← Kurka Labs
Vision

The Third Layer

Intelligence between you and every platform that hates you

Frank said "fix my Google OAuth." What followed was a 10-minute sequence where an AI diagnosed the root cause (Testing mode = 7-day token expiry), navigated Google Cloud Console's credential maze, wrote a token-exchange script, updated production secrets, and verified the pipeline end-to-end.

Frank's only actions: click one button, sign into Google, paste two strings.

This is the Third Layer.

The problem: the platform gap

Today's internet has two layers:

Layer 1 — The platform

Google Cloud, AWS, Cloudflare, GitHub, Stripe, Facebook, Amazon. Each one a maze built around the platform's mental model, not yours.

Layer 2 — The user

"Save my files permanently." "Sell my product." "Publish my writing." Simple intents. Clear goals. No interest in the maze.

The gap between these layers is filled by documentation (platform-centric), Stack Overflow (symptom-focused), YouTube tutorials (stale), and trial-and-error. Nothing understands both your intent and the platform's architecture.

The Third Layer defined

An intent-to-infrastructure translator. Intelligence that understands the user's goals upward and the platform's architecture downward. It bridges the gap — conversationally when it needs information, silently when it doesn't.

Not a chatbot
Chatbots answer questions. The Third Layer acts.
Not RPA
RPA replays recorded steps. The Third Layer reasons.
Not an API wrapper
Wrappers abstract one API. The Third Layer navigates mazes across platforms.
Not no-code
No-code hides complexity. The Third Layer handles complexity you shouldn't see.
Not a single agent
Agents know one maze. The Third Layer knows the topology of mazes.

The maze taxonomy

Developer / infrastructure mazes

Google Cloud

OAuth consent screens, credential types, redirect URIs, Testing vs Production, token lifecycles

"Let my app talk to Drive forever"

AWS

IAM policies, role trust documents, KMS key policies, VPC security groups

"Deploy my app securely"

Cloudflare

Workers vs Pages vs Zones, 47+ token scopes, DNS records, Wrangler CLI

"Put my site on my domain"

GitHub

Classic vs fine-grained PATs, repo permissions, Actions secrets, branch protection

"Let my CI/CD deploy"

Stripe

Webhook signing, API versioning, test vs live mode, Connect onboarding

"Accept payments"

DNS

A vs AAAA vs CNAME vs MX vs TXT, TTL, propagation, registrar confusion

"Point my domain at my app"

Email

SPF vs DKIM vs DMARC, MX records, sending domains, reputation

"Send email from my domain"

Consumer / platform mazes

Facebook

57+ privacy toggles, business vs profile, ad manager, reach algorithm, content moderation appeals

"Show my posts to people who care"

Amazon

Seller Central, FBA vs FBM, A+ content, ad console, brand registry, Buy Box algorithm

"Sell my product and get paid"

eBay

Listing optimization, seller tiers, fee structures, promoted listings, managed payments

"List my item and find a buyer"

Substack

Publication settings, custom domains, paid tiers, welcome emails, recommendation network

"Publish and grow an audience"

Medium

Partner program rules, distribution algorithm, publications, boost eligibility, paywall strategy

"Get my writing read and earn"

LinkedIn

Profile optimization, Sales Navigator, InMail credits, ad targeting, content algorithm

"Get noticed by the right people"

YouTube

Monetization requirements, studio analytics, thumbnail SEO, community guidelines, copyright

"Build an audience for my videos"

Etsy

Shop optimization, search ranking, Etsy Ads, Star Seller metrics, shipping profiles

"Sell my handmade goods"

The pattern is identical. The stakes are different.

Developer mazes are about infrastructure — credentials, APIs, deploy pipelines. Documented poorly, but documented.

Consumer mazes are about optimization — reach, visibility, revenue, settings. Deliberately obscured. Constantly shifting.

Consumer mazes are arguably worse: no documentation culture, deliberate obfuscation (platforms benefit from user confusion), constant change without changelogs, and billions of affected users vs millions of developers.

The Third Layer pattern is the same: user has intent → platform has maze → intelligence bridges the gap.

What exists today and what's missing

Tool
What it does
What's missing
Terraform / Pulumi
Define desired infrastructure state
No diagnosis, no reasoning, no conversation
Copilot / Cursor
Code completion and generation
No platform architecture understanding
Claude Code / Devin
Agentic coding in terminals
Single-session, no persistent infra memory
AWS / GCP chatbots
Platform-specific AI assistant
Single-platform, can't cross mazes
OrchestrateOS
Anti-conversation automation
Eliminates thinking work with the maze

None combine: multi-platform maze navigation + persistent infrastructure memory + conversational intelligence + autonomous action.

Agents vs. the Third Layer

The AI agent hype is building single-platform executors. The Third Layer is the orchestration intelligence that knows which agent to dispatch to which maze, remembers the results, and maintains the cross-platform topology.

Agents are the hands. The Third Layer is the brain that knows the maze topology.

The moat

Maze knowledge isn't secret — documentation is public. The moat is the combination of cross-platform reasoning (an OAuth fix that requires a Wrangler CLI command), undocumented tribal knowledge accumulated over sessions, and persistent user infrastructure context. The moat is accumulated maze knowledge + persistent user context. No one else is building this combination.

"The Third Layer is not a product. It's a category.
The question is who builds it first."