Model Routing & Fallback Design — Market Context

Who’s hiring for this skill, what they pay, and where it’s heading.

Job Market Signal

Model routing is embedded in AI infrastructure and platform engineering roles. It’s a differentiator, not a standalone position.

Titles where routing expertise is valued:

Who’s hiring: LiteLLM (the company behind the OSS project — growing team), Portkey, Martian, Unify AI (building routing products). Every company with >$10K/month LLM spend needs routing: Notion, Stripe, Shopify, Vercel, Databricks, Salesforce. Cloud providers building managed routing: AWS Bedrock (model selection), Azure AI (gateway features), Google Vertex AI. Financial services firms optimizing LLM spend: JPMorgan, Goldman, Capital One. AI consultancies helping clients optimize: Deloitte, Accenture, specialized AI shops.

Remote: ~55% remote-eligible. Infrastructure roles are highly portable.

Industry Demand

Consulting/freelance: Strong and growing. “Help us optimize our LLM costs” is a $15K-$50K engagement. Often combined with cost estimation (Skill 13) and observability (Skill 16). Independent consultants: $200-350/hr for LLM infrastructure optimization.

Trajectory

Appreciating near-term, partial commoditization long-term.

Appreciating now (2026-2028):

• LLM spend is growing faster than budgets. Every company that scales past $10K/month LLM spend discovers they need routing and optimization.
• Model proliferation (Anthropic, OpenAI, Google, Meta, Mistral, DeepSeek, Cohere, and dozens of fine-tuned variants) makes model selection a real engineering problem.
• Provider reliability varies — Anthropic and OpenAI both have had significant outages. Multi-provider fallback is becoming a reliability requirement, not just a cost optimization.

Commoditization risk (2028+):

• Cloud providers are building routing into managed platforms (AWS Bedrock model selection, Azure AI gateway). When “smart routing” is a checkbox, the standalone skill premium compresses.
• LiteLLM and similar OSS tools make basic routing accessible without deep expertise.
• AI gateway startups (Portkey, Martian) are productizing the intelligence that currently requires custom engineering.

Durable value: The tooling layer will commoditize, but the judgment layer won’t: knowing which models to route where, designing quality gates for cascade routing, optimizing prompt variants per model, and building budget-aware policies requires understanding both the models AND the business context. This judgment stays human.

Shelf life: 5-8 years for the full skill set. Basic gateway setup: 2-3 years before it’s fully commoditized. Advanced routing strategies (ML-powered routing, cascade with quality gates, budget-aware policies): 5-8 years.

Strategic Positioning

Model routing pairs with cost estimation (Skill 13) and observability (Skill 16) to form the “LLM infrastructure optimization” package. Key positioning angles:

1. Cost-conscious by default — practitioners who naturally think about cost optimization, not just technical capability, stand out. Develop this instinct by tracking real costs on your own projects.
2. Full-stack routing — designing the routing logic AND evaluating quality across models (connecting to Skills 9-11), not just setting up a proxy. The quality measurement side is the differentiator.
3. Business-case routing — routing decisions are business decisions (which features deserve Opus quality? where is Haiku good enough?). Practitioners who can frame routing in business terms get the senior roles.
4. Entry angle: “I’ll cut your LLM costs 30-50% without degrading quality” is a compelling consulting pitch. Combine with cost estimation (Skill 13) for the full “LLM economics” offering.

Related

• Cost Estimation — Market — overlapping AI FinOps roles
• Guardrails — Market — gateway = routing + safety