Top 15 Reasons the Vibes are
Off with Vibe & AI-assisted
Coding
(And what you can do about it)
Ivan Barajas Vargas,
Co-Founder & CEO of MuukTest
What is Vibe Coding?
https://www.linkedin.com/pulse/vibe-coding-collaborating-ai-build-software-
through-natural-pandey-vkipc/
Prototypes
Mockups
Marketing sites
Small features
What is AI-Assisted Coding?
Autocomplete
Code Refactoring
Scaffolding
Root cause
analysis
Ideation
Unit testing
(Mostly)
Developers currently
spend 19% LONGER*
on tasks when using
GenAI tools
* Measuring the Impact of Early-2025 AI on Experienced Open-Source Developer
Productivity | 10 July 2025
The early results are in, and…
The 15 challenges we discovered
1. Coding Tool
Overload
2. Shadow AI
3. Everyone [thinks
they] can code
Challenges
Lead, choose wisely, and manage closely
➔ Pick one stack and centralize it
➔ Anyone can write a prompt; great teams
write standards
➔ Capture input (not code) from business
& product
Approach
The natural risks of LLMs creep in
4. Hallucinations
5. Hard coded
vulnerabilities
6. Data
Leakage
Password =
myC4Tismy
3V3RYTHING
Challenges
Determine risk tolerance & setup guardrails
➔ Log and communicate problematic
prompts
➔ Use LLMs with solid privacy policies
➔ Review your generated output (and
inputs!) more, early, and often
Approach
Speed & excitement >>> long term goals
7. Architecture
Drift
8. Scalability as
an afterthought
9. Tech Debt
H-e-🏒🏒
Sprint 175:
99% debt
1% roadmap
Challenges
Take the time to onboard AI & learn
continuously
➔ Feed documentation on
architecture and internal best
practices (RAG-based)
➔ Check load performance often
➔ Carefully review the GenAI logs to
understand and tweak approach
Approach
Critical checks get lost in the shuffle
10. Accessibility and
UX overlooked
11. Observability
lacking
12. Nobody owns
the AI tooling
Challenges
Designate ownership and have
accountability
➔ Include UX and Accessibility in
your QA strategy
➔ Work with your senior devs to
instrument observability
➔ Designate strong responsible for
AI tool oversight (e.g, DevOps,
Prod Manager)
Approach
We’re coding fast, but not better code
13. More code = More bugs
14. Less mature code
being commited
15. Devs
burnt out
Challenges
Quality needs to move just as fast as Dev
➔ Developers need to own unit
testing
➔ Establish clear expectations for
code quality
➔ Monitor/adjust the extra load of
senior developers due to AI
Approach
Recap | Path to getting vibe or AI coding to work for you
Join us for dinner tonight!
Wye Hill Kitchen & Brewing
201 South Boylan Avenue
Raleigh, NC 27603
7PM out on the patio
➔ Choose your AI tooling wisely,
and manage it closely
➔ Setup the right guardrails
➔ Take the time to onboard the tool
properly and continuously learn
and manage how it operates
➔ Define clear roles &
responsibilities across eng,
product, QA, business

BoSUSA25 | Ivan Vargas | Top 15 Reasons the Vibes are Off With Vibe Coding

  • 1.
    Top 15 Reasonsthe Vibes are Off with Vibe & AI-assisted Coding (And what you can do about it) Ivan Barajas Vargas, Co-Founder & CEO of MuukTest
  • 2.
    What is VibeCoding? https://www.linkedin.com/pulse/vibe-coding-collaborating-ai-build-software- through-natural-pandey-vkipc/ Prototypes Mockups Marketing sites Small features
  • 3.
    What is AI-AssistedCoding? Autocomplete Code Refactoring Scaffolding Root cause analysis Ideation Unit testing (Mostly)
  • 4.
    Developers currently spend 19%LONGER* on tasks when using GenAI tools * Measuring the Impact of Early-2025 AI on Experienced Open-Source Developer Productivity | 10 July 2025 The early results are in, and…
  • 5.
    The 15 challengeswe discovered 1. Coding Tool Overload 2. Shadow AI 3. Everyone [thinks they] can code Challenges
  • 6.
    Lead, choose wisely,and manage closely ➔ Pick one stack and centralize it ➔ Anyone can write a prompt; great teams write standards ➔ Capture input (not code) from business & product Approach
  • 7.
    The natural risksof LLMs creep in 4. Hallucinations 5. Hard coded vulnerabilities 6. Data Leakage Password = myC4Tismy 3V3RYTHING Challenges
  • 8.
    Determine risk tolerance& setup guardrails ➔ Log and communicate problematic prompts ➔ Use LLMs with solid privacy policies ➔ Review your generated output (and inputs!) more, early, and often Approach
  • 9.
    Speed & excitement>>> long term goals 7. Architecture Drift 8. Scalability as an afterthought 9. Tech Debt H-e-🏒🏒 Sprint 175: 99% debt 1% roadmap Challenges
  • 10.
    Take the timeto onboard AI & learn continuously ➔ Feed documentation on architecture and internal best practices (RAG-based) ➔ Check load performance often ➔ Carefully review the GenAI logs to understand and tweak approach Approach
  • 11.
    Critical checks getlost in the shuffle 10. Accessibility and UX overlooked 11. Observability lacking 12. Nobody owns the AI tooling Challenges
  • 12.
    Designate ownership andhave accountability ➔ Include UX and Accessibility in your QA strategy ➔ Work with your senior devs to instrument observability ➔ Designate strong responsible for AI tool oversight (e.g, DevOps, Prod Manager) Approach
  • 13.
    We’re coding fast,but not better code 13. More code = More bugs 14. Less mature code being commited 15. Devs burnt out Challenges
  • 14.
    Quality needs tomove just as fast as Dev ➔ Developers need to own unit testing ➔ Establish clear expectations for code quality ➔ Monitor/adjust the extra load of senior developers due to AI Approach
  • 15.
    Recap | Pathto getting vibe or AI coding to work for you Join us for dinner tonight! Wye Hill Kitchen & Brewing 201 South Boylan Avenue Raleigh, NC 27603 7PM out on the patio ➔ Choose your AI tooling wisely, and manage it closely ➔ Setup the right guardrails ➔ Take the time to onboard the tool properly and continuously learn and manage how it operates ➔ Define clear roles & responsibilities across eng, product, QA, business