The Future: AI Economy - Which Jobs Will Disappear, Which Will Emerge, Where to Grow

Цели урока

• Understand which professions get automated first and why
• Learn about new roles in the AI industry and their requirements
• Evaluate the stages of AI maturity in companies
• Explore business models for AI developers
• Build a career development strategy for the AI era

McKinsey estimates AI's economic potential at USD 4.4 trillion per year - more than Germany's GDP. Goldman Sachs sees 300 million jobs affected globally. And already today: GitHub Copilot generates 55% of all code in teams that adopted it. The market isn't waiting - it's restructuring now.

• Klarna cut 700 support agents: the AI system handles 2/3 of chats with identical CSAT - saving USD 40M per year, payback in 3 months
• Cursor: 10 engineers, 1 year, USD 100M ARR - an AI IDE taking market share from VS Code through pure product-led growth
• Jensen Huang (NVIDIA): 'Every company will become an AI factory' - and NVIDIA sells the infrastructure for all those factories
• Shopify CEO Tobi Lutke: 'Before hiring anyone, prove that AI can't do this job'

The labor market shifts after ChatGPT

ChatGPT's launch in November 2022 reached 100 million users in two months and pulled AI out of research labs into daily work. In the years since, the shift has been concrete rather than speculative: a new "AI Engineer" role spread across hiring, coding assistants like GitHub Copilot moved from novelty to default, and firms began reorganizing teams around AI-assisted workflows. The honest picture is one of ongoing restructuring, not a single rupture - some tasks automated, new roles created, and the long-term balance still unsettled.

Предварительные знания

• Familiarity with the AI Engineer role and core production patterns (RAG, agents, evaluation)
• Basic grasp of how scaling and new model capabilities are progressing

• Building AI SaaS: Billing, Rate Limiting, Multi-tenancy, API Design
• The Future: Path to AGI - Scaling Laws, Emergent Abilities, the Alignment Problem

Job Automation: Who's First in Line

AI doesn't replace entire professions - it automates **specific tasks**. A profession disappears when >80% of its tasks are automated. A profession transforms when 30-60% are automated. McKinsey calculated: by 2030, AI will affect ~30% of working hours economy-wide, but only 5% of professions will fully disappear - the rest will change.

The key pattern: **routine gets automated first**. If a task can be described by a clear algorithm and has examples - AI can handle it. That's exactly why GitHub Copilot generates 55% of code in adopting teams: boilerplate, tests, CRUD - these are algorithms. If it requires business context, political judgment, or dealing with uncertainty - humans are still irreplaceable.

**McKinsey (2024):** by 2030, AI will affect ~30% of work hours in the economy. But only 5% of professions will be fully automated. The remaining 25% - transformation: the same people doing more with AI's help.

For developers, the situation is paradoxical: AI **automates code** (Copilot, Cursor, Devin) while simultaneously **creating massive demand** for engineers who know how to work with it. LinkedIn tracks a 4.5x growth in AI skills job listings between 2023-2025. Automation of some tasks opens a market for others - and that market is not yet saturated.

New Roles: AI Engineer, Prompt Engineer, AI PM

Every technological revolution destroys some professions and creates others. The internet killed travel agents - but created SEO specialists, UX designers, and DevOps engineers. LinkedIn shows roles with 'AI Engineer' in the title grew 300% in two years. AI is creating its own set of specializations - and demand still outpaces supply.

**AI Engineer** is the most in-demand of the new roles. This isn't an ML researcher (who trains models from scratch), but an **applied engineer** who builds products on top of existing models: RAG pipelines, agents, integrations via function calling. These are exactly the engineers building Cursor, Perplexity, Klarna AI - products with real economics.

**Prompt Engineer as a standalone role** is already fading. Prompt engineering is becoming a skill every developer has - like SQL or Git. A dedicated position survives only in companies with very complex AI pipelines.

Company Transformation: AI-First Organizations

Companies go through **4 stages of AI adoption** - and most are stuck at the first. Understanding these stages helps evaluate an employer's maturity, anticipate hiring risks, and find areas of maximum engineering impact.

**Not every automation succeeds.** Companies that simply "replaced people with AI" without redesigning processes often see quality drops and customer churn. AI requires a new workflow, not just a swap into the old one.

