RACE Programming
By Pavel Khodalev, CTO at First Line Software ·
Layer 1: Science

AI-First Theory

The scientific foundation. Two axioms about AI systems and human tacit knowledge; six theorems that follow from them. Together they describe what happens to the software development lifecycle when AI systems can read, write, and verify code at production grade.

This piece

You are reading the full theory below, about a 6-minute read. Prefer your ears? The whole page is narrated in roughly 5 minutes. Short on time? Watch the 3-minute overview.

Why a theory

Engineering needs a ground truth, not a framework

Most discussions of AI in software development are tactical: which tool, which prompt pattern, which IDE integration. The upstream question: what, formally, is changing?

AI-First Theory answers in terms suitable for reasoning, falsification, and extension. Not a framework. Not a manifesto. A compact set of propositions, stated precisely enough to disagree with, find edge cases in, or extend.

Theory → Manifesto → RACE Programming. Each layer more prescriptive. Each depends on the one above being sound.

The base

Two axioms and the core skill

The foundation everything rests on, for engineers and knowledge workers alike. Two claims taken as premises, and the one skill that follows immediately from them.

  1. Axiom I · What AI is actually good at
    AI excels at exactly two cognitive functions, finding information and generating actionable insights, where it delivers 1000x+ acceleration. Everything else (judgment, intuition, tacit knowledge) remains human.
  2. Axiom II · Humanity always adapts (but this time is different)
    Humans have always adapted to new tools, and AI is no exception. But earlier tools amplified capability; AI replaces entire cognitive functions. It doesn't help you search, it searches for you. That is unprecedented, and the adjustment may take a generation.
  3. Theorem I · Delegation Is the Core Skill
    Delegation is the core skill, not prompting: verbalize context → delegate → validate. Once context is clear, AI executes; the human job is quality control.

Full discussion of the axioms and delegation →

Consequences

Five consequences, two audiences

With the base in place (the two axioms and delegation), the remaining five theorems split by who they are for: three for software engineers, two for knowledge work at large. Read each set together, or any theorem on its own page.

For software engineers

The empirical core: where AI leverage lands, the team structure that follows, and the cost paradox. This is what grounds RACE Programming's design.

  1. Theorem II · The One-Person Orchestra Theorem
    Full SDLC experience + delegation skills + AI tools = a 10x one-person orchestra. A mathematical certainty given the skill set; the only variable is time to adoption.
  2. Theorem III · The Team Compression Theorem
    Engineers without delegation accelerate ~3x, not 10x. Coding speeds up 20x but is only ~30% of delivery. A 3-person micro-team matches a 9-person Scrum team.
  3. Theorem IV · The Paradox of Enterprise Development Cost
    AI raises capacity, but demand (vibe-coders, automation appetite) outpaces it. Efficiency rises; the cost of enterprise development paradoxically does not fall. It keeps growing.

Read the Engineering Theorems together →

For knowledge work

The broader thesis: how the labor market stratifies into a new aristocracy, and why the same fracture spreads to every white-collar field.

  1. Theorem V · AI Creates a New Aristocracy, Not Equality
    AI filters rather than democratizes: 4% elite (10x), 16% adapters (3x micro-teams), 80% pure coders displaced. The pre-AI delegation advantage compounds.
  2. Theorem VI · The Fractal Stratification of Knowledge Work
    Software is Patient Zero. The same three-tier fracture (AI Aristocracy, Synthetic Micro-Teams, Displaced Layer) scales to every white-collar industry.

Read the Knowledge-Work Theorems together →

What this theory is not

Not a claim about AGI, consciousness, or the singularity. Narrow scope: current-generation AI systems (LLMs with tool use and code execution) × structure of software engineering work. Intended to remain valid across model generations, operating above any specific model.

Not a claim that engineers disappear, but a claim about which engineers thrive. Pure code-writing is displaced; engineers who master delegation and own a domain compound their advantage. The theory describes how the market stratifies, not a target to aim for.

Reading order

Fast path: the base (axioms + delegation) → T2 (The One-Person Orchestra Theorem) → T3 (The Team Compression Theorem), both on the Engineering Theorems page. Sufficient to understand why RACE Programming looks different from Scrum. Complete picture: read all six in order, since later theorems build on earlier ones.

FAQ

Frequently asked questions

What is AI-First Theory?
AI-First Theory is the scientific foundation of RACE Programming, consisting of two axioms and six theorems that formalize what happens to software engineering when AI systems can read, write, and verify code at production grade. It is descriptive, not prescriptive: it describes what shifts, not what to do about it.
What are the two axioms of AI-First Theory?
Axiom I (What AI is actually good at): AI excels at exactly two cognitive functions, finding information and generating actionable insights, where it delivers 1000x+ acceleration; everything else (judgment, intuition, tacit knowledge) stays human. Axiom II (Humanity always adapts, but this time is different): humans always adapt to new tools, but earlier tools amplified capability while AI replaces entire cognitive functions. It does the work for you, which is unprecedented.
What are the six theorems of AI-First Theory?
T1 Delegation Is the Core Skill: the foundational skill is delegation, not prompting (verbalize, delegate, validate). T2 The One-Person Orchestra Theorem: full SDLC + delegation + AI = a 10x one-person orchestra. T3 The Team Compression Theorem: engineers without delegation accelerate ~3x, so a 3-person micro-team matches a 9-person Scrum team. T4 The Paradox of Enterprise Development Cost: efficiency rises but demand outpaces capacity, so cost keeps growing. T5 AI Creates a New Aristocracy, Not Equality: 4% elite, 16% adapters, 80% displaced. T6 The Fractal Stratification of Knowledge Work: the same three-tier split scales to every white-collar industry.
Is AI-First Theory a claim about AGI or superintelligence?
No. The theory is narrow and concerns the interaction between current-generation AI systems (large language models with tool use and code execution) and the structure of software engineering work as practiced in industry. It operates at a higher level of abstraction than any specific model and is intended to remain valid across model generations.
How does AI-First Theory relate to RACE Programming?
RACE Programming is the prescriptive framework derived from AI-First Theory via the AI-First Manifesto. The theory describes what changes; the manifesto articulates values; RACE Programming specifies roles, artifacts, ceremonies, and tooling. Each layer gets more concrete and each depends on the one above being sound.
What is the fastest way to understand AI-First Theory?
Read the two axioms first, then Theorem 2 (The One-Person Orchestra Theorem) and Theorem 3 (The Team Compression Theorem). Those three together are sufficient to understand why RACE Programming looks different from Scrum. For the complete picture, read all six theorems in order, since later theorems build on earlier ones.