---
id: ins_software-3-ai-engineer
operator: Swyx (Shawn Wang)
operator_role: Founder Latent Space; coined "AI Engineer" as a distinct role
source_url: https://www.latent.space/
source_type: essay
source_title: "Software 3.0 and the AI Engineer landscape"
source_date: 2026-03-03
captured_date: 2026-05-02
domain: [ai-native, engineering]
lifecycle: [hiring-team-design, ai-workflow]
maturity: frontier
artifact_class: framework
score: { originality: 4, specificity: 4, evidence: 3, transferability: 4, source: 4 }
tier: B
related: []
raw_ref: raw/expert-content/experts/swyx.md
---

# We are in the transition from Software 2.0 to Software 3.0, AI Engineers will build the majority of new applications

## Claim
We're transitioning from Software 2.0 (ML models replace hand-written logic) to Software 3.0 (LLMs *control* the application's logic flow). AI engineers, not traditional software engineers, will build the majority of new applications. The key new discipline is *agent engineering*: designing systems where LLMs control flow, with deterministic tools as their action space.

## Mechanism
In Software 1.0, engineers wrote explicit logic. In 2.0, ML replaced parts of that logic with learned models. In 3.0, the LLM is the orchestrator, it reads inputs, plans steps, calls deterministic tools, evaluates outputs, retries. The skill set required: prompt engineering, evaluation/observability, tool design, context management, and graceful degradation. These overlap with traditional engineering but center the LLM as the runtime, which is a fundamentally different mental model.

## Conditions
Holds when:
- The application can tolerate non-deterministic LLM-led flow.
- The team can hire/develop AI Engineer skills (a real bottleneck).

Fails when:
- Highly regulated systems where deterministic flow is required by law or compliance.
- Latency-sensitive paths where LLM-orchestration overhead is unacceptable.

## Evidence
> "We are in the transition from Software 2.0 to Software 3.0, where AI engineers — not traditional software engineers — will build the majority of new applications, and the key discipline is agent engineering."

· Swyx (Shawn Wang) (synthesized from operator's published work)

## Signals
- Engineering org has named "AI Engineer" roles, not just engineers using AI tools.
- Codebase includes evals, observability, and tool-design as first-class concerns.
- Hiring rubrics include prompt-engineering and agent-design competencies.

## Counter-evidence
For most enterprise software, traditional deterministic logic still dominates and 3.0 patterns are inappropriate. The "AI Engineer" framing is also still consolidating; some argue it's a temporary specialization that will dissolve back into general engineering as tools mature.

## Cross-references
- ins_pm-as-orchestrator-of-agents, adjacent operator (Lenny Rachitsky)