Skip to content
R&D / Innovation

Engineering Taste as Moat

We are a small, geek-driven team. In an era when AI makes tools universally available, the only remaining scarcity is taste — aesthetic judgment about architecture, obsession over detail, commitment to the long game. Nebutra benchmarks against top Silicon Valley R&D orgs: type-safe, observable, Edge-first, multi-tenant, open-source from line one. Not a slogan — this is our daily engineering discipline.

Across the three-layer evolution of the AI stack, our battlefield is the Harness layer — MCP/Skill, tool ecosystems, agent orchestration, A2A. This is where real engineering lives in 2025–2026.

Three-Layer Evolution of the AI Stack

From Weights, to Context, to Harness

The innovation frontier of large models expands outward every two years. 2022 was scaling weights; 2023–2024 was RAG and long-context engineering; 2025–2026, the real marginal value has shifted to the Harness layer — tools, protocols, workflow graphs, and agent infrastructure.

Layer 03 · 2025–2026In progress
← Nebutra's battlefield

Harness Layer ← Nebutra's battlefield

MCP/SkillFunction CallingTool EcosystemsWorkflow GraphsProtocolsMulti-agentA2AOrchestrationSecurityAgent Infrastructure
Layer 02 · 2023–2024

Context Layer

RAGMemoryLong ContextPromptingChain-of-ThoughtKnowledge InjectionContext Engineering
Layer 01 · 2022

Weights Layer

PretrainingRLHFFine-tuningScaling LawAlignmentInstruction-followingFew-shot

Nebutra Sailor treats Harness engineering as a first-class concern. We don't train foundation models, and we don't just do prompting — we ship production-grade Harness: MCP/Skill-driven agents, A2A protocols, observable workflow graphs, and a secure tool ecosystem. This is the next architectural coordinate of AI-native SaaS.

Pillar 01

AI-Native Architecture

Every product designed for the AI era from line one. MCP Agent workflows, RAG retrieval, and multi-model gateways as first-class citizens.

H01

MCP/Skill

Model Context Protocol + Skill packages as first-class citizens. Sailor ships an MCP server adapter, Skill registry, and sandboxed runtime.

H02

A2A Collaboration

Agent-to-Agent protocol — task routing, capability advertisement, tenant-isolated message bus. Agents collaborate like microservices.

H03

Workflow Graphs

Declarative workflow graphs — stateful nodes, replayable, observable. AI behavior is no longer a black-box prompt but a debuggable DAG.

H04

AI Gateway

Multi-model routing + compute scheduling. OpenAI, Anthropic, local OSS unified — dynamic switching by task, cost quota, and SLA.

Pillar 02

Open-source Infrastructure

Sailor is open-source end-to-end. We believe open ecosystems compound more than proprietary stacks.

Why open-source? Because we believe: in an accelerated era, proprietary edge evaporates in 6 months, while open-source compounds over a decade. It frees customers from vendor lock-in, lets the community harden our infrastructure, and subjects every commit to worldwide review. This is our most confident long-term bet.

1,500+GitHub Stars
300+Contributors
42Core Packages
MITLicense
Pillar 03

Engineering Best Practices

Benchmarked against Silicon Valley standards — TDD, PPR, Edge-first, built-in observability. Engineering taste is what separates winners from losers.

P01

Test-Driven

Tests first. Red → Green → Refactor. 80%+ coverage required on every PR.

P02

Partial Pre-Rendering

Next.js 16 Cache Components — static shell + streamed dynamics.

P03

Edge-First

Deploy to global Edge nodes by default — CDN, data, and compute near users.

P04

Type Safety

TypeScript strict + Zod runtime — contracts at both compile and runtime.

P05

Built-in Observability

OpenTelemetry, structured logs, error tracking — wired from day one, never "later".

P06

Single Monorepo

pnpm workspaces + Turborepo — share taste across apps, avoid divergent rot.

P07

Review Culture

All code dual-reviewed by Claude + human, architecture decisions recorded as ADRs.

ORGANIZATIONAL PRINCIPLES

AI Leverage vs. Administrative Decay

Great organizations should not drift into mediocrity and bureaucracy as they scale. When the Nebutra ecosystem and its incubated companies face rapid growth, the following three are the only non-negotiable principles of organizational evolution — each directly constraining hiring, governance, and resource allocation decisions.

Principle 01

AI Automation as the Only Answer

Refuse to respond to complexity by piling on headcount or management layers. Every scaling problem must return to the two acceptable answers — raising the per-person ceiling, and raising AI automation depth.

Principle 02

Code & System Governance Over Human Rule

Delegate standardized operations, testing, and internal flows to AI Agents. Reduce inefficiency, conflict of interest, and internal friction at the root by embedding governance logic as auditable code and systems.

Principle 03

Preserve Creative Purity

Keep human cognition focused on the highest-value strategic insight, architectural innovation, and real delivery. Reserve scarce organizational resources for judgment and creativity that cannot be automated.

Innovation Timeline

The road walked and stations ahead

2025 Q3

Project Inception

Nebutra founded. Sailor architecture blueprint locked — Harness engineering as the core roadmap.

2025 Q4

Sailor Open-Source Launch

42 core packages open-sourced under MIT.

2026 Q1

Harness Deep Integration

MCP/Skill, A2A, and workflow graphs go first-class — natively interop with customer systems.

2026 Q2

Multi-Modal GA

Image, audio, and video understanding integrated into the Harness tool ecosystem end-to-end.

2026 Q3

Day-1 Global

7 languages, cross-border payments, GDPR-ready out of the box.

Join our R&D journey

Build the next generation of AI infrastructure with fellow geeks