News
Research Report
Mind The Agentic Action Gap: Stop Losing Money With AI Agents
Cloudflare wants its network to be the foundational infrastructure for the next wave of AI agents.
The company kicked off Agents Week 2026 with a wave of new announcements. Spanning compute, connectivity, security, identity, economics and developer experience, the company's product blitz touched nearly every layer of the modern agent stack.
Here's a breakdown of the major launches.
Table of Contents
- Agent Memory: Persistent Context Without the Bloat
- Artifacts: Git-Compatible Versioned Storage for Agents
- AI Search: Hybrid Retrieval for Agent Workloads
- Flagship: Feature Flags for AI-Generated Code
- Shared Dictionary Compression: Faster Delivery for Incremental Updates
- Redirects for AI Training: Canonical Tags as Enforced Policy
- isitagentready.com: A Lighthouse Score for the Agentic Web
Agent Memory: Persistent Context Without the Bloat
One of the headlining announcements is the private beta of Agent Memory, a managed service that gives AI agents a way to remember what matters across sessions without consuming precious context window space.
The core problem Agent Memory addresses is "context rot": the degradation in model output quality that occurs when too much information is stuffed into a single context window. Cloudflare's solution is a retrieval-based architecture that extracts meaningful information from agent conversations and surfaces it on demand. Developers can call operations including ingest, remember, recall, list and forget against a named memory profile, accessible via a binding from any Cloudflare Worker or via a REST API for agents running outside of Workers.
The service is designed to work across several agent setups:
- Individual agents, including coding tools like Claude Code or OpenCode
- Custom agent harnesses, such as autonomous background agents used by companies like Ramp, Stripe and Spotify
- Shared team memory, where a memory profile is shared across multiple agents and people, letting coding conventions or architectural decisions accumulate as a durable team asset rather than disappearing when context is pruned
Artifacts: Git-Compatible Versioned Storage for Agents
Cloudflare is also launching Artifacts, a distributed, versioned file system built for agents. Artifacts allows developers to create tens of millions of repositories, fork from any remote source and provide agents with a permanent home for code and data accessible to any standard Git client.
The motivation: existing source control platforms were built for human-paced development. With AI agents generating more code than ever, those systems are straining under the volume. Artifacts is designed to handle the scale — developers can provision per-session repos, create 10,000 forks from a known-good starting point or bootstrap from any existing Git repository.
Artifacts is currently in private beta, with a public beta targeted for early May 2026.
AI Search: Hybrid Retrieval for Agent Workloads
Formerly known as AutoRAG, AI Search has been majorly upgraded and rebranded. The service now offers a plug-and-play search primitive that agents can use without manually configuring vector indexes, chunking pipelines or separate keyword-search infrastructure.
Key new capabilities include:
| Feature | What It Does |
|---|---|
| Hybrid search | Combines vector (semantic) and BM25 (keyword) retrieval in parallel, then fuses results |
| Built-in storage | New instances come with their own vector index and storage; no R2 bucket setup required |
| Metadata boosting | Attach metadata to documents and boost rankings at query time |
| Cross-instance search | Query across multiple AI Search instances in a single API call |
| Dynamic namespaces | Create and delete instances at runtime via the new ai_search_namespaces binding |
AI Search is currently in open beta and free to use within specified limits. Cloudflare has committed to giving at least 30 days' notice before any billing begins.
Flagship: Feature Flags for AI-Generated Code
A new product called Flagship brings native feature flag management to the Cloudflare platform, built on OpenFeature, the CNCF open standard for feature flag evaluation.
The pitch is aimed at agent-driven development workflows: an agent can write new code behind a flag, deploy it safely, exercise the feature in production and roll back without a human needing to intervene at every step. Flagship is built entirely on Cloudflare infrastructure — Workers, Durable Objects and KV — and evaluates flags at the edge without external round-trips. It supports boolean, string, number and JSON flag values.
Flagship is currently in closed beta.
Shared Dictionary Compression: Faster Delivery for Incremental Updates
On the performance side, Cloudflare previewed its roadmap for shared compression dictionary support, a technique that allows browsers and servers to compress responses against a previously cached version of the same resource, sending only the diff.
In internal testing, a 272KB JavaScript bundle compressed with gzip came in at 92.1KB. With shared dictionary compression using the previous version as a reference, that same asset dropped to just 2.6KB — a 97% reduction over the already-compressed file.
The feature is rolling out in three phases, starting with passthrough support, which Cloudflare expects to open as a public beta on April 30, 2026.
Redirects for AI Training: Canonical Tags as Enforced Policy
Cloudflare addressed a growing frustration for web publishers: AI training crawlers that ingest deprecated content despite clear canonical tag signals. The new Redirects for AI Training feature automatically converts <link rel="canonical"> tags into HTTP 301 redirects for verified AI training crawlers, including GPTBot, ClaudeBot and Bytespider.
The company cited a telling data point from its own developer documentation site: AI crawlers visited 4.8 million times over 30 days and consumed deprecated content at the same rate as current content, with canonical tags making no measurable difference. The new toggle, available on all paid Cloudflare plans, enforces the existing content hierarchy for crawlers without affecting human visitors or search indexing.
isitagentready.com: A Lighthouse Score for the Agentic Web
Rounding out the week's launches, Cloudflare introduced isitagentready.com, a free tool that scores any website on how well it supports AI agent interactions. The scanner checks for:
- A robots.txt file
- An llms.txt file
- Whether the site serves structured data
- Whether it can deliver content in markdown on request
- Other signals that determine whether an agent can discover, read and interact with the site effectively
Scores are organized across four dimensions: Discoverability, Content, Bot Access Control and Capabilities. For each failing check, the tool generates a prompt that developers can hand directly to a coding agent to implement the fix.
Webinar
Apr
23
AI for Your DXP: Connect What You Have, Transform How You Work
Most AI strategies stop at the platform—but work happens elsewhere. Bring intelligence into Teams, email, tickets and CRM.
Register
Webinar
Apr
27
Where Contact Center AI Is Actually Headed
A straight read on where the contact center platform market is going and what it means for your next technology decision.
Register
Webinar
May
12
How Marketing Teams Can Take Control of the Web Experience
See how Veo and Kvalifik put marketing in the driver's seat, shipping faster and growing qualified bookings by 150%.
Register
Webinar
May
14
Two Audiences, One WordPress. Is Yours Ready?
The practical steps to making your WordPress portfolio ready for both human visitors and AI agents.
Register
Webinar
On demand
The State of Enterprise Site Search: From Basic Utility to Strategic Advantage
Join CMSWire and SearchStax for a conversation about how enterprise IT and marketing leaders are moving beyond basic site search.
Watch Now
Webinar
On demand
Content Leaders Collective: Navigating Content Decisions at Scale
Discover how content leaders are modernizing content operations, avoiding costly missteps and preparing for scale and AI.
Watch Now
The site itself also exposes a stateless MCP server at /.well-known/mcp.json, allowing any MCP-compatible agent to run scans programmatically. Agent readiness checks have also been added to Cloudflare Radar and the Cloudflare URL Scanner API.
The cloud, according to Cloudflare, is not built for agents, and the shift from web applications to autonomous agents requires a different unit of compute, a different security model and a different storage primitive Cloudflare — infrastructure it's now actively shipping.