Software development continues to evolve quickly, but the last 12 months have seen some particularly significant changes in the way custom software is designed and developed.
Unsurprisingly, Artificial Intelligence (AI) has been the biggest driver of this change. AI is now having an impact on both sides of software development: it is changing the tools we as developers use to build software, while also creating new functionality that can be incorporated into the applications we develop for our clients.
For businesses considering custom software development, these changes have the potential to reduce development time, make prototyping more cost-effective and allow more sophisticated functionality to be delivered within existing development budgets.
AI is changing the development process
AI-assisted development has moved well beyond developers simply using ChatGPT to generate a function or troubleshoot an error.
Development tools can now use AI agents that work across an entire application codebase. These agents can investigate existing code, implement functionality, write automated tests, identify bugs, refactor code and assist with documentation.
At Codium, we see these tools as another evolution in the software development process. Developers have progressively moved from writing low-level code to using frameworks, libraries, cloud services and other tools that allow increasingly complex applications to be developed efficiently. AI-assisted development is the next significant step in this progression.
It does, however, place greater importance on good software engineering practices. AI can generate a large amount of code quickly, but an experienced developer still needs to determine whether that code is secure, maintainable, scalable and appropriate for the application's overall architecture.
Used correctly, AI allows developers to spend less time on repetitive development tasks and creating boilerplate files, and more time solving the business and technical problems that make custom software valuable.
Faster development makes Discovery more important
One area where we are already seeing a considerable benefit from AI is prototyping.
During the early stages of a software project, there are often technical assumptions that need to be tested before committing to a larger build. Previously, producing a functional proof of concept could require several days or weeks of development.
AI-assisted development now allows us to test some of these concepts considerably faster. Development teams can create small applications, integrate third-party APIs, test AI models and demonstrate working functionality earlier in the project.
This provides businesses with an opportunity to validate ideas and technical approaches before committing to the cost of developing the complete solution.
One recent example was an app prototype we developed to validate real-time video object classification and lookup using a combination of Apple's ARKit and LLM-based image analysis. Building and testing the concept early allowed us to validate the technical approach before considering a larger development project.
At the same time, as AI makes it easier to produce large amounts of code quickly, the Discovery Phase of a software project becomes more important, not less.
As the effort required to produce code decreases, greater emphasis needs to be placed on understanding what should be built and why. Before development begins, businesses still need to understand the problem being solved, the users involved, the required workflows and where the software will create measurable value.
AI is becoming part of business software
AI isn't only changing how we develop software. It is also becoming another technology that is be incorporated into the applications we build.
Over the last few years, much of the discussion around AI in business software has focused on chatbots. The opportunities are now considerably broader.
Modern applications can use AI to analyse documents, extract information, search internal business knowledge, summarise large amounts of data, classify information, generate content and assist users with complex business processes.
We are also seeing increased interest in AI agents. Unlike a traditional chatbot that simply provides a response, an agent can be provided with tools that allow it to perform controlled actions within an application.
For example, an agent could review historical information, retrieve information from several business systems, prepare a quote, create a task or update a customer record.
Development frameworks are also starting to provide better support for this functionality. Laravel, which we use extensively at Codium, now provides an AI SDK for integrating AI providers, agents, tools, structured outputs and embeddings directly into Laravel applications.
This makes it considerably easier to integrate AI functionality into existing business software rather than treating AI as a completely separate system.
One challenge we continue to find when integrating AI into business software is the non-deterministic nature of Large Language Models (LLMs). Unlike traditional software, where the same inputs can be expected to produce a predictable result, an LLM may produce different outputs from the same request. This becomes particularly important when AI-generated outputs are relied upon for business-critical decisions or workflows. Appropriate validation, guardrails and human oversight therefore remain important considerations when designing AI functionality into business software.
Integrations and automation are becoming more important
Most businesses already use a number of different software platforms to operate their business. This can include accounting systems, CRMs, document management platforms, payment providers, cloud storage and industry-specific applications.
As AI becomes more capable, the ability to securely integrate these systems becomes increasingly important.
An AI agent becomes considerably more useful when it can access the information required to complete a task and, where appropriate, interact with other business systems.
For example, rather than simply asking an AI system to summarise a customer enquiry, an integrated application could retrieve the customer's previous orders, review relevant pricing information, prepare a response and create a follow-up task for a staff member.
This is where we see some of the more practical opportunities for AI in custom business software. Rather than adding AI for the sake of having an AI feature, it can be incorporated into existing workflows to reduce repetitive work and help staff make better use of the information already available within the business.
Security and software architecture remain important
The ability to develop software faster also introduces some new risks.
AI-assisted and "vibe coding" tools have made it possible for people with limited development experience to create surprisingly capable applications. These tools can be useful for prototypes, internal tools and testing ideas, but producing an application quickly doesn't remove the engineering requirements of production software.
Applications still need appropriate authentication and permissions, secure database design, backups, logging, monitoring, testing and disaster recovery.
AI functionality introduces additional considerations. If an AI agent has access to business information or can perform actions within an application, careful consideration needs to be given to what it can access and what it is permitted to do.
Depending on the application, this can include user permissions, approval processes, audit trails and restrictions around the actions an AI agent can perform without human approval.
These are architectural decisions that should be considered during the design of the software rather than added after the application has been developed.
Where does this take custom software development?
We are still relatively early in understanding where AI, LLMs and agents will take custom software development. The technology and the tools available to developers are changing quickly, and both our industry and our clients are continuing to identify where they can provide meaningful value.
What is becoming clearer is that the opportunity extends well beyond simply generating code or adding a chatbot to an existing application. LLMs and agents have the potential to assist with, automate or accelerate increasingly complex business processes, particularly where those processes involve large amounts of information, repetitive decision-making or interaction across multiple systems.
We expect the next stage will be increasingly focused on identifying which parts of a business can benefit from these capabilities and how they can be safely incorporated into existing software and workflows.
At Codium, we are continuing to explore these technologies through our own development processes, prototypes and client projects. While the capabilities are evolving quickly, the fundamentals remain the same: understanding the business problem, identifying where technology can create measurable value, and designing software that applies the right technology to solve it.
