Quality Bar Maintainability Composability As Primary Success Metric

Issue 87 Edition 2026-03-28 5 min read

General

Sources: 1 • Confidence: High • Updated: 2026-04-12 10:21

Key takeaways

Matt Webb states that a strong foundation for agentic and developer productivity is high-quality libraries that encapsulate hard problems behind interfaces that make the correct approach the easiest approach.
Matt Webb reports that his current practice is best described as "vibing" rather than "coding" or "vibe coding".
Matt Webb asserts that agentic coding often solves problems by exhaustively iterating until the problem is eliminated, even at extremely high token and compute cost.
Matt Webb states that the desired outcome for AI coding agents is fast solutions that remain maintainable, adaptive, and composable so improvements elsewhere can lift the whole stack.
Matt Webb asserts that in a "vibing" workflow, developers may read fewer lines of code while making more architecture-level decisions.

Matt Webb states that a strong foundation for agentic and developer productivity is high-quality libraries that encapsulate hard problems behind interfaces that make the correct approach the easiest approach.
Matt Webb states that the desired outcome for AI coding agents is fast solutions that remain maintainable, adaptive, and composable so improvements elsewhere can lift the whole stack.

Matt Webb reports that his current practice is best described as "vibing" rather than "coding" or "vibe coding".
Matt Webb asserts that in a "vibing" workflow, developers may read fewer lines of code while making more architecture-level decisions.

Matt Webb asserts that agentic coding often solves problems by exhaustively iterating until the problem is eliminated, even at extremely high token and compute cost.

What is the empirical distribution of token usage, wall-clock time, and iteration counts for agentic coding runs in this workflow, and how often do runs enter low-marginal-utility loops?
Do agent-produced solutions that are generated quickly actually meet the stated maintainability/adaptability/composability criteria when evaluated over weeks/months of change and incident response?
How large is the effect of high-quality shared libraries/interfaces on reducing defects, rework, and unsafe agent behaviors versus ad-hoc implementations in comparable projects?
In "vibing" workflows, how do code review, architecture review, and testing practices change, and what is the measurable impact on defect escape rate and maintainability?
Is there any direct operator/product/investor decision-readthrough in the corpus (e.g., explicit procurement choices, process mandates, or resource allocation changes) tied to these deltas?

Value may accrue to high quality shared libraries and interface layers that make correct behavior the default for both humans and agents, shifting competitive focus from task completion to maintainability and composability outcomes.
Tooling that detects and rate limits unbounded agent iteration could become a key control point as agentic coding can brute force solutions with high token and compute burn.
Developer workflows may shift toward architecture level decision making with less code reading, increasing demand for governance, testing, and review systems aligned to agent generated changes and lifecycle quality.

Benchmarks and customer reports start emphasizing maintainability, adaptability, and composability measured over weeks or months, with consistent improvement tied to shared libraries and interface design.
Published telemetry shows distributions of token usage, wall clock time, and iteration counts, plus measurable reductions in low marginal utility loops after adding loop detection or cost controls.
Process metrics show architecture review and testing practices evolving in vibing workflows, with lower defect escape rates or faster incident response without loss of maintainability.

Longitudinal evaluations show fast agent produced solutions degrade maintainability and increase rework or incidents, with composability benefits not materializing in real change management.
Cost and iteration profiles remain dominated by runaway loops, and added controls fail to meaningfully reduce token and compute burn or improve time to stable solutions.
Comparative studies find high quality libraries and interfaces do not materially reduce defects, unsafe behaviors, or rework versus ad hoc implementations in similar projects.