← d3dev

AI Writes the Code and Humans Still Write the Rules

Bharath Salla · UX Collective (Medium) · Feb 2026
https://uxdesign.cc/ai-writes-the-code-and-humans-still-write-the-rules-a2058ca0734c
ai-codingvibe-codingdeveloper-toolsstartupsmarket-analysis
At a Glance

The AI coding market hit $7.37B in 2025. Lovable reached $100M ARR in 8 months. But the METR study found senior devs actually take 19% longer with AI tools — despite believing they're 20% faster. AI platforms are the best prototyping tools ever built, and potentially dangerous production tools without human oversight.

AI Writes the Code and Humans Still Write the Rules

Metadata

FieldValue
TitleAI writes the code and humans still write the rules
AuthorBharath Salla
SourceUX Collective (Medium)
Linkhttps://uxdesign.cc/ai-writes-the-code-and-humans-still-write-the-rules-a2058ca0734c
Date PublishedFeb 2026
Date Downloaded2026-02-26
Tagsai-coding, vibe-coding, developer-tools, startups, market-analysis

At a Glance

The AI coding market hit $7.37B in 2025. Lovable reached $100M ARR in 8 months. But the METR study found senior devs actually take 19% longer with AI tools — despite believing they're 20% faster. AI platforms are the best prototyping tools ever built, and potentially dangerous production tools without human oversight.

Quotes

The 50th prompt produces measurably worse code than the 5th.

— Multiple independent studies

AI doesn't know what it doesn't know. It generates plausible code, not correct code.

— Bharath Salla

A year ago, they would have built their product from scratch. But now 95% of it is built by an AI. You don't need a team of 50 engineers anymore.

— Garry Tan, CEO, Y Combinator

AI coding platforms are the best prototyping tools ever built and potentially dangerous production deployment tools if used without human oversight.

— Article conclusion

Sam's TLDR;

Comprehensive market analysis of the AI coding explosion. Lovable hit $100M ARR in 8 months, 25% of YC startups ship 95% AI-generated code, but the METR study found senior devs actually take 19% longer with AI tools despite feeling 20% faster. The honest take: AI platforms are the best prototyping tools ever built but potentially dangerous for production without human oversight. The real story isn't that devs got faster — it's that the definition of "developer" exploded.

Key Points

Market Leaders

PlatformFocusARRNotable
GitHub CopilotEnterprise IDE integration$1B+82% of devs use it, market leader by user count
Claude CodeTerminal agent, multi-file reasoning$1B+"Runaway market leader" among LLM providers for coding
Cursor (Anysphere)VS Code-based power tool$500M+$9.9B valuation
LovableNo-code app builder$200M$100M in 8 months, targeting $1B by summer 2026
Bolt.newBrowser-based speed demon$100M+Fastest time-to-demo, but quality degrades over iterations
ReplitFull cloud IDE + AI Agent$100M$10M to $100M in 9 months
WindsurfIDE with agentic flow modeAcquired $3BBought by OpenAI, May 2025. Highest code quality benchmarks (8.5/10)
v0 by VercelUI component specialistFree+proHighest quality score (9/10), React + shadcn/ui

Three User Tribes

TribeWhoWhy AI
Non-technical foundersDesigners, marketers, domain expertsCan build without hiring engineers. Gartner: 70% of new apps built outside IT by end of 2025 [5]
Technical startup foundersSkilled engineers who could code manuallyDramatically faster. 25% of YC W25 batch. Teams <10 reaching $10M revenue [5]
Enterprise developersLarge orgs (63% of market)Code review, docs, tests, function completion. 61% integrated into CI/CD [2]

The 9 Real Problems

  • Security holes — no platform auto-audits, prompt injection risks
  • Code degrades over time — inconsistency grows with project size
  • Perception gap — feels fast, actually slower for complex work
  • Runway token costs — unpredictable, can spike during debugging
  • Compliance blind spots — no SOC2, no HIPAA defaults, no audit trails
  • Vendor lock-in — migration is painful (especially Bolt, Replit)
  • "Almost right" frustration — 66% of devs cite this as top AI frustration [2]
  • Technical debt explosion — 4x code cloning, 7.2% delivery stability drop [8]
  • No production-ready code — every benchmark reaches this conclusion [3]
  • Full Summary

    The article is a thorough, data-backed analysis of the AI coding platform landscape as of early 2026. It traces the arc from Karpathy coining "vibe coding" in February 2025 through the explosive growth of platforms like Lovable ($100M ARR in 8 months), Replit ($10M to $100M in 9 months), and Cursor ($9.9B valuation).

    The central thesis is nuanced: AI coding tools have genuinely democratized software creation — millions of people who couldn't build software now can. That's real. But the tools are fundamentally prototyping tools, not production tools. Every benchmark study reaches the same conclusion: no AI platform produces code you can ship without significant manual finishing.

    The most striking data point is the METR study [3] showing experienced developers actually took 19% longer with AI tools while believing they were 20% faster. This "perception gap" suggests AI doesn't uniformly accelerate development — it compresses the skill gap, making juniors faster while potentially slowing seniors on complex work.

    Security findings are alarming: 170 of 1,645 Lovable apps exposed user data [4], and "VibeScamming" attacks can inject backdoors via prompt manipulation. Combined with the finding that 29% of developers skip manual review of AI-generated code [2], this creates genuine risk.

    The article identifies a common "graduate workflow" pattern: prototype fast in Lovable/Bolt, validate with real users, then rebuild properly in Cursor or traditional development once the idea is proven. "Start vibe, graduate to code."

    The author's final framing: "AI coding platforms are the best prototyping tools ever built and potentially dangerous production deployment tools if used without human oversight." The printing press analogy is apt — it didn't replace writers, it changed what writing meant. AI coding is doing the same to software development.

    References

    1. [1]CB Insights — Coding AI Market Share December 2025. https://www.cbinsights.com/research/report/coding-ai-market-share-december-2025/
    2. [2]Stack Overflow Developer Survey 2025 — AI Section. https://survey.stackoverflow.co/2025/ai/
    3. [3]METR — Measuring Impact of Early 2025 AI on Developer Productivity. https://metr.org/blog/2025-07-10-early-2025-ai-experienced-os-dev-study/
    4. [4]Guardio Labs — VibeScamming Research. https://guard.io/labs/vibescamming-from-prompt-to-phish-benchmarking-popular-ai-agents-resistance-to-the-dark-side
    5. [5]Y Combinator Blog — AI Coding Trends. https://www.ycombinator.com/blog/
    6. [6]CB Insights — AI Software Development Market Map. https://www.cbinsights.com/research/ai-software-development-market-map/
    7. [7]Menlo Ventures — 2025 State of Consumer AI. https://menlovc.com/perspective/the-state-of-consumer-ai/
    8. [8]METR Study Academic Paper. https://arxiv.org/abs/2507.09089