You're already standing on it — let's build the next platform on top

You teach 3 million developers how systems scale.
Yours can scale on the same edge you already trust.

Educative is already a Cloudflare customer — DNS, edge, Bot Management, and Cloudflare Images are in production today. The next-decade footprint is the developer platform itself: in-browser code, Cloud Labs sandboxes, Fenzo's tutor traffic, course-corpus retrieval, and DevPath's enterprise tenancy — on the primitives Cloudflare already ships.

NS: mario.ns.cloudflare.com  ·  server: cloudflare  ·  /cdn-cgi/image in production on the homepage  ·  __cf_bm bot cookie set on every request

What's already running on Cloudflare today

DNS & CONTROL PLANE
Authoritative DNS via mario / megan.ns.cloudflare.com
EDGE / WAF / BOT
Bot Management active — __cf_bm cookie on every response
IMAGE OPTIMIZATION
/cdn-cgi/image transforms in production on testimonials, course tiles, author photos
DEVPATH B2B
devpath.com on the same Cloudflare account — identical CF + GCP stack

The footprint today, and the footprint that fits the product roadmap.

Educative's product surface is expanding in the directions Cloudflare's developer platform was built for: per-learner sandboxes, AI tutors, agentic course material, mobile (Educative Go), DevPath enterprise tenancy. The current Cloudflare footprint covers the front door. The expansion is the building behind it.

Cloudflare today at Educative

The front door — DNS, edge, bot, images

Authoritative DNS on Cloudflare nameservers. Edge serves the Next.js app. Bot Management protects the public site. /cdn-cgi/image transforms run on every author photo, testimonial avatar, and course thumbnail.

  • Cloudflare DNS — full delegation
  • Bot Management — __cf_bm cookie set on hero requests
  • Cloudflare Images / Polish — live, in /cdn-cgi/image URLs
  • Same setup mirrored on devpath.com

Where the footprint goes next

The developer platform — compute, AI, tenancy, storage

The same edge becomes the runtime for in-browser code, Cloud Labs sandboxes, Fenzo's tutor inference, course-corpus retrieval, DevPath per-tenant isolation, and the static asset tier currently on GCS.

  • Workers + Containers — Cloud Labs sandboxes at the edge
  • AI Gateway — observability + caching for Fenzo and course AI
  • Vectorize + R2 — course-corpus retrieval and asset storage
  • Workers for Platforms — per-learner / per-DevPath-tenant isolation

Nine primitives, mapped to Educative's actual product surface.

Each one maps to something you ship today (Cloud Labs, Fenzo, Compilers, DevPath, Educative Go, Daily Coding Challenge) or something on the public roadmap. Status tags reflect what's live in your stack now.

PRIMITIVE 01 Live on CF

DNS, edge, Bot Management, Images

You already trust Cloudflare with the front door for both educative.io and devpath.com. Image transforms run on the homepage today. This is the foundation everything else snaps onto.

DNS Bot Mgmt Images WAF
PRIMITIVE 02 Highest-leverage next

Workers + Containers for Cloud Labs

Cloud Labs spins up ephemeral cloud sandboxes (AWS / GCP / Azure / K8s) per learner. Containers on Workers gives you per-learner, per-lesson, isolated runtimes at the edge — sub-second cold start, billed only while attached, no per-region capacity planning.

Workers Containers Durable Objects Sandboxes
PRIMITIVE 03 Roadmap fit

AI Gateway in front of Fenzo + course AI

Fenzo is an AI tutor. Course AI grades submissions, runs mock interviews, gives code feedback. That's enormous inference volume across OpenAI (verified in your TXT) and likely Anthropic. AI Gateway gives one logged, cached, rate-limited, budget-capped hop — per-course attribution included.

AI Gateway Semantic cache Per-tenant cost BYO keys
PRIMITIVE 04 Roadmap fit

Vectorize + R2 for course corpus

2,300+ interactive courses — that's a fixed, ever-growing corpus. Vectorize makes Fenzo answer "which lesson teaches this concept?" cleanly. R2 holds the raw lesson assets with zero egress, replacing GCS for the static tier.

