Vibe Coding in Access with Juan Soto

^{[Editor's note: This recap is from a video recorded about six months ago. While most Access content is evergreen, AI technology is evolving rapidly. The concepts Juan discusses are still relevant today–especially his insights into the short-, medium-, and long-term implications of AI for Access developers–but the technology demonstrated has advanced appreciably since his talk, and will continue to do so for the foreseeable future.]}

Ever wondered if you could create an Access database by simply describing what you want—no coding required?

On April 4, 2025, Microsoft Access MVP Juan Soto demonstrated "vibe coding"—a revolutionary approach where AI handles the technical implementation while you focus on business requirements. Using Microsoft Copilot and Google's Gemini Canvas, Juan showed how Access developers can leverage AI to generate tables, queries, forms, reports, and even VBA code through natural conversation. While the technology isn't perfect yet, it's advancing rapidly and already proving useful for everything from creating test data to generating complex class modules.

Whether you're curious about AI-assisted development or looking for ways to boost your productivity, this presentation offers a fascinating glimpse into how Access development might look in the near future—and what it means for professional developers.

The State of AI in Access Development

Current Microsoft Access AI Status

• No native Copilot integration in Access (unlike Excel, PowerPoint, and Outlook)
• No AI initiatives currently on the official Microsoft Access roadmap
• Access developers must use external AI tools like Copilot Web or Google Gemini
• Copilot for Microsoft 365 ($30/month) provides data privacy protections

Important Privacy Considerations

• Public AI tools (ChatGPT, Claude, etc.) don't protect your data privacy
• Data submitted to public AI may be used for training or other purposes
• Copilot for Microsoft 365 keeps work data siloed and protected (indicated by shield icon)
• Never upload confidential client data to public AI services

Vibe Coding Explained

What is Vibe Coding?

• A new development approach where AI does the coding while you provide direction
• Developer acts as business consultant, giving feedback and instructions to AI
• Partnership model: you define requirements, AI implements solutions
• Term emerged within the last month and a half
• Represents shift from traditional coding to AI-assisted development

The Process

• Describe what you want in natural language
• Review AI-generated code and functionality
• Provide feedback on what needs adjustment
• Iterate until the solution meets your requirements
• No direct coding required (in theory)

Demo 1: Building a Safety Inspection Database with Copilot

Initial Requirements

Juan started with a clear business scenario:

• Create database to log safety inspections during campus walks
• Track violations by building, floor, and department
• Record date, time, and location of inspections
• Assign violations to departments or outside vendors
• Track status: Open, Investigating, or Closed

Creating the Database Schema

Table Generation

• AI initially suggested only two tables (Inspections, Violations)
• Required prompting to add Buildings, Departments, and Vendors tables
• Generated proper DDL (Data Definition Language) queries
• Created appropriate primary keys and foreign key relationships

Implementation Challenges

• AI provided individual CREATE TABLE statements rather than single combined query
• Some syntax errors when trying to combine all statements
• Required running each CREATE TABLE query separately
• Demonstrated need for Access-specific knowledge to execute queries

Populating Test Data

VBA Approach

• Initial attempt using INSERT statements failed with syntax errors
• AI suggested VBA code to insert fake data instead
• Generated well-structured VBA procedure with proper formatting
• Code successfully inserted sample data across all tables
• Excellent code quality with proper indentation and comments

Creating Queries

Fire Safety Query

• Requested query to find all fire-related violations
• Initial query failed to return results
• Second attempt with refined logic worked correctly
• Demonstrated AI's ability to self-correct when given feedback

Form Creation Attempts

Mixed Results

• AI couldn't generate importable form definition files
• VBA-based form creation code had errors
• Successfully created basic form controls but with limitations
• Generated useful code structure despite execution failures
• Illustrated current limitations of AI for Access form generation

Report Generation

Report Creation Challenges

• AI provided VBA code to create report programmatically
• Code quality was excellent with proper naming conventions and indentation
• Generated supporting parameter query successfully
• Report creation code failed to execute despite well-structured code
• Demonstrated gap between code quality and functional execution

Advanced VBA: Class Modules

Impressive Class Generation

• Created proper class module for inspection data retrieval
• Generated test code to use the class
• Properly declared variables using explicit object types (Access.Report, DAO.Database)
• Excellent indentation and code documentation
• Code executed successfully after proper module naming
• Rating: 9.5 out of 10 for code quality

Demo 2: Web Application with Google Gemini Canvas

The Canvas Tool

• New Google tool released the same week as the presentation
• Free to use (unlike Copilot for Microsoft 365)
• Specialized for creating web applications
• Provides live preview of running application

Creating Safety Violation Tracking Web App

Initial Requirements

• Notate safety violations during campus walks
• Support multiple buildings
• Assign violations to responsible parties
• View open violations at a glance
• Display status in list format (Open, Close, Investigating)
• Click email fields to contact responsible parties

