From Vibe Coding to VIABLE Coding

The data on AI coding productivity is in, and it's counterintuitive: senior engineers benefit most. A new framework explains why—and what it means for the future of software.

Cursor just hit a $10 billion valuation. GitHub Copilot crossed 15 million users. Every major IDE now ships with AI autocomplete. The assumption driving all of it: AI will flatten the skill gap, letting junior developers punch above their weight.

Here's the finding that should reframe how we think about AI coding tools: senior engineers get more value from them than juniors do.

This inverts the common narrative. We've been told AI democratizes expertise—that it removes the need for deep technical judgment by letting anyone prompt their way to working code. The data says otherwise.

A University of Chicago study of tens of thousands of developers found that agent acceptance rates increased approximately 6% for every standard deviation of experience. Senior developers wrote more "plans" before generating code. They used AI for architecture and strategy, not just autocomplete.

Meanwhile, a randomized controlled trial by the nonprofit METR found that experienced open-source developers were actually 19% slower when using AI tools—even though they believed they were 20% faster.

These findings aren't contradictory. They reveal something important: AI amplifies judgment, not just output. Engineers who can plan, evaluate, and verify gain leverage. Engineers who can't accumulate debt.

This is the difference between building skyscrapers and building treehouses.

Vibe Coding vs. VIABLE Coding

"Vibe coding" is Andrej Karpathy's term for building software by prompting AI without examining the code it produces. Close your eyes, describe what you want, and trust the output.

Michael Truell, CEO of the $29.3 billion AI coding company Cursor, rejects the term for his product. "If you close your eyes and you don't look at the code and you have AIs build things with shaky foundations," he said at Fortune's AI Brainstorm conference, "as you add another floor, and another floor, things start to kind of crumble."

Vibe coding optimizes for speed. What Truell is describing—and what the data supports—is something else. Call it VIABLE coding: AI-assisted development optimized for durable, production-grade value.

The framework:

V — Vision: Clear intent, constraints, and success criteria before prompting
I — Inspection: Human review of structure, foundations, and assumptions
A — Action: Executing code, running it, observing behavior
B — Behavior: How the system performs over time, under load, at edge cases
L — Longevity: Maintainability, extensibility, team understanding
E — Evidence: Tests, metrics, PRs, benchmarks—proof, not vibes

This isn't anti-AI. It's a bet on AI as a multiplier of professional judgment rather than a replacement for it.

The Evidence Is Messier Than Vendors Admit

Company-reported productivity gains are real but cherry-picked.

Cursor claims Salesforce saw 30% productivity uplift. Salesforce's engineering blog confirms they've designated Cursor as a "default coding agent" with 85% weekly adoption. Case studies show specific wins: 85% reduction in legacy code coverage time, 30% faster test failure triage.

The University of Chicago study found organizations using Cursor merged 39% more pull requests, with no significant increase in revert rates.

But broader industry data tells a more complicated story.

Bain & Company's September 2025 report called real-world savings "unremarkable." Among companies that rolled out AI coding tools, developer adoption remained low. Those using AI assistants saw 10-15% productivity gains—but the saved time often wasn't redirected to higher-value work. "So even those modest gains don't translate into positive returns," Bain noted.

One reason: developers only spend 20-40% of their time actually writing code. Speed up coding by 30% and you've improved maybe 10% of the job.

GitClear's analysis of 211 million lines of code found concerning quality trends:

Code churn (lines reverted within two weeks) nearly doubled since 2021
Duplicated code blocks increased 8x during 2024
Refactoring activity dropped from 24% of changes to under 10%
2024 was the first year copy-pasted lines exceeded moved lines

Google's DORA 2024 report found AI adoption improved code quality by 3.4% but reduced delivery stability by 7.2%.

The pattern: speed without inspection creates fragile systems. Vibe coding generates output. VIABLE coding generates value.

Why Senior Engineers Win

The senior engineer advantage makes sense through the VIABLE lens.

Experienced developers bring Vision—they know what success looks like before they start prompting. They can Inspect AI output against mental models built over years. They understand Behavior at scale because they've debugged production systems at 3am. They care about Longevity because they've inherited unmaintainable code.

Juniors often lack these capabilities. They accept AI output because they can't evaluate it. They optimize for working code, not lasting code. They're playing the short game because they don't yet understand the long one.

This is why the METR study showed experienced developers getting slower on individual tasks while the Chicago study showed organizations getting faster overall. The value isn't in typing speed—it's in judgment applied at the right moments.

AI doesn't replace the need for expertise. It raises the stakes on having it.

Cursor's Bet

Cursor's $29.3 billion valuation is a bet on VIABLE coding.

Their roadmap confirms it: more autonomy for well-specified tasks, but with verification built in. Team-level features for code review across human and AI output. Background agents that work while you do other things—then surface results for human inspection.

Their recent acquisition of code review startup Graphite extends this logic. As Truell put it: "The way engineering teams review code is increasingly becoming a bottleneck. Over the past 2.5 years, Cursor has made it much faster to write production code. However, for most engineering teams, reviewing code looks the same as it did three years ago."

Writing code is the Action phase. Cursor is expanding into Inspection, Behavior, and Evidence—the phases where senior judgment compounds.

They're not betting against AI autonomy. They're betting that the highest ceiling in software belongs to teams that use AI to amplify judgment, not bypass it.

The Stakes

The choice between vibe coding and VIABLE coding isn't about tools. It's about what kind of software we're willing to build.

For prototypes, demos, and throwaway projects, vibes are fine. Close your eyes. Ship fast. Move on.

For production systems—the ones that handle money, health data, infrastructure, safety—the foundations matter. Someone needs to look under the floorboards.

The data suggests that someone should be experienced. And that AI makes them more valuable, not less.

"We have lots to do to continue building out the technology, the product, and the team," Truell said when asked about an IPO.

Building skyscrapers takes patience. VIABLE takes time.

Sources: University of Chicago/Suproteem Sarkar study via Cursor blog (November 2025); METR RCT study (July 2025); Bain & Company Technology Report 2025; GitClear AI Code Quality Research 2025; Google DORA State of DevOps 2024; Salesforce Engineering Blog; Fortune AI Brainstorm conference (December 2025).