Vectorize R2 Zero egress RAG
PRIMITIVE 05 DevPath wedge

Workers for Platforms = per-tenant isolation

DevPath sells Educative to enterprises with their own catalog, branding, SSO, and learner data. Workers for Platforms gives every B2B tenant its own Worker namespace — their own keys, egress, logs, model routing — with one shared control plane. The boundary is enforced by infra.

Workers for Platforms DevPath Multi-tenant SOC 2
PRIMITIVE 06 High-leverage

Workers for the in-browser code runtime

The Compilers product runs code in-browser today, "no setup, no config." Workers + isolates is exactly that primitive at planet scale. Cheaper than VM-per-execution, sub-millisecond cold start, ~50ms median latency to any learner globally.

Workers Isolates Compilers In-browser
PRIMITIVE 07 Mobile + offline

Workers KV + D1 for Educative Go

Educative Go (iOS + Android) needs low-latency, globally distributed reads for course progress, leaderboard state, and the Daily Coding Challenge. KV + D1 give that without standing up regional Postgres clusters for a mobile launch.

KV D1 Educative Go Daily Challenge
PRIMITIVE 08 Quiet workhorse

Queues + Workflows for grading pipelines

AI code feedback, mock interview scoring, AI-graded submissions — all classic async pipelines. Queues + Workflows give you durable, replayable, observable pipelines without standing up a separate worker fleet or Kafka.

Queues Workflows Durable Objects Async
PRIMITIVE 09 Enterprise security

Zero Trust for Educative + DevPath ops

You ship SOC 2. As DevPath grows into Fortune 500 enterprises, identity-aware access to author tooling, content CMS, GCP consoles, and the model environments becomes the next compliance ask. Cloudflare Access layers on the edge you already have.

Access Tunnel SOC 2 Compliance

Cloud Labs is the most obvious next move.

Per-learner sandboxes are exactly the problem Workers + Containers was designed for. Today: a learner clicks "start lab," a VM spins up somewhere, they wait, the lesson timer starts. Tomorrow: a learner clicks, an isolate or container is attached in <100ms from the nearest of 330+ POPs, and the lab is already warm.

Cloud Labs on Workers + Containers, sketched

Per-learner, per-lesson, attached at the edge. Billed only while alive.
LEARNER CLICKS "START LAB"
Browser request hits CF edge
~30ms anywhere on earth
WORKER ROUTES
Spawn container, attach session
Durable Object holds state
CONTAINER LIVE
Lab runtime, ready in <1s
Idle → auto-suspend in 60s
What changes for Educative: No per-region capacity planning. No standing fleet of EC2 instances. Billing reflects what learners actually used, not what was reserved. Cold-start measured in milliseconds, not seconds. And the labs run on the same edge that already serves your homepage.

Fenzo + course-AI = a real inference cost line.

The AI Mock Interviewer alone is one of your sharpest product wedges — "more effective than the $200 human coaches I used for practice" (Ammar D., testimonial on your homepage). Every mock interview = many LLM turns. Multiplied across 3M+ developers, that's a serious inference bill. AI Gateway turns it into a managed cost line with a 30–60% reduction lever.

A back-of-the-envelope, not a quote
Modeled across Fenzo + AI Mock Interview + AI code feedback at $15 / M blended tokens
3M DEVS × CONSERVATIVE USAGE
~30M turns/mo
Fenzo conversations + mock interview turns + per-submission AI code feedback, sampled at modest engagement.
SEMANTIC CACHE HIT RATE
35–55%
Course-specific queries cluster enormously. "Explain CAP theorem" gets asked a thousand times a week. Cache once.
ANNUAL INFERENCE SAVINGS
$0.8M–$2.1M
Before counting per-course-author attribution at content review time, which alone changes how you prioritize the next 100 AI courses.
The real win isn't the savings, it's the attribution. When the System Design course, the Grokking Coding Interview course, and Fenzo each have their own AI cost line — broken out by author, by skill path, by DevPath tenant — you can decide which AI features deserve to scale, which need a cheaper model, and which DevPath enterprise tier should bear which cost. That data lives inside AI Gateway by default.

