The platform for AI productivity. The foundation for your next services practice.
Tze Yoong Tee
Principal Partner Architect, APJC
agentee.zeytee.com
City list: Cloudflare, https://www.cloudflare.com/network/. Approximate city-centre coordinates adapted from GeoNames, https://www.geonames.org/, licensed CC BY 4.0, https://creativecommons.org/licenses/by/4.0/. Connection arcs are illustrative.
01
Cloudflare
The connectivity cloud — one network, every layer.
*AI routing depends on configuration; Workers AI also provides document conversion and web search. Browser Run handles Gadget exports and can be omitted. †Basin Pipelines (formerly Pipelines): optional PRODUCT_ANALYTICS binding; absent from the default release configuration. Audited Oct 22, 2026 · public source a679442 (Oct 20) · Bindings · AI routing · Analytics hook · Product names
FOUNDATION
Built on Workers, by the Workers team.
WORKSPACE DURABLE OBJECT
One OverseerDurableObject per workspace
Access checks · orchestration strongly consistent workspace state
SUPERVISOR
Workspace supervisor
Controls capabilities, orchestrates agents and manages Gadget and Gatekeeper facets.
Own SQLite database
WORKER LOADERLoads the Gadget's exported class
GADGET FACET
AI-built application
Dynamic Worker code; no direct outbound network access.
Isolated SQLite
INTEGRATION WORKERSupplies a Gatekeeper class
GATEKEEPER FACET
Governed connection
Scopes remote-service access and mediates actions.
Agents are a different workload than websites — they chain tool calls, wait on LLMs, and need to run close to whatever they're calling.
DESIGN NEED
WHAT YOUR WORKFLOW NEEDS
CLOUDFLARE FOUNDATION
Execution
Run generated app code in isolation
Workers + Dynamic Workers
State
Coordinate shared, interactive work
Durable Objects + isolated facets
AI traffic
Choose models and govern requests
AI Gateway + model providers
Access
Connect approved systems and users
Gatekeepers + configured Zero Trust
Complement your existing cloud investments. Deep managed ML training and GPU pipelines remain hyperscaler strengths; this foundation focuses on serving and securing connected AI workflows.
ARCHITECTURE, NOT ASSEMBLY
Purpose-built, not assembled from parts.
On traditional cloud, an agent means stitching together six services before writing any agent logic.
Compute service
Database
Message queue
Container runtime
Secrets manager
API gateway
On Cloudflare, it's one integrated framework:
Agents SDK
Durable Objects — state lives with the agent
Workflows — durable retries + human-in-the-loop
THE PRICING ARGUMENT
Pay for thinking, not for waiting.
Agents spend most of their time waiting — on the LLM, on a human approval, on a scheduled trigger.
Traditional VM / container alive 24 hours
24 hours
Cloudflare Durable Object alive 24 hours, hibernates when idle
10 seconds
Workload illustration, not a price quote: CPU billing tracks compute, while Durable Object duration depends on activity and hibernation. Requests, storage, models and other service charges also apply.
THE NEW PRIMITIVE · CODE MODE
Don't call tools — write code.
Models are trained to write code. So give the agent one code tool, not fifty — it writes a plan that runs in a sandbox and returns only what matters.
DIRECT TOOL CALLS
Every step round-trips the model
LLM→tool→LLM→tool→LLM→tool
CODE MODE
One program composes it all in-sandbox
LLM→code · many calls · one result
fewer hopsless repeated model orchestration
// the model writes this — not 5 separate tool calls const projects = await codemode.list_projects({ status: "active" }); const blocked = []; for (const p of projects)
blocked.push(...await codemode.list_tasks({ projectId: p.id })); return blocked.filter(t => t.status === "blocked");
A Developer Platform primitive — code runs in an isolated Dynamic Worker, so credentials stay out of the model context. On MCP server portals, Code Mode is on by default.
05
Build solutions. Grow your services practice.
Your expertise connects the platform to customer outcomes — from first use case to ongoing operations.
WHAT YOU CAN BUILD A PRACTICE AROUND
Six services. One connected practice.
01 · ADVISE
AI readiness & discovery
Customers need a valuable starting point. Prioritize workflows, assess access and establish a baseline.
Deliver: use-case backlog + pilot brief
02 · CODIFY
Context & skills engineering
Expert knowledge is scattered. Turn trusted sources and working practices into maintained skills.
Deliver: validated context + skill library
03 · CONNECT
Enterprise integration
Agents need governed data access. Configure Gatekeepers and MCP; build custom connectors where needed.
Deliver: scoped, tested system connections
04 · BUILD
Apps & automation
Generic chat cannot finish every job. Build connected Gadgets, team applications and repeatable workflows.
