AI Engineer Conference

The Cure for the Vibe Coding Hangover

Corey J. Gallon presents a systematic framework that transforms AI agents from erratic novices into reliable implementation partners through precise planning, atomic features, and multi-sensory validation.

Corey J. Gallon
December 2024
~57 minutes

The Vibe Coding Problem

The promise of AI coding agents is seductive: describe what you want, watch the code appear. But reality delivers frustration—hallucinated APIs, circular dependencies, and architectures that collapse under their own complexity.

Corey J. Gallon diagnoses the root cause: "vibe coding", where engineers skip planning and rely on prompt engineering to compensate for incomplete thinking. The cure? A two-phase framework that separates Planning (human-led architectural thinking) from Implementation (agent-assisted execution).

Planning produces artifacts through five systematic steps that externalize thinking and enable autonomous agent work. Critically, planning is purely human work—agents assist as thinking partners but humans make every decision. Only after complete specifications exist does the implementation loop begin.

Key Insights

Core principles and foundational concepts

Define Vibe Coding: The Problem with 'Feels Productive' AI Development

Vibe coding is the low-spec zero planning approach to AI accelerated development that feels productive but results in brittle, unmaintainable demo-ware. The 'hangover' is the resulting despair when you try to build maintainable, understandable software this way.

Notable Quotes:

1:03-1:24

"Vibe coding is the low-spec zero planning approach to AI accelerated development that feels productive but results in brittle unmaintainable demo-ware. The hangover is the resulting despair when you try to build maintainable, understandable software."

5:30-5:58

"Treating AI coding agents as pure productivity tools just to crank out code faster, using AI to generate software and not learning anything from the process and then 6 months later being no better as engineers having plateaued or worse becoming dependent on AI... That's AI dependency. That's the opposite of AI augmentation."

Architect vs. Implementer: Clear Role Separation

You are the architect and the agent is the implementer. You own the thinking—architecture, interfaces, system intent, structure, design decisions, and tradeoffs. The agent handles the doing—implementation, typing code, following patterns, implementing tests, and boilerplate. Delegate the doing, not the thinking.

Notable Quotes:

6:36-7:29

"You are the architect and the agent is the implementer. You own the thinking and that means architecture and interfaces, the intent of the system, the structure, design decisions and the associated tradeoffs. And then the agent handles the doing... Delegate the doing, not the thinking."

Specification > Prompt Engineering

Specification is far greater than prompt engineering. Prompt engineering treats AI interactions as an optimization problem rather than a communication problem—trying to find magic words that produce the right output rather than clearly defining what right means. Specifications are structured, precise definitions of requirements, behavior, interfaces, and acceptance criteria.

Notable Quotes:

8:37-9:38

"Specification is far greater than prompt engineering. Prompt engineering treats AI interactions as an optimization problem rather than a communication problem... Specifications are structured, precise definitions of requirements, of behavior, of interfaces and acceptance criteria."

9:38-9:44

"Write the blueprint, not the prompt. Specifications are structured, precise definitions... Prompts are conversational and ambiguous."

Define Done Before Implementing

Starting implementation without executable tests and observable success criteria means that agents lack clear completion criteria and immediate feedback. They can't self-validate, self-correct, and don't know when they're done. By defining tests up front, we give agents clear stop conditions that enable immediate feedback and self-correction.

Notable Quotes:

10:00-10:41

"Starting implementation without executable tests and observable success criteria means that agents lack clear completion criteria and immediate feedback. They can't self-validate, they can't self-correct, and they don't know when they're done... By defining tests up front, we give agents clear stop conditions."

The Learning Value Proposition

Every step in the framework creates specific learning opportunities. You're not just shipping software—you're building yourself. The software is valuable, but the engineer that you become is exponentially more valuable. Always be learning or you may as well be learning.

Notable Quotes:

6:06-6:16

"Every step in the framework creates specific learning opportunities. So that you're not just shipping software, you're building yourself. The software is valuable, but the engineer that you become is exponentially more valuable."

The 5-Step Planning Framework

Planning transforms vague ideas into atomic, sequenced, fully specified features ready for implementation. This is purely human work—agents assist as thinking partners, but you make every decision.

1

Master Project Specification

Transform user intent into technical blueprint with user stories, technical contracts, and acceptance criteria

2

Feature Specifications

Individual feature specs with three-level technical contracts (interface, output, error handling) and validation contracts

3

Dependency Discovery

Binary dependency test: 'Does feature A REQUIRE feature B's output to work?' Create initial dependency matrix

4

Dependency Resolution

Validate and clean dependencies, detect and resolve circular dependencies with four resolution strategies

5

Implementation Planning

Topological sort for phase organization with binary success criteria, validation strategies, and task assignment

Multi-Sensory Feedback System

Unlike traditional testing that only validates outputs, the implementation loop uses three digital sensors to observe what agents actually create. This enables agents to self-validate, self-correct, and know when they're done.