DevPath is a tenancy problem. Workers for Platforms is the answer.

Every DevPath customer wants their own catalog, their own branding, their own SSO, their own learner data, their own AI-budget cap. That's not a feature flag — that's an isolation boundary. Workers for Platforms lets you spin a fresh, isolated Worker namespace per enterprise tenant, all governed from the same control plane.

Per-DevPath-tenant isolation, sketched

Each enterprise tenant gets its own Worker namespace. Same edge, same observability, completely isolated state, egress, and AI budget.
🏢
acme.devpath
🏢
megacorp.devpath
🏢
fintech.devpath
🏢
retailco.devpath
🏢
healthsys.devpath
Shared control plane — Workers for Platforms + AI Gateway + Bot Mgmt
one runtime · one observability surface · zero cross-tenant data leakage by construction

What's running where today, and where the footprint moves.

Everything here is sourced from public DNS, response headers, and the rendered HTML of educative.io and devpath.com. Cloudflare rows in indigo are already in production.

The current stack, with Cloudflare overlaid

Indigo rows = already running on Cloudflare. Orange rows = the expansion footprint.
LAYER
WHAT EDUCATIVE RUNS TODAY
CLOUDFLARE FIT
DNS
Cloudflare (mario / megan.ns.cloudflare.com)
✅ Live — the foundation everything else snaps onto
EDGE + BOT
Cloudflare__cf_bm bot cookie on every response
✅ Live — Bot Management active
IMAGE OPTIMIZATION
Cloudflare Images / Polish/cdn-cgi/image in production
✅ Live — testimonials, course tiles, author photos
APP RUNTIME
Next.js on GCP (Cloud Run or GKE) — via: 1.1 google
+ Workers in front for routing, A/B, edge logic
STATIC ASSETS
GCS bucket — cdn.educative.io resolves to 34.49.49.99
+ R2 (zero egress) as the cheaper, faster static tier
CLOUD LABS / SANDBOXES
Likely VMs on a hyperscaler per-learner
+ Workers + Containers — per-learner edge sandbox
AI / LLM ACCESS
OpenAI (verified on apex), likely Anthropic, Cursor in dev
+ AI Gateway as the one observable, cached, budgeted hop
SEARCH / RAG
2,300+ courses — likely Pinecone / pgvector / Elastic
+ Vectorize as the native, edge-resident vector tier
DEVPATH B2B TENANCY
Shared app with tenant config, same CF + GCP stack
+ Workers for Platforms — per-tenant isolation by construction
MOBILE / OFFLINE
Educative Go (iOS / Android, just launched)
+ KV + D1 for low-latency global reads
EMAIL
Google Workspace MX, HubSpot + SendGrid + HelpScout
+ Cloudflare Email Security as defense-in-depth (optional)
CONTENT SECURITY
SOC 2 certified, x-frame / x-content-type headers in place
+ Zero Trust to author tools / CMS / model envs as enterprise B2B grows

Why expand the footprint now?

Three things make this the right quarter: (1) Fenzo and AI Mock Interview just shipped and the inference volume is climbing fast — AI Gateway gets dramatically cheaper to instrument before that traffic is locked into direct provider relationships. (2) Educative Go just launched on iOS and Android — the global low-latency tier for mobile state is a greenfield choice. (3) DevPath enterprise growth means the per-tenant isolation story has to be airtight for the next round of Fortune 500 procurement.

And the foundation is already there. Cloudflare DNS, Bot Management, and Images are already in production on educative.io and devpath.com. There is no procurement event to start, no security review to begin, no new vendor MSA to negotiate. The MSA, the SOC 2 mapping, the DPA — all exist. Expansion is a roadmap choice, not a vendor decision.

Worth a 30-minute conversation with the team building this?

You already know the front door works. The interesting conversation is which of these primitives is closest to your current sprint — Cloud Labs on Workers + Containers, AI Gateway in front of Fenzo, or Workers for Platforms behind DevPath. I'd rather hear what's actually on your roadmap than guess.

Matt Holscher Calendar  → Reply by email