For engineers, this means: maximum value isn't in writing code, but in **designing AI systems**. Klarna didn't just swap people for a model - the entire support workflow was redesigned: routing, escalation, training loops. That system design is what pays 2-3x more than coding alone.

AI Developer Economics: Freelance, API-as-Product, Indie AI

AI has dramatically lowered the barrier to building products. A single developer with LLM API access creates in a week what used to take a team of 5 and 3 months. Cursor hit USD 100M ARR with 10 engineers - numbers unthinkable in the pre-AI era. This creates a new economy of **indie AI developers**.

• **API-as-Product** - a wrapper over an LLM with domain logic: legal AI, medical AI, financial analysis. 60-80% margin with good prompt engineering
• **AI-powered SaaS** - a full product with AI inside: Jasper (USD 80M ARR), Copy.ai (USD 30M ARR). High competition, need a moat
• **AI Consulting** - implementing AI in companies. USD 150-300 per hour for experienced specialists. Huge demand, limited supply
• **Custom AI Solutions** - contract development of AI pipelines: RAG, agents, automation. USD 50K-200K per project
• **AI Education** - courses, mentoring, content. The market is growing exponentially

**"A wrapper over ChatGPT" is not an insult.** Stripe is a wrapper over banking APIs. Twilio is a wrapper over telecommunications. The value is in convenience, domain expertise, and reliability. A product only loses when the moat is just a prompt that can be copied in 5 minutes.

A freelance AI developer in 2025 earns 1.5-3x more than a backend developer at the same level. Demand for AI integrations is growing faster than supply: Upwork reports AI-related contracts up 220% in 2024 alone. This is a window of opportunity - in 3-5 years the market will saturate and the premium will compress.

Career Strategy in the AI World: T-Shaped Skills

AI amplifies those who know how to use it and devalues those it can replace. GitHub published data: developers with Copilot complete tasks 55% faster. The strategic question: **how to become someone AI amplifies rather than replaces?**

The answer is **T-shaped skills**: deep expertise in one area (the vertical) + a broad set of adjacent skills (the horizontal). AI makes the horizontal nearly free: code in an unfamiliar language, a UI mockup, a SQL query - generated in seconds. Value concentrates in the vertical that AI cannot replicate.

• **System design** - designing complex systems. AI generates components, but a human defines how they interact
• **Domain expertise** - deep knowledge of a subject area (fintech, medtech, legal). AI doesn't understand business context
• **Communication** - explaining complex things simply, negotiation, leadership. Interpersonal skills can't be automated
• **Judgment under uncertainty** - making decisions with insufficient data. AI is good at optimization, poor at strategy
• **AI orchestration** - the ability to decompose a task into parts where AI is effective and parts where a human is needed

**Practical rule:** dedicate 20% of weekly time to skills AI can't automate. Read about the business domain, practice system design, participate in architectural discussions. In 2 years, this pays off many times over.

The paradox of the AI era: the more routine gets automated, the more expensive **unique human experience** becomes. An engineer with 5 years in fintech understands why PSD2 constrains architecture, how AML rules shape UX, where compliance requires a human in the loop. Cursor hasn't read Basel IV. That expertise doesn't compress into a prompt.

Summary

• Tasks get automated, not professions: routine patterns with clear algorithms go first - boilerplate, L1 support, template reports
• AI Engineer is the fastest-growing role on the market: an applied engineer building RAG systems, agents, and integrations on top of existing models
• 4 stages of company AI maturity: from a ChatGPT widget in support to AI-native products like Cursor and Perplexity
• The barrier to indie AI dropped dramatically: one engineer with LLM API access builds in 2 weeks what used to take a 5-person team a quarter
• T-shaped skills are the winning bet: AI gives the horizontal for free, value concentrates in deep vertical expertise

What's Next

The AI economy shapes demand for new skills. The next step is understanding how AI becomes personal: assistants with full life context, memory systems, and privacy questions.

• Personal AI — AI assistants with long-term memory and full user context - the next frontier
• AI as an Operating System — A new interface between humans and computers - where the industry is heading

Связанные уроки

• aie-46-ai-coding-assistants — Automation debate centers on coding assistants
• aie-47-autonomous-agents — Autonomous agents drive job automation pressure
• aie-49-ai-marketplace — Indie AI economics build on AI-as-product
• aie-52-capstone-project — New AI roles need full-cycle building skills
• sd-10-microservices — AI-orchestrated work mirrors decomposed service teams
• ml-01