Deliver: tested apps + workflow blueprints
05 · GOVERN
Security & governance
Controls must match the workflow. Design identity, resource scope, approval policy and usage visibility.
Deliver: control design + operating runbooks
06 · OPERATE
Managed AI operations
Systems and knowledge change. Maintain integrations, curate skills and review model usage and quality.
Deliver: support + improvement reviews
WHO DOES WHAT
One platform. Clear responsibilities.
Cloudflare · platform
Compute, storage and AI primitives
Security and access capabilities
Open-source Cloudflare OS reference
Product documentation and enablement
You · services
Discovery and solution design
Context, integrations and custom apps
Deployment and control configuration
Adoption, support and improvement
Customer · ownership
Business priorities and success criteria
Data owners and access authorization
Policy decisions and output validation
Service acceptance and rollout decisions
Agree account ownership, deployment responsibilities and support boundaries for each engagement.
SERVICE SPOTLIGHT · CONTEXT & SKILLS ENGINEERING
Turn customer expertise into reusable capability.
Business Knowledge
Customer-approved sources
Expert Workflows
How experts do the work
↓
Partner + Customer Experts
Select, structure and validate
↓
Shared Context + Skills
Reusable collections with scoped sharing
↓
AI Agent Workspace
Surfaced and ready-to-use in Cloudflare OS
Capture expert practice, validate it with the customer and maintain it as a service.
1
Select trusted knowledge
Identify owners, approved sources and access boundaries.
2
Capture expert practice
Document the steps, decisions and exceptions that matter.
3
Package reusable skills
Give agents task instructions and relevant business context.
4
Validate with customer teams
Check output quality, permissions and human review.
5
Maintain and improve
Update skills as customer systems and processes evolve.
YOUR KNOWLEDGE ENGINEERING DELIVERABLES
Two file types, one knowledge layer.
Skill File
The customer's repeatable way of working — expressed as task instructions.
Example: prepare a weekly account review
Steps, checks and escalation conditions
Reusable across customer-authorized teams
One page ideal · .md
Context File
The customer's business knowledge — available where the task needs it.
Example: products, terminology and service guidance
Approved reference material and business rules
Owned and maintained with customer subject experts
One–two pages · .md
A REPEATABLE DELIVERY MOTION
From first workflow to ongoing service.
Scope the engagement around customer readiness, deliverables and clear decision points.
01
Discover
QUALIFY VALUE
Workflow and owner
Readiness and baseline
Agreed pilot brief
02
Design
AGREE THE SCOPE
Architecture and access
Data and model choices
Acceptance criteria
03
Build
PROVE THE WORKFLOW
Skills and integrations
Apps and control tests
Customer validation
04
Operate
ACCEPT INTO SERVICE
Runbooks and ownership
Support and releases
Usage and quality reviews
05
Expand
SCALE WITH EVIDENCE
Measured pilot outcomes
Next workflow backlog
Reusable solution patterns
Each stage ends with an agreed customer decision. Timeline, staffing and support terms follow the actual scope.
PROVE CUSTOMER VALUE
Measure what matters to the customer.
Establish the baseline with the customer first. Use the scorecard in service reviews and expansion decisions.
Workflow efficiency
Completion time, manual steps and human review effort per task.
Quality and adoption
Output acceptance, corrections, repeat usage and user feedback.
Cost per task
Model, platform and service costs against delivered value.
Governance fit
Resource scope, approvals, access reviews and operating ownership.
Evidence before expansion. Targets are agreed for the customer's workflow during discovery. Productivity gains and savings are measured, not assumed.
THE PARTNER BUSINESS MODEL
Deliver a project. Build a recurring relationship.
01 · PROJECT SERVICES
Establish the capability
Discovery, architecture, integration, context engineering and application delivery.
Commercial model: scoped project fee.
02 · MANAGED SERVICES
Operate and improve
Integration maintenance, skills curation, usage reviews, user support and controlled releases.
Commercial model: recurring service fee.
03 · REUSABLE SOLUTIONS
Scale your expertise
Develop industry workflow patterns, connector packages and validated delivery accelerators.
Commercial model: packaged implementation.
Illustrative engagement structures; agree scope and service terms with the customer. Track model and platform consumption separately from your services fees.
06
Build it internally. Take it to customers.
Start with one workflow, prove your delivery approach and package a repeatable offering.
YOUR FIRST PARTNER PILOT
Prove it in your own business.
1
Select one internal workflow
Choose a repeated task, name its owner and capture the current baseline.
2
Deploy and connect a scoped environment
Use the public repository and deployment guide. Configure approved integrations, identity and review controls.
3
Measure it and document the pattern
Validate outputs and operating effort. Capture a demo, runbook and customer-facing service brief.