A developer backlash against Cloudflare’s AI-era platform push

A personal blog post titled Cloudflare’s AI Psychosis has drawn attention on Hacker News, where the summary page listed 108 points and 90 comments. The author says they work at a small AI startup, rely on Cloudflare, and are neither fully happy nor fully unhappy as a customer. Their central complaint is not that Cloudflare has lost its core value, but that its rapid expansion into AI, developer platforms, and full-stack cloud features has made the experience feel fragmented.

The post contrasts today’s Cloudflare with the company many developers first adopted years ago: a quiet infrastructure layer that handled DNS, absorbed attacks, cached static assets, reduced bandwidth, and sent useful reporting. In infrastructure, boring is often praise. It means dependable, predictable, and easy to reason about.

The critique: too many products, too little coherence

The author argues that Cloudflare remains commercially strong and still operates critical web infrastructure. They also claim that Cloudflare routes roughly one third of daily web requests and note that the company’s stock was at an all-time high at the time of writing. But the criticism focuses on a different issue: the upper layers of the platform now feel driven by launch cadence and AI positioning rather than by finished, durable developer primitives.

Several product areas are used as examples:

• Storage includes D1, Durable Objects with SQLite, KV, R2, Queues, and Hyperdrive.
• Compute includes Workers, Dynamic Workers, Sandboxes, Containers, and multiple code execution paths.
• AI tooling includes Agents SDK, Flue, Project Think, Cloudflare OS, Workers AI, and AI Search.
• Observability is described as still incomplete and as something that gets added after launches rather than being fully integrated from the start.

The storage example is especially pointed. D1 is Cloudflare’s serverless SQLite offering; KV is a key-value store; R2 is object storage; Hyperdrive is described as a connection pooling and caching layer for external PostgreSQL or MySQL databases. The author’s complaint is that Cloudflare still lacks a first-class managed PostgreSQL product that feels native to the platform, even though PostgreSQL remains a common choice for serious applications.

Compute and AI amplify the same pattern

On the compute side, the author sees too many overlapping ways to run code. Workers, Dynamic Workers, Sandboxes, and Containers each have different trade-offs around isolation, startup behavior, pricing, and bindings. The problem is not that different workloads require different runtimes; it is that developers may struggle to identify the default, canonical place to run an application.

The AI layer receives similar criticism. AI Search, formerly AutoRAG, is described as a managed pipeline built on R2, Vectorize, and Workers AI. RAG, or retrieval-augmented generation, means retrieving relevant information from a knowledge base before asking a model to generate an answer. The author says this can be useful for demos or hackathons, but argues it lags stronger RAG platforms or well-built open-source stacks in quality, filtering, hybrid search, and visibility.

Documentation and observability become trust issues

The post also argues that Cloudflare’s documentation and observability have not kept pace with product expansion. Observability is the ability to understand a system through logs, metrics, and traces. The author’s complaint is that these capabilities should be treated as production essentials, not as pieces bolted on after new products are announced.

Documentation is framed as part of the product, not secondary marketing. Incomplete pages, aging examples, and insufficiently precise references reduce confidence, especially for infrastructure customers who need stable behavior and versioned guidance.

Why the criticism matters

The blog post is opinionated and emotionally written, not a neutral benchmark. Still, it captures a tension facing many infrastructure companies in the AI cycle. Cloud providers want to move up the stack, compete for developers, and attach themselves to agent and AI workflows. But infrastructure buyers continue to value reliability, clarity, and boring consistency.

Cloudflare’s durable advantage remains its network, security, caching, and edge execution footprint. Its platform ambitions will depend less on how many AI frameworks it launches and more on whether it can make the existing surface area feel coherent, observable, and well documented. AI may help attract developers, but trust is still the core product for an infrastructure company.