Building Realistic Success with Generative AI
Over the last 2 years, there has been significant excitement in the area of Artificial Intelligence, specifically in the form of Generative AI. Many entrepreneurs and investors are asking themselves what can be built using these types of technologies.
Appricotsoft has assisted many clients in their efforts to become productive and successful with AI. Some examples include the launch of an MVP for AI products and providing guidance to help many projects avoid the pitfalls of being over-engineered based on Code Generation Tools and dependent on them. Through these experiences, we have come to learn that while AI has the potential to be a tremendous tool for speeding up the development and creation of new products and technologies, it is not a cure-all solution. The ability to distinguish between what is realistic and what is unrealistic is essential when trying to deliver a successful product that can be scaled and commercialized.
Let’s look at a few examples.
What You Can Build With AI (And Should)
1. Productivity Tools Assisted by AI
For example, writing assistants, email classifiers, task generators and meeting summarizers. These tools don’t need to be 100% accurate, they only need to help make workflows easier to manage.
Examples of these products include:
- Marketing co-pilots that write campaigns for you.
- Email triage tools that automatically place labels on incoming email messages.
- AI note-taking tools that listen to, and summarize, what you said during a meeting.
As you may have noticed from the examples provided, these products work very well because there is still a human element involved in the decision-making process. When a person is involved in this process, the person uses their own intelligence and thought processes to enhance the output produced by the AI tool and therefore makes it better suited for the final decision.
2. Developer and Internal Automation Tools
For example, Code Scaffolding and Test Support Tools assist developers by eliminating the need to do repetitive or redundant tasks. The use of AI tools can drastically reduce the amount of repeated work performed by Engineers and Quality Assurance Teams.
Examples of these tools include:
- Generating Unit Tests from Code Snippets.
- Refactoring Legacy Code through Suggestion.
- Drafting API Documentation through Annotated Endpoints.
At Appricotsoft, we utilize the Unison Framework to implement these types of tools to create a faster and more consistent delivery process. However, every piece of output produced with AI assistance is still subjected to review from a human before it is released into production.
3. Data Labelling/Tagging/Enrichment
AI has a huge potential to speed up back office activity as well as help enrich product offerings when the right data is used to train AI, while still leaving room for and working in tandem with human operators.
For example, AI has been used to:
- Tag product catalogs.
- Pre-screen moderation queues.
- Provide automated sorting functions for HR software.
All of these applications have a solid foundation in terms of being safe, efficient and practical as there is an opportunity to exercise human over-site or revert back should an issue arise.
What You Should Not (or Cannot) Create With AI (Yet)
1. Key Operational Decisions
If the application you are developing will have a significant impact on a person’s life (eg; a loan decision, medical diagnosis, or legal advice), generative AI cannot be trusted enough to go into production or production-like environments.
Why is this?
- Large Language Models (LLMs) will make errors. OpenAI has also acknowledged this.
- Large Language Models do not understand cause and effect or context outside of their training data.
- Errors are unpredictable and generally very difficult to identify through testing.
Even with the most extreme guardrails in place (if this/then that), these systems are very fragile. Therefore, if you develop an application in either FinTech, HealthTech, or LegalTech, it is critical to have AI assist you with the creation of the application, not make critical operational decisions for you.
2. Codebases Fully Generated by AI
The dream scenario is to be able to outline your product idea and receive a functioning MVP. The reality is that you will receive something that appears to “work,” but will be neither secure nor scalable nor maintainable.
We have had several instances where clients have come to us after producing a minimally viable product (MVP) based solely on AI-generated code, and the results were indicative of the real challenges developers face:
- Lack of error handling or logging
- Unclear application architecture
- There has been no testing conducted
- Has serious security vulnerabilities
- Contains large amounts of hidden technical debt
As a result, our audit and review services of AI-generated code become the foundation for refactoring and improving these products.
👉 Related: 20 Major Mistakes We Keep Getting in the Code Reviews of AI
3. “Autonomous” Products
Many entrepreneurs fantasize about developing AI technology that learns from its users, alters itself accordingly, and evolves autonomously. This idea is pure fantasy until a version 1 (MVP) is built. Building adaptive AI requires:
- Structured, labelled, clean, and adequate data (for learning);
- Performance Monitoring and Feedback Loops.
- A limited/completely controlled setting to prevent the development of AI from becoming random/evolving.
Until after version 1 is released, continue evolving this AI autonomously through feedback, usage, experience.
What Should Founders Focus On?
🛠 AI as A Tool, Not a Strategy
You shouldn’t consider AI a product. Instead, consider AI a tool in your toolbox (similar to a database, payment processor, or front-end framework). So, our question is not “Can we use AI?” but rather “What will AI do for our customers?”
🔍 Check for Technical Viability Before Investing Time & Resources on Building
Ask for a technical evaluation early on in order to avoid spending months on building something. This is especially important if you plan to use AI within your product. Appricotsoft provides technical audits for MVPs, so check out our services if you want to avoid common pitfalls.
📊 Build an MVP with a Consumer Focus First
Think about what your product’s core value is before developing an AI-powered solution. Tailor the use of AI around enhancing the customer experience versus being a replacement of customer service. Having a functioning, straightforward, well-tested MVP is always superior to developing a complex AI prototype.
How Appricotsoft Helps Founders Build Real AI Products
Assistance building a successful AI product is difficult for those just getting started out and already have been through the experience of building an AI product that failed. This is why we created the Unison Framework. The Unison Framework allows developers to create AI-first products fast and predictably while providing an optimal level of safety by allowing them to perform feasibility checks before they begin building.
Here’s how we assist:
✅ Feasibility – Helping you understand what AI can safely do and what must be built manually.
✅ Clean execution – AI tools will support our team of engineers in producing final results, but the final outcome will always belong to the engineer team.
✅ Recovery from messes – Have you created something with AI that turned out to be a complete disaster? Let us assist you in getting your product back on track by running a thorough code review and supporting you through the coding process to make the necessary fixes.
✅ Prototyping & Production – From early stages of building a product through to the finished product, we are there to support you every step of the way as you create a quality prototype and launch the product to market.
We’ve been able to support many startups in the United States and Europe with AI projects due to our radical transparency in our work approach, weekly demos, and lack of black boxes.
A Quick Evaluation: How Can AI Assist Me?
A Quick Evaluation: How Can AI Assist Me?
- If you have a lot of repetitive, structured tasks associated with your product;
- A human is involved in making the most critical decisions;
- Precision and accuracy can enhance your products/services, but are not crucial elements of your products to maintain a sense of professionalism;
- There are opportunities for testing AI output in a safe environment before exposing users;
- And you’re open to changing/refactoring the code that was generated by AI – then let’s collaborate!
If you did not check off most of these boxes, now would be a good time to reach out to me so we can discuss what to do next and how to get started on building your product in the proper way.
More from the Appricotsoft Blog:
Conclusion
AI provides an incredible ability to build faster, develop smart testing methods, and deliver superior solutions to users, when applied effectively. However, AI will not replace any human effort, so it is essential for every developer to think carefully about the application of AI and consider how it will support human-based efforts.
Appricotsoft has made it their mission to provide developers throughout the globe with the best resources and guidance to navigate through the confusion created by the hype surrounding AI development. If you are interested in developing an AI-based solution that is viable and successful, we are ready to help you on your journey!