Visual Sense

Screenshots, rendering, state, code structure—what the user sees

UI rendering verification
Component state inspection
Layout consistency checks

Auditory Sense

Logs, errors, warnings, stack traces—what the system reports

Error log analysis
Warning detection
Stack trace interpretation

Tactile Sense

User workflows, API interactions, performance validation—how interactions respond

Button click testing
Form submission flows
API response validation

The Implementation Loop:

Write Code → Execute & Sense → Test & Validate → Correlate & Analyze → Refine → Loop until all tests pass AND all sensors report clean execution

Circular Dependency Resolution

When features depend on each other in circles, implementation becomes impossible. Apply these four strategies in priority order to break cycles and enable atomic implementation.

First

Dependency Elimination

Apply binary dependency test again - remove non-essential dependencies

Re-examine: Does Feature A ACTUALLY need Feature B's output?

Second

Revised Specification

Rethink contracts and interfaces - features don't need each other's output

Can we restructure the interface to remove this dependency?

Third

Feature Splitting

Split feature into smaller atomic units - may not have been atomic enough

Break into smaller pieces that can be implemented independently

Last Resort

Consolidation

Last resort - acknowledge it gets messy, combine features

When elimination, revision, and splitting all fail

Practical Takeaways for Engineers

Planning is Human Work

Agents assist as thinking partners but humans make every architectural decision. Don't outsource thinking.

Atomic Features Matter

Features must be implementable in a single focused session. If specifications feel scattered, split the feature.

Binary Dependencies

Each dependency either exists or doesn't—no partial requirements. This eliminates ambiguity about feature coupling.

No 20% Implementation Tracking

Features are either complete or not. Progress tracking is binary to avoid false confidence in partial work.

Dependency Graphs Save Time

Visual matrices reveal circular dependencies immediately, preventing weeks of rework.

Templates Force Clarity

Well-structured templates guide thinking and capture results in artifacts agents can actually use.

Notable Quotes

0:15-1:24
Opening story: The vibe coding cycle

"Inspiration strikes. You've got an idea and you know exactly how you're going to build it. You fire up your favorite AI coding agent. You jam in those prompts and then you hand it over. Hey, look at him go. He's done it. That is to say, you've done it. The app works. This is what 10x engineering really feels like. You're a genius. A rebel in the AI revolution. But then Monday rolls around. You want to add a feature or you want to change the way that it works and you realize that you don't understand it. You can't maintain it and you have to throw most or all of it away."

7:39-8:27
On slow down to go fast

"Slow down and iterate in order to go fast. The problem is the starting over cycle. Without deliberate iteration on validated work, you end up repeatedly starting from scratch. And so 3 months in, you've had multiple abandoned attempts instead of consistently improving on one single system. Deliberate iteration enables compounding returns on both understanding and on productivity."

37:13-37:17
On binary completion criteria

"There is no 20% implementation tracking. Is a feature implemented or is it not?"

43:00-43:07
On implementation being the only AI coding step

"This is the only one step in this entire framework that is the AI writes code."

39:02-39:48
On multi-sensory feedback

"The agent implements code and then executes it while gathering feedback through these digital senses. The visual sense - what renders, the auditory sense - what the system reports, and the tactile sense - how interactions respond. This sensory feedback provides rich diagnostic information about what's actually happening in the application."

46:24-46:47
On correlating tests with sensory feedback

"This integrated analysis reveals both what failed from the tests and why it failed from sensory feedback which enables targeted refinement by the agent."

40:14-40:43
On context optimization

"Without systematic context assembly, you're dumping entire planning documents into agent sessions, which wastes context window on irrelevant information."

54:23-54:32
On template systems

"Without templates, you reinvent specification structure every time. And starting from scratch is exhausting."

3:37-4:14
On framework philosophy

"The framework is comprised of three pillars: Principles (the philosophy), Process (the workflow), Tools (accelerators/enablers)... The framework is not just about building faster. It's about learning as we go."

1:52-2:46
On target audience

"This talk is NOT for you if programming is a job and not a craft that you're refining... If you're satisfied just having AI do it for you without needing to understand how or why... if vibe coding gets you what you need and that's good enough."

About the Speaker

Corey J. Gallon

Software Engineer & Framework Designer

Corey J. Gallon is a software engineer specializing in AI-augmented development who has been building with AI coding agents since 2022. His systematic framework emerged from real-world experience building production applications including specialized litigation support systems, real-time appliance monitoring packages, and digital publishing platforms.

Framework Resource: vibecodinghangover.com

"These aren't toys, these are real software applications that do real work every day."

Explore More AI Engineering Insights

Dive deeper into AI engineering with our comprehensive collection of talks, analyses, and case studies from leading experts.

Analysis based on Corey J. Gallon's talk "The Cure for the Vibe Coding Hangover" at AI Engineer Conference 2024. All quotes and timestamps verified from the full transcript.

Framework and methodology presented by Corey J. Gallon. Not officially endorsed by any organization.