Rigidity–Flexibility Pendulum

Software methodology has swung between documentation fidelity and execution velocity, each pole rational for its era's bottleneck — and agentic synthesis swings it back toward formal specification.

Methodology oscillates between documentation fidelity and execution velocity; each swing was rational for the bottleneck of its time.

  • Formal rigidity (Waterfall/RUP, 1970s–1990s) — high fidelity, low velocity. Royce actually prescribed staged phases with adjacent-phase feedback, early prototyping, and customer involvement; pedagogy flattened that into a feedback-free cascade. Strengths: traceability, reviewability, contracts that survive staff turnover. Weaknesses: feedback latency across specification freezes and late, expensive defects. Its rigorous artifacts turn out to be exactly the machine-interpretable intent agents need.
  • Iterative flexibility (Agile/Scrum, 2001–2020) — velocity bought by trading fidelity for tacit communication. Two-week sprints gave rapid validation, but rationale lived in conversation and tribal memory. Rational while humans were the only executors; brittle for agents, which cannot reconstruct undocumented intent from context (see tacit knowledge erosion).
  • Point acceleration (Copilot era, 2021–2024) — token-level autocomplete inside unchanged Agile structures. Short contexts and unreliable multi-file reasoning kept architecture and specification human, so methodology did not move.
  • Automated rigidity (spec-driven agentic development, 2025–2026) — multi-million-token contexts and multi-agent orchestration make whole cross-module features synthesizable from one formal document, with documented 3–5× velocity gains — conditional on rigorous up-front specification, without which quality degrades steeply under the ambiguity tax.

Use this history when justifying agentic specification discipline to skeptics: the return to Royce-style artifacts is not nostalgia but inversion. Synthesis is fast, so durable, complete, inspectable specifications are the scarce input. The mechanism is BDUF in agentic engineering; the economic statement is the bottleneck inversion to intent clarity.

Sources

  • Nguyen & Nguyen (2026), ch. 2