Technology Choice

• Initial attempt with AngularJS didn't work
• Switched to React framework
• AI generated complete working application
• Created responsive grid-based interface

Application Features Generated

• Tables for Buildings and Violations
• Grid view with status filtering (All, Open, Close, Investigating)
• Forms to add/edit buildings
• Form to add new violations with dropdowns
• Automatic email field population based on department selection
• 10 days of sample violation data

Iteration and Refinement

• AI responded to feedback to improve interface
• Added edit capabilities for buildings and departments
• Created proper table relationships
• Some features didn't work consistently (violation editing)
• Demonstrated both potential and current limitations

Code Quality Assessment

Strengths Observed

• Proper variable declarations with explicit object types
• Excellent code indentation and formatting
• Meaningful procedure and variable names
• Comprehensive code comments
• Appropriate use of error handling patterns
• Knowledge of Access-specific best practices

Current Limitations

• Not all generated code executes successfully
• Form and report generation less reliable than table/query creation
• Sometimes requires multiple iterations to get working code
• May need developer knowledge to troubleshoot and implement
• Inconsistent results between similar requests

Alternative AI Tools for Developers

Tools Mentioned

• Cursor.com: IDE integration for real-time coding assistance
• GitHub Copilot: Code completion and suggestions within development environment
• Claude.ai: Content generation for websites and LinkedIn
• ChatGPT: General-purpose AI assistant
• Sora: Image generation
• Google AI Studio: Free Google AI experimentation platform

Specialized Applications

• Creating AI agents for repetitive tasks
• Lead generation and research automation
• CRM data enrichment
• Sales process automation

The Future of Access Development

Short Term (Next 5-10 Years)

• Access developers still needed, similar to how COBOL developers remain in demand
• Role shifting from coder to business consultant
• AI as tool to increase productivity, not replacement
• Need for expertise in translating business requirements
• Critical thinking required to validate AI-generated solutions

The Learning Curve

• AI can't design optimal solutions without human guidance
• Business knowledge remains essential
• Developer must direct AI toward desired outcomes
• Quality assurance and testing still require human expertise

Professional Considerations

• Risk of presenting AI-generated code as custom work
• Responsibility for accuracy of AI-generated queries and reports
• Client expectations vs. AI capabilities
• Need for thorough testing and validation
• Professional liability for incorrect results

Practical Applications

Use Cases Where AI Excels

• Generating test data for database tables
• Creating class modules with proper structure
• Converting business requirements to initial schema designs
• Generating boilerplate code and standard patterns
• Creating documentation and code comments

Areas Requiring Caution

• Complex query logic requiring accuracy
• Financial calculations and critical business logic
• Form and report generation
• Production-ready code without testing
• Any client-facing functionality

Best Practices for AI-Assisted Development

Getting Started

• Start with small, well-defined projects
• Use AI for learning and exploration
• Always test AI-generated code thoroughly
• Keep detailed prompts and conversations
• Iterate based on results

Organizational Adoption

• Don't mass-deploy licenses without training
• Identify "heroes" willing to experiment
• Let early success drive broader adoption
• Provide proper education on AI tool usage
• Balance cost vs. productivity gains

Juan's AI Journey

Personal Development Strategy

• Maintains collections of AI websites in Edge
• Created AI agents for lead generation and research
• Uses Claude.ai for content creation
• Leverages Copilot for finding files and summarizing communications
• Pursuing AI MVP status alongside Access MVP renewal
• Planning AI-focused user group chapter
• Specializing in AI for manufacturing and operations

Key Insights

• AI development moving faster than presenters can keep up
• Technology from months ago already outdated
• Continuous learning essential
• Integration with existing workflows critical

Conclusion

Vibe coding represents an exciting evolution in Access development, where AI handles implementation details while developers focus on business requirements and quality assurance.

While current AI tools show impressive capabilities—particularly for generating well-structured VBA code and database schemas—they're not yet reliable enough to replace experienced developers. The technology excels at creating starting points, generating test data, and handling boilerplate code, but still requires human expertise for validation, debugging, and ensuring solutions meet business needs. As AI continues its rapid advancement,

Access developers who embrace these tools while maintaining their business consulting expertise will be well-positioned for the future.

Recording

The full recording is available on YouTube:

Join Live!

Want to get even more out of these presentations? Join the live Access User Group events! The next upcoming events are listed on the AUG Event Calendar.

Attending live gives you the opportunity to:

• Interact directly with presenters during Q&A sessions
• Network with other Access developers
• Share your own experiences and challenges
• Get immediate answers to your specific questions
• Participate in group discussions

With multiple user groups across different time zones (and languages!), you're sure to find a meeting time that works for your schedule.

Acknowledgements

• Base cover image generated by FLUX-schnell
• Initial draft generated by Claude-Sonnet-4.5