind — software model)	FinancialContent
Agentic content monetisation standards	Akamai predicts that 'agentic commerce will scale only where trust and permission are clearly established.' Agent identity frameworks (Visa Trusted Agent Protocol, Skyfire KYA) will require CDN-level enforcement of who agents are and what they can access.	mixed	CDNs that build agent identity and permission infrastructure early win new platform revenue; laggards become 'dumb pipes'.	All three — early movers win higher-margin platform roles	Akamai AI Pulse

So even if some tailwinds exist, Claude Managed Agents threaten to transform some of those tailwinds into headwinds. 

The key headwind is that Anthropic runs Managed Agents on its own cloud infrastructure. Users simply describe the agent, and Anthropic handles sandboxing, authentication, and tool execution. 

This keeps substantial compute spend inside Anthropic's own stack rather than at CDN edge nodes.

 

Will AI "Eat Enterprise Software?"

If you are an investor in enterprise software, you are aware there is a fear that language models are going to disrupt the traditional enterprise software market and firms. And that fear seems to be playing out in equity prices.

At one level there is concern that the traditional pricing model (per-seat licenses) will be challenged.

At another level there is concern that increasingly-capable AI models will displace the need for many enterprise software functions. 

Investors are essentially moving from views that “software eats the world” (so invest) to “software is dead” (so stay away). Near-term turbulence is inevitable. 

But it also is possible to argue that long term, there will be more enterprise software, even as AI adoption accelerates. 

More to the point, though language models enable natural language interfaces, automate routine tasks and generating insights from vast datasets, they arguably cannot replace enterprise software. 

Enterprise systems are engineered for reliability, security, scalability, and regulatory compliance in high-stakes environments. Moreover, enterprises often deal with proprietary data silos, strict data privacy laws and mission-critical uptime that general-purpose models cannot easily replicate. 

Aspect	Role of Enterprise Software	Role of General-Purpose Models	Coexistence Example
User Interface and Interaction	Provides structured dashboards, forms, and workflows for consistent, role-based access.	Enables natural language querying and conversational interfaces for ad-hoc exploration.	Models integrated as chatbots within enterprise resource planning systems (querying inventory via plain English without navigating menus).
Data Management and Security	Handles secure storage, compliance (audit trails, encryption), and integration with legacy databases.	Analyzes unstructured data or generates summaries, but relies on external data feeds.	Enterprise tools feed sanitized data to s for insights, while maintaining control over sensitive information ( GDPR-compliant AI assistants in CRM).
Automation and Workflow	Executes rule-based, repeatable processes like approvals or batch processing with high reliability.	Automates creative or variable tasks, such as generating custom reports or code.	Models suggest workflow optimizations within HCM platforms, but the core execution remains in the enterprise system (auto-drafting performance reviews in Workday).
Analytics and Insights	Delivers predefined key performance indicators, business information tools, and real-time dashboards with deterministic accuracy.	Provides probabilistic predictions, trend spotting, or scenario modeling from natural language prompts.	Hybrid BI where enterprise software runs core analytics, and models enhance with exploratory queries ("What if" simulations in financial planning tools).
Customization and Scalability	Supports enterprise-grade customization via APIs, modules, and cloud scaling for thousands of users.	Offers flexible, on-demand generation but struggles with consistent scaling or versioning.	Models used to generate custom code snippets for enterprise integrations, deployed within the platform (auto-building plugins for Salesforce).
Compliance and Auditing	Ensures regulatory adherence with built-in logging, versioning, and certification	Lacks inherent auditability; outputs can be opaque or inconsistent.	Enterprise systems log  interactions as auditable events, using AI for efficiency while meeting standards (fraud detection in banking software).

It’s a bit analogous to the traditional choices between general-purpose and application-specific processing. Sometimes one makes more sense than the other, but both coexist. 

AI-enabled or AI-centric software is moving up the stack of what a product is. So consumer experiences of products include vastly more software content than in prior years.

Sometimes a general-purpose approach will suffice. But not always. ASICs still make sense as well. 

And AI will often allow software to become more capable, rather than replacing it, which is the common concern today. 

Domain experience, codified in enterprise software, arguably will be just as important tomorrow as it is today. 

But investors, at the moment, seem more focused on the near-term negative impact on enterprise software company fortunes. 

In some cases, that concern is exacerbated by huge new capital spending requirements for AI infrastructure. 

An adage suggests "markets can stay irrational longer than you can stay solvent." And that is the reality some investors might be facing in the near term.