The Five Eras of the Software Engineer

The developer writes all code by hand, learns the domain through immersion, and uses AI occasionally for search or quick answers. This is where most engineers were trained, and where institutional knowledge still lives. The IDE, Stack Overflow, and documentation are the primary tools.

This stage represents the baseline of software engineering as practiced for decades. Every line of production code originates from a human typing it. Quality depends entirely on the developer's skill, code review from peers, and manual testing. Most formal CS education still trains engineers for this mode of work.

Only 14% of professional developers still avoid AI tools entirely (GitHub Octoverse 2025). This stage is rapidly becoming legacy, though the foundational skills it developed — debugging intuition, systems thinking, deep language knowledge — remain essential at every subsequent era.

The developer codes with AI tools — GitHub Copilot, ChatGPT sidebar, Tabnine — accepting suggestions, generating boilerplate, and accelerating routine tasks. The developer remains the primary author, understands every line, and uses AI as an accelerator. The ratio is approximately 60% manual / 40% AI.

This is where most organizations are today. 84% of developers now use AI coding tools regularly. Copilot alone has 20M users and is deployed across 90% of Fortune 100. Task speed-ups of up to 55% are documented. The trap is staying here — treating AI as "better autocomplete" rather than a catalyst for deeper workflow change.

The danger at this stage is well-documented: 16 of 18 CTOs surveyed in 2025 reported production disasters from unreviewed AI code. The METR study found experienced developers took 19% longer with AI but believed they were 20% faster — a dangerous perception gap.

AI does most of the implementation. The developer's primary role shifts to: defining specifications, reviewing AI output, making architecture decisions, validating tests, and maintaining quality gates. The developer approves rather than writes. The ratio is approximately 20% manual / 80% AI.

The defining term for this era is "Agentic Engineering" — spec-first, accountable, disciplined. Three terms describe the spectrum: Vibe Coding (Karpathy, Feb 2025) is the undisciplined end — prompting LLMs while you "forget that the code even exists." Vibe Engineering (Willison, Oct 2025) is the experienced middle. Agentic Engineering (2026 consensus) is the disciplined end — designing agentic loops, writing specifications, delegating to agents, spending time on architecture and review.

The critical insight from DORA 2025: this transition requires organizational transformation, not just better tools. AI magnifies the strengths of high-performing teams and the dysfunctions of struggling ones. Teams that skip governance, review processes, and cultural work see quality regression — bug rates climb 9% even as velocity increases.

The developer coordinates multiple AI agents to deliver solutions, defines system boundaries, and manages agent workflows. Steve Yegge calls the highest level "Gas Town" — custom orchestrators managing agent fleets. Code becomes an intermediate artifact, not the primary deliverable. The ratio is approximately 5% manual / 95% AI.

This era is characterized by the "10x engineer" becoming the engineer who manages 10 agents. Cursor CEO Michael Truell disclosed that 35% of Cursor's internally merged pull requests are created by autonomous agents. OpenAI engineers run 4–8 parallel Codex agents simultaneously. At Anthropic, approximately 90% of Claude Code itself was written using Claude Code. Teams shrink from 6–10 people to 3–5 while producing 2–5x output.

The verification challenge dominates: as Addy Osmani (Google) frames it, the scarce resource is judgment, not typing speed. Code review of machine output becomes the critical bottleneck. The developer must orchestrate, verify, and synthesize outputs from multiple parallel agents while maintaining architectural coherence.

The developer's primary product is system design and constraints, not implementation. Strategic architecture, business problem decomposition, and AI system governance become the core activities. Fleet-scale agent systems with AI supervisors coordinating groups of coding agents are the tool landscape. The developer defines "what" and "why" — never "how."

This era remains largely theoretical but is already visible at the frontier. The SDLC transforms into: Intent, Orchestration, Parallel Execution, Synthesis, and Governance. Human role is architect of the agent system and governor of outputs. AI handles primary implementation, testing, deployment, and monitoring.

Premium skills at this stage center on critical thinking, AI output evaluation, security oversight, and organizational communication. Gartner predicts 90% of enterprise engineers will use AI assistants by 2028. The ELEKS L5 maturity level — AI-autonomous with routine changes auto-approved — is the organizational equivalent.